diff --git a/evidence/parallel-llm-fix/README.md b/evidence/parallel-llm-fix/README.md
new file mode 100644
index 000000000..8da2e3d5c
--- /dev/null
+++ b/evidence/parallel-llm-fix/README.md
@@ -0,0 +1,184 @@
+# Parallel LLM lease fix: commands and captured output
+
+Implementation commit: 78f7012.
+The base commit is f6ece41d31e1908649c0b4540bfc53d189caaf7e.
+The historical diagnosis is in ../parallel-llm-lease. Those old numbers are
+not post-fix results. All links below are literal captured output, including
+failures. Commands run from the repository root unless a different cwd is stated.
+
+## Kernel and SDK checks
+
+From packages/sdk:
+
+| Literal command | Captured output |
+| --- | --- |
+| `npm run test:prep` | [prep.txt](prep.txt) |
+| `npm run typecheck` | [typecheck-final.txt](typecheck-final.txt) |
+| `npm run typecheck:tests` | [test-types-final.txt](test-types-final.txt) |
+| `npm run build` | [build.txt](build.txt) |
+| `npx vitest run tests/authored-parallel-llm.test.ts` | [regression-pass.txt](regression-pass.txt) |
+| `npx vitest run tests/cli-probe.test.ts tests/authored-preflight.test.ts tests/communication-preflight.test.ts tests/bundle-preflight.test.ts tests/preflight.test.ts tests/cli.test.ts tests/authored-parallel-agents.test.ts tests/authored-parallel-llm.test.ts` | [final-focused.txt](final-focused.txt) |
+| `npm test` | [sdk-suite.txt](sdk-suite.txt) |
+
+The kernel build is a fixture prerequisite; no kernel code changed.
+The complete SDK suite was attempted, not declared green. It includes environment
+and fixture failures. To check one failing group on the unchanged base:
+
+```sh
+git worktree add --detach /tmp/flows-lease-baseline f6ece41d31e1908649c0b4540bfc53d189caaf7e
+ln -s /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/node_modules /tmp/flows-lease-baseline/packages/sdk/node_modules
+cd /tmp/flows-lease-baseline/packages/sdk
+npx vitest run tests/hosted-extension-isolation.test.ts
+```
+
+Output: [baseline-hosted.txt](baseline-hosted.txt). This reproduces 15 failures:
+missing bubblewrap and missing local Surface build files. This comparison does
+not establish the cause of every failure in the complete suite.
+
+## Mutation verification
+
+```sh
+python evidence/parallel-llm-fix/mutate.py cache
+python evidence/parallel-llm-fix/mutate.py async
+```
+
+[mutate.py](mutate.py) saves the selected source file as bytes, replaces exactly
+the chosen implementation, captures the test failure, restores the saved bytes,
+asserts equality, and reruns the same command. The commands and both complete
+outputs are pasted in [mutations.md](mutations.md).
+
+- Cache mutation: construct a fresh authored preflight for each call instead of
+  sharing it. Both capacities run nine probes instead of one.
+- Async mutation: route the async probe entry through the synchronous driver.
+  Two 45-second cold models starve the durable root lease; the test fails with
+  `lease_conflict: attempt has no active worker lease`.
+- These protections overlap. With async probing retained, removing the cache
+  does **not** expire child leases; the slow test fails its probe-count assertion.
+  It would be false to claim cache-only removal causes expiry after Phase 2.
+
+For the original child-lease failure, the new test was also copied to the
+unchanged base worktree, with the existing kernel explicitly selected:
+
+```sh
+cp packages/sdk/tests/authored-parallel-llm.test.ts /tmp/flows-lease-baseline/packages/sdk/tests/authored-parallel-llm.test.ts
+cd /tmp/flows-lease-baseline/packages/sdk
+RELAYFLOWD_BIN=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd npx vitest run tests/authored-parallel-llm.test.ts -t 'parallel llm capacity 1.*completes nine'
+```
+
+Output: [baseline-child-lease.txt](baseline-child-lease.txt).
+
+## Original diagnosis harness, rerun after the fix
+
+```sh
+cp evidence/parallel-llm-lease/diagnosis-harness.test.ts packages/sdk/tests/zz-scratch-repro.test.ts
+cd packages/sdk
+SCRATCH_PROBE_MS=4000 SCRATCH_CAP=1 npx vitest run tests/zz-scratch-repro.test.ts
+rm tests/zz-scratch-repro.test.ts
+```
+
+Output: [diagnosis-post-fix.txt](diagnosis-post-fix.txt). Its identification
+count includes the nine actual wrapper sessions; only one identify/auth round
+is preflight. This scratch harness is historical instrumentation, not the
+regression assertion; the committed tests discover every run journal from disk.
+
+## Provider repro
+
+```sh
+git fetch origin feat/examples-prompt-lab
+git worktree add --detach /tmp/flows-prompt-lab origin/feat/examples-prompt-lab
+git -C /tmp/flows-prompt-lab merge --no-edit relayflow/flows-software-garden-860ae350
+```
+
+The fetched branch was 735f2e0. Merge refused unrelated histories:
+[worktree-merge.txt](worktree-merge.txt). Instead the isolated example worktree
+used this checkout's built CLI and dependencies:
+
+```sh
+ln -s /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/node_modules /tmp/flows-prompt-lab/examples/prompt-lab/node_modules
+cd /tmp/flows-prompt-lab/examples/prompt-lab
+RELAYFLOWD_BIN=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js run evidence/runtime-findings/runtime-parallel-llm-repro.flow.ts --local-agent --agent-capacity 1 --input '{"n":1}' --data-dir /tmp/flows-verified-capacity1
+RELAYFLOWD_BIN=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js run evidence/runtime-findings/runtime-parallel-llm-repro.flow.ts --local-agent --agent-capacity 4 --input '{"n":1}' --data-dir /tmp/flows-verified-capacity4
+```
+
+Outputs: [capacity 1](repro-final-capacity1.txt), [capacity 4](repro-final-capacity4.txt).
+Earlier runs are also retained as repro-capacity1.txt and repro-capacity4.txt.
+
+```sh
+python evidence/parallel-llm-fix/scan-journals.py /tmp/flows-verified-capacity1 /tmp/flows-verified-capacity4
+```
+
+Output: [repro-final-journals.txt](repro-final-journals.txt). Every SQLite journal
+is inspected, including the root and deterministic children.
+
+## prompt-lab Promise.all restoration
+
+[prompt-lab-parallel.patch](prompt-lab-parallel.patch) restores parallel calls
+in runEngine and both previously serialized loops in new-agency. It is a
+downstream verification patch, not a change to the example branch's history.
+
+The example's existing proof script was copied to prove-fixed.sh with
+`npx flows` replaced by this checkout's absolute `node .../dist/cli.js`
+command. That exact script is preserved here. Run it from the example directory:
+
+```sh
+RELAYFLOWD_BIN=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd FLOWS_DATA_DIR=/tmp/flows-prompt-lab-data bash prove-fixed.sh /tmp/flows-prompt-lab-proof
+```
+
+Output: [prompt-lab.txt](prompt-lab.txt); full commands, model-step progress,
+reviewer edit and resume failure: [prompt-lab-run/](prompt-lab-run/).
+The parallel run produced a nine-row grid and reached both human gates, but
+the final resume failed with `unawaited_step`. Therefore acceptance box 3
+is **not fully verified**. No fix to that lifecycle failure is claimed here.
+
+```sh
+python evidence/parallel-llm-fix/scan-journals.py /tmp/flows-prompt-lab-data
+```
+
+Output: [prompt-lab-journals.txt](prompt-lab-journals.txt).
+
+## Standalone packaging
+
+```sh
+node scripts/build-standalone-cli.mjs bun-linux-x64 /tmp/flows-parallel-fixed
+```
+
+Output: [standalone-build-final.txt](standalone-build-final.txt). The same build
+command in the base worktree, with outfile /tmp/flows-parallel-base, is captured
+in [standalone-baseline-build.txt](standalone-baseline-build.txt).
+
+From the example directory, smoke commands:
+
+```sh
+RELAYFLOWD_BIN=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd /tmp/flows-parallel-fixed run evidence/runtime-findings/runtime-parallel-llm-repro.flow.ts --local-agent --agent-capacity 4 --input '{"n":1}' --data-dir /tmp/flows-final-repro-capacity4
+RELAYFLOWD_BIN=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd /tmp/flows-parallel-base run evidence/runtime-findings/runtime-parallel-llm-repro.flow.ts --local-agent --agent-capacity 4 --input '{"n":1}' --data-dir /tmp/flows-base-standalone-data
+```
+
+Both refuse importing the authored flow: [changed](standalone-smoke.txt),
+[base](standalone-baseline-smoke.txt). Build success is not a claim of standalone
+runtime success. The Node CLI provider repro above is separate.
+
+## Intermediate checks retained for transparency
+
+regression.txt has the initial test's wrong journal-count expectation (it omitted
+the deterministic f.done child). related.txt records the initial async signal
+classification failure, fixed before related-pass.txt. Their commands were,
+respectively, the regression command above and:
+
+```sh
+npx vitest run tests/cli-probe.test.ts tests/preflight.test.ts tests/cli.test.ts tests/authored-parallel-agents.test.ts tests/communication-environment-preflight.test.ts
+```
+
+static-refusals.txt used:
+
+```sh
+npx vitest run tests/authored-preflight.test.ts tests/cli-probe.test.ts tests/preflight.test.ts
+```
+
+Authentication material is not included. claude-auth.txt contains only the
+output of:
+
+```sh
+claude auth status | python -c 'import json,sys; d=json.load(sys.stdin); print(json.dumps({k:d.get(k) for k in ("loggedIn","authMethod")}))'
+```
+
+Captured stdout/stderr and patch files retain tool-generated whitespace verbatim.
diff --git a/evidence/parallel-llm-fix/baseline-child-lease.txt b/evidence/parallel-llm-fix/baseline-child-lease.txt
new file mode 100644
index 000000000..2744752f7
--- /dev/null
+++ b/evidence/parallel-llm-fix/baseline-child-lease.txt
@@ -0,0 +1,63 @@
+
+ RUN  v2.1.9 /tmp/flows-lease-baseline/packages/sdk
+
+ ❯ tests/authored-parallel-llm.test.ts (6 tests | 1 failed | 5 skipped) 55669ms
+   × parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 55668ms
+     → step_failed: journal step "llm-1" completed with lease_expired
+Inspect: flows replay 01M36D0HK698MR5P0FHJ0YC4W6 --at llm-1
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight
+AuthoredFlowExecutionError: step_failed: journal step "llm-1" completed with lease_expired
+Inspect: flows replay 01M36D0HK698MR5P0FHJ0YC4W6 --at llm-1
+ ❯ Module.readCompletedStepOutput src/authored-step-output.ts:85:11
+     83|       ...edges,
+     84|     });
+     85|     throw new AuthoredFlowExecutionError('step_failed', message, reaso…
+       |           ^
+     86|   }
+     87|   await recordAuthoredChild(journal, context.rootRunId, {
+ ❯ WorkerSlots.run src/worker-slots.ts:44:14
+ ❯ Object.llm src/authored-worker-step.ts:212:22
+ ❯ Module.observeStep src/progress.ts:48:20
+ ❯ AuthoredFlowOperation.begin src/authored-flow-operation.ts:174:23
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯
+
+⎯⎯⎯⎯⎯⎯ Unhandled Errors ⎯⎯⎯⎯⎯⎯
+
+Vitest caught 1 unhandled error during the test run.
+This might cause false positive tests. Resolve unhandled errors to make sure your tests are not affected.
+
+⎯⎯⎯⎯ Unhandled Rejection ⎯⎯⎯⎯⎯
+Error: Agent lease is already expired for 01M36D0HK698MR5P0FHJ0YC4W6/llm-1.
+ ❯ armExpiry src/worker-lease.ts:20:13
+     18|     const remaining = deadline - Date.now();
+     19|     if (!Number.isFinite(remaining) || remaining <= 0) {
+     20|       throw new Error(`Agent lease is already expired for ${dispatch.r…
+       |             ^
+     21|     }
+     22|     latestDeadline = deadline;
+ ❯ Module.withWorkerLease src/worker-lease.ts:47:5
+ ❯ LlmWorker.execute src/llm-worker.ts:60:46
+ ❯ JournalClient.onDispatch src/llm-worker.ts:49:26
+ ❯ JournalClient.emit node:events:508:20
+ ❯ JournalClient.onLine src/journal-client.ts:168:12
+ ❯ JournalClient.onData src/journal-client.ts:144:33
+ ❯ Socket.<anonymous> src/journal-client.ts:109:43
+ ❯ Socket.emit node:events:508:20
+ ❯ addChunk node:internal/streams/readable:564:12
+
+This error originated in "tests/authored-parallel-llm.test.ts" test file. It doesn't mean the error was thrown inside the file itself, but while it was running.
+The latest test that might've caused the error is "completes nine calls without expired child leases during slow preflight". It might mean one of the following:
+- The error was thrown, while Vitest was running this test.
+- If the error occurred after the test had been completed, this was the last documented test before it was thrown.
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+
+ Test Files  1 failed (1)
+      Tests  1 failed | 5 skipped (6)
+     Errors  1 error
+   Start at  05:49:49
+   Duration  56.91s (transform 599ms, setup 0ms, collect 1.06s, tests 55.67s, environment 0ms, prepare 46ms)
+
diff --git a/evidence/parallel-llm-fix/baseline-hosted.txt b/evidence/parallel-llm-fix/baseline-hosted.txt
new file mode 100644
index 000000000..66c3d3282
--- /dev/null
+++ b/evidence/parallel-llm-fix/baseline-hosted.txt
@@ -0,0 +1,413 @@
+
+ RUN  v2.1.9 /tmp/flows-lease-baseline/packages/sdk
+
+ ❯ tests/hosted-extension-isolation.test.ts (22 tests | 15 failed) 286ms
+   × hosted extension capability isolation > executes the exact capability-only handler for a queued receipt 8ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > executes the exact capability-only handler for a duplicate receipt 2ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > launches through the captured process primitive after builtin export synchronization 4ms
+     → promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+   × hosted extension capability isolation > ignores inherited launcher overrides and decodes manifests with the captured Buffer intrinsic 3ms
+     → promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+   × hosted extension capability isolation > streams verified bytes when the live store is replaced and no writable staging path exists 207ms
+     → promise rejected "Error: Hosted extension sandbox exited wi… { code: '…' }" instead of resolving
+   × hosted extension capability isolation > mounts pinned private Surface bytes when the live package changes before launch 1ms
+     → ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-Mim52z/dist/flow.js'
+   × hosted extension capability isolation > refuses Surface runtime bytes that differ from the reviewed pin before launch 1ms
+     → ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-6b3BQt/dist/flow.js'
+   × hosted extension capability isolation > refuses oversized Surface files through the bounded descriptor reader 1ms
+     → ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-IqCOBL/dist/flow.js'
+   × hosted extension capability isolation > shields verified Surface files before async settlement 2ms
+     → ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-7aruFD/dist/flow.js'
+   × hosted extension capability isolation > preserves a typed host refusal while disclosing only a fixed marker to the child 5ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to be Error: private Cloud policy detail { code: '…' } // Object.is equality
+   × hosted extension capability isolation > denies ambient credentials, host files, writes, network, subprocesses, and undeclared context verbs 7ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > enforces OS address-space and data bounds on native Buffer allocation 8ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_unsupported', …(1) }
+   × hosted extension capability isolation > blocks extra handler fields and authority-bearing receipt fields at the parent port 3ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+   × hosted extension capability isolation > constructs adapter authority with the captured freeze intrinsic 3ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > writes the Surface manifest and protocol without inherited toJSON behavior 3ms
+     → promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+
+⎯⎯⎯⎯⎯⎯ Failed Tests 15 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > executes the exact capability-only handler for a queued receipt
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > executes the exact capability-only handler for a duplicate receipt
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:221:26
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > launches through the captured process primitive after builtin export synchronization
+AssertionError: promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:283:11
+    281|           input: descriptor(),
+    282|           babysitterTurn: { queue: async () => ({ receiptId: 'receipt-…
+    283|         })).resolves.toEqual({ completionReason: 'success', capability…
+       |           ^
+    284|       } finally {
+    285|         process.execPath = originalExecPath;
+
+Caused by: Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:277:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > ignores inherited launcher overrides and decodes manifests with the captured Buffer intrinsic
+AssertionError: promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:373:11
+    371|           input: descriptor('delivery-options'),
+    372|           babysitterTurn: { queue: async () => ({ receiptId: 'receipt-…
+    373|         })).resolves.toEqual({ completionReason: 'success', capability…
+       |           ^
+    374|       } finally {
+    375|         Buffer.prototype.toString = bufferToString;
+
+Caused by: Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:367:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[3/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > streams verified bytes when the live store is replaced and no writable staging path exists
+AssertionError: promise rejected "Error: Hosted extension sandbox exited wi… { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:431:7
+    429|         return { receiptId: 'receipt-1', status: 'queued' };
+    430|       } },
+    431|     })).resolves.toEqual({ completionReason: 'success', capabilityCall…
+       |       ^
+    432|     expect(calls).toBe(1);
+    433|     expect(readFileSync(join(installed.directory, 'babysitter.flow.ts'…
+
+Caused by: Error: Hosted extension sandbox exited without a valid completion (exit 1): /tmp/hosted-bwrap-wrapper-o4PuPs/bwrap-wrapper:20
+if (child.error) throw child.error;
+                 ^
+
+Error: spawnSync /usr/bin/bwrap ENOENT
+    at Object.spawnSync (node:internal/child_process:1103:20)
+    at spawnSync (node:child_process:911:24)
+    at Object.<anonymous> (/tmp/hosted-bwrap-wrapper-o4PuPs/bwrap-wrapper:19:15)
+    at Module._compile (node:internal/modules/cjs/loader:1809:14)
+    at Object..js (node:internal/modules/cjs/loader:1940:10)
+    at Module.load (node:internal/modules/cjs/loader:1530:32)
+    at Module._load (node:internal/modules/cjs/loader:1332:12)
+    at wrapModuleLoad (node:internal/modules/cjs/loader:255:19)
+    at Module.executeUserEntryPoint [as runMain] (node:internal/modules/run_main:154:5)
+    at node:internal/main/run_main_module:33:47 {
+  errno: -2,
+  code: 'ENOENT',
+  syscall: 'spawnSync /usr/bin/bwrap',
+  path: '/usr/bin/bwrap',
+  spawnargs: [
+    '--unshare-all',
+    '--die-with-parent',
+    '--new-session',
+    '--clearenv',
+    '--cap-drop',
+    'ALL',
+    '--dir',
+    '/usr',
+    '--ro-bind',
+    '/usr/lib',
+    '/usr/lib',
+    '--ro-bind',
+    '/usr/lib64',
+    '/usr/lib64',
+    '--ro-bind',
+    '/usr/lib',
+    '/lib',
+    '--ro-bind',
+    '/usr/lib64',
+    '/lib64',
+    '--proc',
+    '/proc',
+    '--dev',
+    '/dev',
+    '--tmpfs',
+    '/tmp',
+    '--dir',
+    '/runtime',
+    '--dir',
+    '/extension',
+    '--dir',
+    '/extension/node_modules',
+    '--dir',
+    '/extension/node_modules/@relayflows',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface/dist',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface/dist/helpers',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface/dist/triggers',
+    '--dir',
+    '/extension/src',
+    '--perms',
+    '0500',
+    '--ro-bind-data',
+    '4',
+    '/runtime/node',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '5',
+    '/runtime/runner.mjs',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '6',
+    '/extension/node_modules/@relayflows/surface/package.json',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '7',
+    '/extension/node_modules/@relayflows/surface/index.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '8',
+    '/extension/node_modules/@relayflows/surface/runtime.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '9',
+    '/extension/node_modules/@relayflows/surface/dist/flow.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '10',
+    '/extension/node_modules/@relayflows/surface/dist/helpers/providers.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '11',
+    '/extension/node_modules/@relayflows/surface/dist/provider-trigger.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '12',
+    '/extension/node_modules/@relayflows/surface/dist/schedule.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '13',
+    '/extension/node_modules/@relayflows/surface/dist/triggers.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '14',
+    '/extension/node_modules/@relayflows/surface/dist/triggers/github.js',
+    '--perms',
+    ... 27 more items
+  ]
+}
+
+Node.js v25.6.0
+
+ ❯ Object.<anonymous> ../../../hosted-bwrap-wrapper-o4PuPs/bwrap-wrapper:19:15
+ ❯ refuse src/hosted-extension-protocol.ts:135:21
+ ❯ ChildProcess.<anonymous> src/hosted-extension-protocol.ts:234:21
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[4/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > mounts pinned private Surface bytes when the live package changes before launch
+Error: ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-Mim52z/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:437:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[5/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > refuses Surface runtime bytes that differ from the reviewed pin before launch
+Error: ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-6b3BQt/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:462:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[6/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > refuses oversized Surface files through the bounded descriptor reader
+Error: ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-IqCOBL/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:483:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[7/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > shields verified Surface files before async settlement
+Error: ENOENT: no such file or directory, copyfile '/tmp/flows-lease-baseline/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-7aruFD/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:507:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[8/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > preserves a typed host refusal while disclosing only a fixed marker to the child
+AssertionError: expected Error: bubblewrap is unavailable { code: '…' } to be Error: private Cloud policy detail { code: '…' } // Object.is equality
+
+- Expected
++ Received
+
+- [Error: private Cloud policy detail]
++ [Error: bubblewrap is unavailable]
+
+ ❯ tests/hosted-extension-isolation.test.ts:560:5
+    558|       provider: 'github', eventType: 'pull_request.labeled', deliveryI…
+    559|     });
+    560|     await expect(runVerifiedNativeExtensionSandbox({
+       |     ^
+    561|       artifact: await artifact(source), manifest: validateFlowExtensio…
+    562|       babysitterTurn: { queue: async () => { throw refusal; } },
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[9/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > denies ambient credentials, host files, writes, network, subprocesses, and undeclared context verbs
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:624:13
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[10/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > enforces OS address-space and data bounds on native Buffer allocation
+AssertionError: expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_unsupported', …(1) }
+
+- Expected
++ Received
+
+- Object {
++ PluginError {
+    "code": "plugin_unsupported",
+-   "message": StringMatching /(?:Failed to allocate memory|Array buffer allocation failed)/u,
+  }
+
+ ❯ tests/hosted-extension-isolation.test.ts:646:5
+    644|     });
+    645|     let calls = 0;
+    646|     await expect(runVerifiedNativeExtensionSandbox({
+       |     ^
+    647|       artifact: installed,
+    648|       manifest: validateFlowExtensionManifest(manifest()),
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[11/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > blocks extra handler fields and authority-bearing receipt fields at the parent port
+AssertionError: expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+
+- Expected
++ Received
+
+- Object {
+-   "code": "plugin_event_unroutable",
++ PluginError {
++   "code": "plugin_unsupported",
+  }
+
+ ❯ tests/hosted-extension-isolation.test.ts:675:5
+    673|     });
+    674|     let calls = 0;
+    675|     await expect(runVerifiedNativeExtensionSandbox({
+       |     ^
+    676|       artifact: await artifact(source), manifest: validateFlowExtensio…
+    677|       babysitterTurn: { queue: async () => { calls += 1; return { rece…
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[12/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > constructs adapter authority with the captured freeze intrinsic
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:745:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[13/15]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > writes the Surface manifest and protocol without inherited toJSON behavior
+AssertionError: promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:788:11
+    786|           babysitterTurn: { queue: async () => ({ receiptId: 'receipt-…
+    787|           timeoutMs: 3_000,
+    788|         })).resolves.toEqual({ completionReason: 'success', capability…
+       |           ^
+    789|       } finally {
+    790|         if (previous === undefined) delete (Object.prototype as { toJS…
+
+Caused by: Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:781:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[14/15]⎯
+
+ Test Files  1 failed (1)
+      Tests  15 failed | 7 passed (22)
+   Start at  05:47:28
+   Duration  2.51s (transform 1.11s, setup 0ms, collect 2.01s, tests 286ms, environment 0ms, prepare 76ms)
+
diff --git a/evidence/parallel-llm-fix/baseline-worktree.txt b/evidence/parallel-llm-fix/baseline-worktree.txt
new file mode 100644
index 000000000..f7834fc95
--- /dev/null
+++ b/evidence/parallel-llm-fix/baseline-worktree.txt
@@ -0,0 +1,2 @@
+Preparing worktree (detached HEAD f6ece41)
+HEAD is now at f6ece41 port(drive-cloud): emit the tick as a v2 kernel spec (#455)
diff --git a/evidence/parallel-llm-fix/build.txt b/evidence/parallel-llm-fix/build.txt
new file mode 100644
index 000000000..e83dbde7b
--- /dev/null
+++ b/evidence/parallel-llm-fix/build.txt
@@ -0,0 +1,4 @@
+
+> @relayflows/sdk@2.0.29 build
+> tsc && node scripts/make-cli-executable.mjs
+
diff --git a/evidence/parallel-llm-fix/claude-auth.txt b/evidence/parallel-llm-fix/claude-auth.txt
new file mode 100644
index 000000000..dad7b0d96
--- /dev/null
+++ b/evidence/parallel-llm-fix/claude-auth.txt
@@ -0,0 +1 @@
+{"loggedIn": true, "authMethod": "oauth_token"}
diff --git a/evidence/parallel-llm-fix/diagnosis-post-fix.txt b/evidence/parallel-llm-fix/diagnosis-post-fix.txt
new file mode 100644
index 000000000..6cd9a422e
--- /dev/null
+++ b/evidence/parallel-llm-fix/diagnosis-post-fix.txt
@@ -0,0 +1,25 @@
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+stdout | tests/zz-scratch-repro.test.ts > scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1
+{
+  "elapsedMs": 4716,
+  "probeCount": 11,
+  "byKind": {
+    "identify": 10,
+    "auth": 1
+  },
+  "completionReason": "success"
+}
+
+stdout | tests/zz-scratch-repro.test.ts > scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1
+lease_expired attempts: []
+
+ ✓ tests/zz-scratch-repro.test.ts (1 test) 4792ms
+   ✓ scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1 4792ms
+
+ Test Files  1 passed (1)
+      Tests  1 passed (1)
+   Start at  05:46:03
+   Duration  5.97s (transform 576ms, setup 0ms, collect 1.01s, tests 4.79s, environment 0ms, prepare 46ms)
+
diff --git a/evidence/parallel-llm-fix/fetch.txt b/evidence/parallel-llm-fix/fetch.txt
new file mode 100644
index 000000000..c67ee1129
--- /dev/null
+++ b/evidence/parallel-llm-fix/fetch.txt
@@ -0,0 +1,3 @@
+From https://github.com/agentworkforce/flows
+ * branch            feat/examples-prompt-lab -> FETCH_HEAD
+ * [new branch]      feat/examples-prompt-lab -> origin/feat/examples-prompt-lab
diff --git a/evidence/parallel-llm-fix/final-focused.txt b/evidence/parallel-llm-fix/final-focused.txt
new file mode 100644
index 000000000..ba2021626
--- /dev/null
+++ b/evidence/parallel-llm-fix/final-focused.txt
@@ -0,0 +1,35 @@
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/preflight.test.ts (67 tests) 148ms
+ ✓ tests/cli.test.ts (70 tests) 2801ms
+   ✓ flows check CLI > binds a checked relative wrapper to the flow directory for worker execution 726ms
+   ✓ flows check CLI > resolves a bare PATH-resolved claude with no declared model, in an isolated PATH 532ms
+ ✓ tests/cli-probe.test.ts (6 tests) 861ms
+ ✓ tests/bundle-preflight.test.ts (4 tests) 1443ms
+   ✓ bundle execution preflight > ignores surrounding cache configuration on a verified cache hit 626ms
+   ✓ bundle execution preflight > uses the built alias for a nameless flow even in a digest-only cache directory 786ms
+ ✓ tests/communication-preflight.test.ts (13 tests) 42ms
+ ✓ tests/authored-preflight.test.ts (3 tests) 98ms
+ ✓ tests/authored-parallel-agents.test.ts (8 tests) 15720ms
+   ✓ authored steps under local workers with capacity > runs more concurrent f.llm calls than the worker holds side by side, never more than its capacity 2201ms
+   ✓ authored steps under local workers with capacity > completes more concurrent f.agent calls than the worker holds: the overflow waits for a slot instead of parking 3688ms
+   ✓ authored steps under local workers with capacity > runs agents in distinct working directories side by side (the kernel carries cwd) 831ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a symlink alias of the same directory 2110ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a directory nested inside the other 1920ms
+   ✓ authored steps under local workers with capacity > never starts queued agents once the body has failed 1990ms
+   ✓ authored steps under local workers with capacity > never starts a queued agent when the agent holding the only slot fails 2053ms
+   ✓ authored steps under local workers with capacity > parks the overflow when the body is not told the capacity (the defect this closes) 927ms
+ ✓ tests/authored-parallel-llm.test.ts (6 tests) 119261ms
+   ✓ parallel llm capacity 1 > deduplicates concurrent preflight probes 5453ms
+   ✓ parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 8940ms
+   ✓ parallel llm capacity 1 > keeps the durable root lease alive across two cold models 47791ms
+   ✓ parallel llm capacity 4 > deduplicates concurrent preflight probes 3310ms
+   ✓ parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 7300ms
+   ✓ parallel llm capacity 4 > keeps the durable root lease alive across two cold models 46466ms
+
+ Test Files  8 passed (8)
+      Tests  177 passed (177)
+   Start at  05:50:30
+   Duration  123.32s (transform 2.98s, setup 0ms, collect 10.31s, tests 140.37s, environment 1ms, prepare 555ms)
+
diff --git a/evidence/parallel-llm-fix/mutate.py b/evidence/parallel-llm-fix/mutate.py
new file mode 100644
index 000000000..c585dbc5f
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutate.py
@@ -0,0 +1,44 @@
+"""Revert one fix, capture failure, restore byte-for-byte, capture success."""
+import pathlib
+import subprocess
+import sys
+
+root = pathlib.Path(__file__).resolve().parents[2]
+sdk = root / 'packages/sdk'
+mode = sys.argv[1]
+if mode == 'cache':
+    path = sdk / 'src/authored-worker-step.ts'
+    before = 'await check(authoring)'
+    after = 'await authoredPreflight(flowPath)(authoring)'
+    pattern = 'deduplicates|completes nine'
+elif mode == 'async':
+    path = sdk / 'src/cli/cli-probe.ts'
+    before = 'const sequence = probeSequence(...args);'
+    after = 'return probeCli(...args);\n  const sequence = probeSequence(...args);'
+    pattern = 'parallel llm capacity 1.*keeps the durable'
+else:
+    raise ValueError(mode)
+original = path.read_bytes()
+assert before.encode() in original
+
+def run(label):
+    command = ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', pattern]
+    log = root / 'evidence/parallel-llm-fix' / f'mutation-{mode}-{label}.txt'
+    with log.open('w') as out:
+        out.write(f'cwd: packages/sdk\ncommand: {command!r}\n')
+        out.flush()
+        result = subprocess.run(command, cwd=sdk, stdout=out, stderr=subprocess.STDOUT)
+        out.write(f'\nexit: {result.returncode}\n')
+    return result.returncode
+
+try:
+    path.write_bytes(original.replace(before.encode(), after.encode(), 1))
+    failed = run('reverted')
+finally:
+    path.write_bytes(original)
+assert path.read_bytes() == original
+print(f'{mode}: restored byte-for-byte', flush=True)
+passed = run('restored')
+assert failed != 0, 'mutation did not fail'
+assert passed == 0, 'restored fix did not pass'
+print(f'{mode}: reverted exit={failed}; restored exit={passed}', flush=True)
diff --git a/evidence/parallel-llm-fix/mutation-async-restored.txt b/evidence/parallel-llm-fix/mutation-async-restored.txt
new file mode 100644
index 000000000..8b8fe3d1f
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutation-async-restored.txt
@@ -0,0 +1,15 @@
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'parallel llm capacity 1.*keeps the durable']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/authored-parallel-llm.test.ts (6 tests | 5 skipped) 47947ms
+   ✓ parallel llm capacity 1 > keeps the durable root lease alive across two cold models 47946ms
+
+ Test Files  1 passed (1)
+      Tests  1 passed | 5 skipped (6)
+   Start at  05:55:31
+   Duration  49.09s (transform 558ms, setup 0ms, collect 978ms, tests 47.95s, environment 0ms, prepare 43ms)
+
+
+exit: 0
diff --git a/evidence/parallel-llm-fix/mutation-async-reverted.txt b/evidence/parallel-llm-fix/mutation-async-reverted.txt
new file mode 100644
index 000000000..fcf670840
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutation-async-reverted.txt
@@ -0,0 +1,32 @@
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'parallel llm capacity 1.*keeps the durable']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ❯ tests/authored-parallel-llm.test.ts (6 tests | 1 failed | 5 skipped) 91068ms
+   × parallel llm capacity 1 > keeps the durable root lease alive across two cold models 91067ms
+     → lease_conflict: attempt has no active worker lease
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > keeps the durable root lease alive across two cold models
+JournalProtocolError: lease_conflict: attempt has no active worker lease
+ ❯ JournalClient.onLine src/journal-client.ts:165:27
+    163|       if (pending.timer !== undefined) clearTimeout(pending.timer);
+    164|       if (res.ok) pending.resolve(res.result);
+    165|       else pending.reject(new JournalProtocolError(res.error.code, res…
+       |                           ^
+    166|     } else {
+    167|       const ev = msg as ServerEvent;
+ ❯ JournalClient.onData src/journal-client.ts:144:33
+ ❯ Socket.<anonymous> src/journal-client.ts:109:43
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯
+
+ Test Files  1 failed (1)
+      Tests  1 failed | 5 skipped (6)
+   Start at  05:53:59
+   Duration  92.22s (transform 560ms, setup 0ms, collect 978ms, tests 91.07s, environment 0ms, prepare 46ms)
+
+
+exit: 1
diff --git a/evidence/parallel-llm-fix/mutation-async.txt b/evidence/parallel-llm-fix/mutation-async.txt
new file mode 100644
index 000000000..797559d89
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutation-async.txt
@@ -0,0 +1,2 @@
+async: restored byte-for-byte
+async: reverted exit=1; restored exit=0
diff --git a/evidence/parallel-llm-fix/mutation-cache-restored.txt b/evidence/parallel-llm-fix/mutation-cache-restored.txt
new file mode 100644
index 000000000..2cd7ed98b
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutation-cache-restored.txt
@@ -0,0 +1,18 @@
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'deduplicates|completes nine']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/authored-parallel-llm.test.ts (6 tests | 2 skipped) 20071ms
+   ✓ parallel llm capacity 1 > deduplicates concurrent preflight probes 2799ms
+   ✓ parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 8690ms
+   ✓ parallel llm capacity 4 > deduplicates concurrent preflight probes 1311ms
+   ✓ parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 7270ms
+
+ Test Files  1 passed (1)
+      Tests  4 passed | 2 skipped (6)
+   Start at  05:53:37
+   Duration  21.25s (transform 569ms, setup 0ms, collect 1.01s, tests 20.07s, environment 0ms, prepare 49ms)
+
+
+exit: 0
diff --git a/evidence/parallel-llm-fix/mutation-cache-reverted.txt b/evidence/parallel-llm-fix/mutation-cache-reverted.txt
new file mode 100644
index 000000000..2b544490f
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutation-cache-reverted.txt
@@ -0,0 +1,66 @@
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'deduplicates|completes nine']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ❯ tests/authored-parallel-llm.test.ts (6 tests | 4 failed | 2 skipped) 22493ms
+   × parallel llm capacity 1 > deduplicates concurrent preflight probes 3643ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+   × parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 10266ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+   × parallel llm capacity 4 > deduplicates concurrent preflight probes 1273ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+   × parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 7311ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 4 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > deduplicates concurrent preflight probes
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 4 > deduplicates concurrent preflight probes
+AssertionError: expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+
+- Expected
++ Received
+
+- 1
++ 9
+
+ ❯ Object.assertProbes tests/authored-parallel-llm.test.ts:61:27
+     59|   }
+     60|   return { fixture, client, assertJournals, assertProbes(count: number…
+     61|     expect(lines(probes)).toHaveLength(count);
+       |                           ^
+     62|   }, assertCapacity() {
+     63|     const intervals = lines(spans) as Array<{ start: number; end: numb…
+ ❯ tests/authored-parallel-llm.test.ts:86:10
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/4]⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight
+AssertionError: expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+
+- Expected
++ Received
+
+- 1
++ 9
+
+ ❯ Object.assertProbes tests/authored-parallel-llm.test.ts:61:27
+     59|   }
+     60|   return { fixture, client, assertJournals, assertProbes(count: number…
+     61|     expect(lines(probes)).toHaveLength(count);
+       |                           ^
+     62|   }, assertCapacity() {
+     63|     const intervals = lines(spans) as Array<{ start: number; end: numb…
+ ❯ tests/authored-parallel-llm.test.ts:99:10
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/4]⎯
+
+ Test Files  1 failed (1)
+      Tests  4 failed | 2 skipped (6)
+   Start at  05:53:12
+   Duration  23.80s (transform 596ms, setup 0ms, collect 1.13s, tests 22.49s, environment 0ms, prepare 42ms)
+
+
+exit: 1
diff --git a/evidence/parallel-llm-fix/mutation-cache.txt b/evidence/parallel-llm-fix/mutation-cache.txt
new file mode 100644
index 000000000..03abbfc85
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutation-cache.txt
@@ -0,0 +1,2 @@
+cache: restored byte-for-byte
+cache: reverted exit=1; restored exit=0
diff --git a/evidence/parallel-llm-fix/mutations.md b/evidence/parallel-llm-fix/mutations.md
new file mode 100644
index 000000000..842945e9e
--- /dev/null
+++ b/evidence/parallel-llm-fix/mutations.md
@@ -0,0 +1,180 @@
+# Mutation commands and literal captured output
+
+## cache
+
+Command: `python evidence/parallel-llm-fix/mutate.py cache`
+
+### reverted
+
+```text
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'deduplicates|completes nine']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ❯ tests/authored-parallel-llm.test.ts (6 tests | 4 failed | 2 skipped) 22493ms
+   × parallel llm capacity 1 > deduplicates concurrent preflight probes 3643ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+   × parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 10266ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+   × parallel llm capacity 4 > deduplicates concurrent preflight probes 1273ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+   × parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 7311ms
+     → expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 4 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > deduplicates concurrent preflight probes
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 4 > deduplicates concurrent preflight probes
+AssertionError: expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+
+- Expected
++ Received
+
+- 1
++ 9
+
+ ❯ Object.assertProbes tests/authored-parallel-llm.test.ts:61:27
+     59|   }
+     60|   return { fixture, client, assertJournals, assertProbes(count: number…
+     61|     expect(lines(probes)).toHaveLength(count);
+       |                           ^
+     62|   }, assertCapacity() {
+     63|     const intervals = lines(spans) as Array<{ start: number; end: numb…
+ ❯ tests/authored-parallel-llm.test.ts:86:10
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/4]⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight
+AssertionError: expected [ 'test-model', 'test-model', …(7) ] to have a length of 1 but got 9
+
+- Expected
++ Received
+
+- 1
++ 9
+
+ ❯ Object.assertProbes tests/authored-parallel-llm.test.ts:61:27
+     59|   }
+     60|   return { fixture, client, assertJournals, assertProbes(count: number…
+     61|     expect(lines(probes)).toHaveLength(count);
+       |                           ^
+     62|   }, assertCapacity() {
+     63|     const intervals = lines(spans) as Array<{ start: number; end: numb…
+ ❯ tests/authored-parallel-llm.test.ts:99:10
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/4]⎯
+
+ Test Files  1 failed (1)
+      Tests  4 failed | 2 skipped (6)
+   Start at  05:53:12
+   Duration  23.80s (transform 596ms, setup 0ms, collect 1.13s, tests 22.49s, environment 0ms, prepare 42ms)
+
+
+exit: 1
+
+```
+
+### restored
+
+```text
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'deduplicates|completes nine']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/authored-parallel-llm.test.ts (6 tests | 2 skipped) 20071ms
+   ✓ parallel llm capacity 1 > deduplicates concurrent preflight probes 2799ms
+   ✓ parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 8690ms
+   ✓ parallel llm capacity 4 > deduplicates concurrent preflight probes 1311ms
+   ✓ parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 7270ms
+
+ Test Files  1 passed (1)
+      Tests  4 passed | 2 skipped (6)
+   Start at  05:53:37
+   Duration  21.25s (transform 569ms, setup 0ms, collect 1.01s, tests 20.07s, environment 0ms, prepare 49ms)
+
+
+exit: 0
+
+```
+
+Restoration check:
+
+```text
+cache: restored byte-for-byte
+cache: reverted exit=1; restored exit=0
+
+```
+
+## async
+
+Command: `python evidence/parallel-llm-fix/mutate.py async`
+
+### reverted
+
+```text
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'parallel llm capacity 1.*keeps the durable']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ❯ tests/authored-parallel-llm.test.ts (6 tests | 1 failed | 5 skipped) 91068ms
+   × parallel llm capacity 1 > keeps the durable root lease alive across two cold models 91067ms
+     → lease_conflict: attempt has no active worker lease
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > keeps the durable root lease alive across two cold models
+JournalProtocolError: lease_conflict: attempt has no active worker lease
+ ❯ JournalClient.onLine src/journal-client.ts:165:27
+    163|       if (pending.timer !== undefined) clearTimeout(pending.timer);
+    164|       if (res.ok) pending.resolve(res.result);
+    165|       else pending.reject(new JournalProtocolError(res.error.code, res…
+       |                           ^
+    166|     } else {
+    167|       const ev = msg as ServerEvent;
+ ❯ JournalClient.onData src/journal-client.ts:144:33
+ ❯ Socket.<anonymous> src/journal-client.ts:109:43
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯
+
+ Test Files  1 failed (1)
+      Tests  1 failed | 5 skipped (6)
+   Start at  05:53:59
+   Duration  92.22s (transform 560ms, setup 0ms, collect 978ms, tests 91.07s, environment 0ms, prepare 46ms)
+
+
+exit: 1
+
+```
+
+### restored
+
+```text
+cwd: packages/sdk
+command: ['npx', 'vitest', 'run', 'tests/authored-parallel-llm.test.ts', '-t', 'parallel llm capacity 1.*keeps the durable']
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/authored-parallel-llm.test.ts (6 tests | 5 skipped) 47947ms
+   ✓ parallel llm capacity 1 > keeps the durable root lease alive across two cold models 47946ms
+
+ Test Files  1 passed (1)
+      Tests  1 passed | 5 skipped (6)
+   Start at  05:55:31
+   Duration  49.09s (transform 558ms, setup 0ms, collect 978ms, tests 47.95s, environment 0ms, prepare 43ms)
+
+
+exit: 0
+
+```
+
+Restoration check:
+
+```text
+async: restored byte-for-byte
+async: reverted exit=1; restored exit=0
+
+```
diff --git a/evidence/parallel-llm-fix/prep.txt b/evidence/parallel-llm-fix/prep.txt
new file mode 100644
index 000000000..c719b4217
--- /dev/null
+++ b/evidence/parallel-llm-fix/prep.txt
@@ -0,0 +1,5 @@
+
+> @relayflows/sdk@2.0.29 test:prep
+> ( cd ../../kernel && sh ../ops/cargo.sh build ) && ( [ ! -d ../../testdata/preflight ] || find ../../testdata/preflight -name '*-cli' -type f -exec chmod +x {} + )
+
+    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.07s
diff --git a/evidence/parallel-llm-fix/prompt-lab-journals.txt b/evidence/parallel-llm-fix/prompt-lab-journals.txt
new file mode 100644
index 000000000..b8d5c96a5
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-journals.txt
@@ -0,0 +1 @@
+{"directory": "/tmp/flows-prompt-lab-data", "journals": 29, "lease_expired": []}
diff --git a/evidence/parallel-llm-fix/prompt-lab-parallel.patch b/evidence/parallel-llm-fix/prompt-lab-parallel.patch
new file mode 100644
index 000000000..fda705b68
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-parallel.patch
@@ -0,0 +1,71 @@
+diff --git a/examples/prompt-lab/jobs/new-agency.ts b/examples/prompt-lab/jobs/new-agency.ts
+index 75801b8..ee2bc65 100644
+--- a/examples/prompt-lab/jobs/new-agency.ts
++++ b/examples/prompt-lab/jobs/new-agency.ts
+@@ -60,12 +60,11 @@ export async function newAgency(job: Job, input: NewAgencyInput): Promise<void>
+   }
+ 
+   // Run on fake visits — the background job; the grid fills when it finishes.
+-  const ran: { q: PiledQuestion; outputs: Record<string, Output> }[] = [];
+-  for (const c of plan.coverage) { // sequential: see runEngine
++  const ran = await Promise.all(plan.coverage.map(async c => {
+     const q = byId.get(c.questionId)!;
+     const shelf = c.patientIds.map((id) => patient.get(id)!);
+-    ran.push({ q, outputs: await runEngine(f, prompts.get(q.questionId)!.text, ask(q), shelf) });
+-  }
++    return { q, outputs: await runEngine(f, prompts.get(q.questionId)!.text, ask(q), shelf) };
++  }));
+   const rows = sortRows(ran.flatMap(({ q, outputs }) => Object.entries(outputs).map(([patientId, ai]): Row => ({
+     questionId: q.questionId, patientId, pile: q.pile,
+     highlight: highlights(ai, q.pile, prompts.get(q.questionId)!.firstPass),
+@@ -92,8 +91,7 @@ export async function newAgency(job: Job, input: NewAgencyInput): Promise<void>
+   // are commit candidates (Prompt QA'd); shared/mismatch get a proposed rewrite only.
+   const changes = changed(edited);
+   const changedQs = [...new Set(changes.map((r) => r.questionId))].sort();
+-  const rewrites: { q: PiledQuestion; rows: Row[]; prompt: string; passed: boolean }[] = [];
+-  for (const qid of changedQs) { // sequential: see runEngine
++  const rewrites = await Promise.all(changedQs.map(async qid => {
+     const q = byId.get(qid)!;
+     const rowsFor = changes.filter((r) => r.questionId === qid)
+       .map((r) => ({ patient: patient.get(r.patientId)! as Patient, ai: r.ai, target: r.target, notes: r.notes }));
+@@ -102,14 +100,13 @@ export async function newAgency(job: Job, input: NewAgencyInput): Promise<void>
+     const rows = changes.filter((r) => r.questionId === qid);
+     if (q.pile !== "agency-specific") {
+       const { prompt } = await llm<{ prompt: string }>(f, iterate(ask(q), existing, brief, rowsFor, []));
+-      rewrites.push({ q, rows, prompt, passed: true });
+-      continue;
++      return { q, rows, prompt, passed: true };
+     }
+     const r = await untilQaPasses<{ prompt: string }>(f,
+       (findings) => iterate(ask(q), existing, brief, rowsFor, findings),
+       (w) => promptQa(w.prompt, ask(q), snap.guidelines, brief));
+-    rewrites.push({ q, rows, prompt: r.value.prompt, passed: r.passed });
+-  }
++    return { q, rows, prompt: r.value.prompt, passed: r.passed };
++  }));
+ 
+   // A first-pass prompt is a candidate only once it ran on a shelf patient and
+   // you reviewed the rows; a gap-only draft waits for a patient that covers it.
+diff --git a/examples/prompt-lab/jobs/shared.ts b/examples/prompt-lab/jobs/shared.ts
+index 732b0f5..165d23f 100644
+--- a/examples/prompt-lab/jobs/shared.ts
++++ b/examples/prompt-lab/jobs/shared.ts
+@@ -66,16 +66,10 @@ export async function untilQaPasses<T>(
+ /**
+  * Run one prompt on fake visits: the same engine nurses use. One llm step per patient.
+  *
+- * Sequential on purpose. Every model call in this flow is independent and was
+- * written as Promise.all, but relayflows 2.0.29 loses the run when concurrent
+- * f.llm calls queue past their 30s lease (a stale completion becomes a fatal
+- * protocol_error): see evidence/runtime-findings/runtime-parallel-llm-repro.txt. Restore
+- * Promise.all here and in jobs/new-agency.ts when flows#561 and flows#560 ship.
+  */
+ export async function runEngine(f: Ctx, promptText: string, ask: Ask, patients: readonly Patient[]): Promise<Record<string, Output>> {
+-  const outputs: Record<string, Output> = {};
+-  for (const p of patients) outputs[p.id] = await llm<Output>(f, engine(promptText, ask, p));
+-  return outputs;
++  return Object.fromEntries(await Promise.all(patients.map(async p =>
++    [p.id, await llm<Output>(f, engine(promptText, ask, p))] as const)));
+ }
+ 
+ export interface Plan { coverage: { questionId: string; patientIds: string[] }[]; gaps: { questionId: string; brief: string }[] }
diff --git a/evidence/parallel-llm-fix/prompt-lab-run/01-job1-run.txt b/evidence/parallel-llm-fix/prompt-lab-run/01-job1-run.txt
new file mode 100644
index 000000000..fe29f9aff
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-run/01-job1-run.txt
@@ -0,0 +1,86 @@
+$ node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js run --no-observer-link --data-dir /tmp/flows-prompt-lab-data --local-agent prompt-lab.flow.ts --input {"job":"new-agency","reviewer":"prompt-lab-reviewer","lab":"/tmp/flows-prompt-lab-proof/lab","agency":"sunrise","visitType":"soc"}
+○ run-1 (deterministic) 0.00s
+✓ run-1 (deterministic) 0.09s completionReason: success
+○ llm-2 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CGZ7DHQPB1W0JJAA7J4QK" step "llm-2" (llm) is running under a worker lease until 1790142116388.
+↻ llm-2 (llm) 2.11s
+WAITING [worker_lease] Run "01M36CGZ7DHQPB1W0JJAA7J4QK" step "llm-2" (llm) is running under a worker lease until 1790142126392.
+↻ llm-2 (llm) 12.12s
+WAITING [worker_lease] Run "01M36CGZ7DHQPB1W0JJAA7J4QK" step "llm-2" (llm) is running under a worker lease until 1790142136397.
+↻ llm-2 (llm) 22.12s
+WAITING [worker_lease] Run "01M36CGZ7DHQPB1W0JJAA7J4QK" step "llm-2" (llm) is running under a worker lease until 1790142146407.
+↻ llm-2 (llm) 32.15s
+✓ llm-2 (llm) 36.16s completionReason: success
+○ llm-3 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CJ0G6YRB25A7T31HS3VPE" step "llm-3" (llm) is running under a worker lease until 1790142150461.
+↻ llm-3 (llm) 0.02s
+✓ llm-3 (llm) 6.21s completionReason: success
+○ run-4 (deterministic) 0.00s
+✓ run-4 (deterministic) 0.10s completionReason: success
+○ llm-5 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CJ6N8K5CE2DS40BS9Z7D6" step "llm-5" (llm) is running under a worker lease until 1790142156766.
+↻ llm-5 (llm) 0.02s
+WAITING [worker_lease] Run "01M36CJ6N8K5CE2DS40BS9Z7D6" step "llm-5" (llm) is running under a worker lease until 1790142166767.
+↻ llm-5 (llm) 10.02s
+WAITING [worker_lease] Run "01M36CJ6N8K5CE2DS40BS9Z7D6" step "llm-5" (llm) is running under a worker lease until 1790142176769.
+↻ llm-5 (llm) 20.02s
+✓ llm-5 (llm) 23.93s completionReason: success
+○ llm-6 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CJY16AV5R77J4EZWVW4HF" step "llm-6" (llm) is running under a worker lease until 1790142180700.
+↻ llm-6 (llm) 0.02s
+✓ llm-6 (llm) 4.56s completionReason: success
+○ run-7 (deterministic) 0.00s
+✓ run-7 (deterministic) 0.09s completionReason: success
+○ llm-8 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CK2J805CEYS2GY3ZQCMXR" step "llm-8" (llm) is running under a worker lease until 1790142185343.
+↻ llm-8 (llm) 0.01s
+WAITING [worker_lease] Run "01M36CK2J805CEYS2GY3ZQCMXR" step "llm-8" (llm) is running under a worker lease until 1790142195346.
+↻ llm-8 (llm) 10.01s
+✓ llm-8 (llm) 18.39s completionReason: success
+○ run-9 (deterministic) 0.00s
+✓ run-9 (deterministic) 0.09s completionReason: success
+○ llm-10 (llm) 0.00s
+○ llm-11 (llm) 0.00s
+○ llm-12 (llm) 0.00s
+○ llm-13 (llm) 0.00s
+○ llm-14 (llm) 0.00s
+○ llm-15 (llm) 0.00s
+○ llm-16 (llm) 0.00s
+○ llm-17 (llm) 0.00s
+○ llm-18 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CKMKNARNYRM1MR44ZZZTS" step "llm-10" (llm) is running under a worker lease until 1790142203819.
+↻ llm-10 (llm) 0.05s
+WAITING [worker_lease] Run "01M36CKMKWDXZKEPC20SAYR7A4" step "llm-11" (llm) is running under a worker lease until 1790142203828.
+↻ llm-11 (llm) 0.05s
+WAITING [worker_lease] Run "01M36CKMM3C5D595A9GN2R4728" step "llm-12" (llm) is running under a worker lease until 1790142203835.
+↻ llm-12 (llm) 0.05s
+WAITING [worker_lease] Run "01M36CKMMB96EFV04W6YMC1SNP" step "llm-13" (llm) is running under a worker lease until 1790142203844.
+↻ llm-13 (llm) 0.05s
+✓ llm-13 (llm) 2.42s completionReason: success
+WAITING [worker_lease] Run "01M36CKPZ4YF1RQZAEFD0NFZY4" step "llm-14" (llm) is running under a worker lease until 1790142206235.
+↻ llm-14 (llm) 2.43s
+✓ llm-12 (llm) 3.25s completionReason: success
+WAITING [worker_lease] Run "01M36CKQS75YSENHNFMA9TFP6A" step "llm-15" (llm) is running under a worker lease until 1790142207070.
+↻ llm-15 (llm) 3.27s
+✓ llm-11 (llm) 3.40s completionReason: success
+WAITING [worker_lease] Run "01M36CKQXZKW8JBET2SW0ZCC6Y" step "llm-16" (llm) is running under a worker lease until 1790142207221.
+↻ llm-16 (llm) 3.42s
+✓ llm-14 (llm) 5.53s completionReason: success
+WAITING [worker_lease] Run "01M36CKT0AJB4PY0Z4VGEVN0S4" step "llm-17" (llm) is running under a worker lease until 1790142209351.
+↻ llm-17 (llm) 5.60s
+✓ llm-15 (llm) 6.50s completionReason: success
+WAITING [worker_lease] Run "01M36CKTYT88664BJ884HWWTS1" step "llm-18" (llm) is running under a worker lease until 1790142210323.
+↻ llm-18 (llm) 6.52s
+✓ llm-16 (llm) 7.14s completionReason: success
+✓ llm-18 (llm) 9.18s completionReason: success
+✓ llm-10 (llm) 9.54s completionReason: success
+✓ llm-17 (llm) 9.85s completionReason: success
+○ run-19 (deterministic) 0.00s
+✓ run-19 (deterministic) 0.09s completionReason: success
+○ human-20 (deterministic) 0.00s
+⏸ human-20 (human) 0.01s
+PARKED [run_parked] Run "01M36CGX2HC59PAVGKV0SCFW6S" is waiting for prompt-lab-reviewer to answer human-20: "Config workbench · sunrise soc: 9 rows on 3 questions (6 highlighted, listed first); 1 gap brief(s) queued for the test patient manager.\nEdit target answer / confidence / explanation / notes in /tmp/flows-prompt-lab-proof/lab/work/sunrise-soc/e4a1ccb3/grid.json. Your first pass persists as the target for each question × patient.\nyes = persist targets and run iteration on every changed row; no = stop without persisting."
+Answer with: flows answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-20 yes|no
+Then continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+RUN 01M36CGX2HC59PAVGKV0SCFW6S parked
+exit=3
diff --git a/evidence/parallel-llm-fix/prompt-lab-run/02-reviewer-edit.diff b/evidence/parallel-llm-fix/prompt-lab-run/02-reviewer-edit.diff
new file mode 100644
index 000000000..9ba0e3939
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-run/02-reviewer-edit.diff
@@ -0,0 +1,13 @@
+# reviewer edit: /tmp/flows-prompt-lab-proof/lab/work/sunrise-soc/e4a1ccb3/grid.json
+@@ -144,9 +144,9 @@
+     "target": {
+       "answer": "Ongoing",
+       "confidence": "High",
+-      "explanation": "Today's SOC visit documents the left heel as an open area 2.0 x 1.5 cm with moderate serous drainage and a pink wound bed with yellow slough, so the wound is not healed despite the discharge summary stating it was closed. The current direct assessment supersedes the stale referral statement."
++      "explanation": "Today's visit notes describe an open 2.0 x 1.5 cm left heel wound with serous drainage and slough; the discharge summary's closed status is outdated."
+     },
+-    "notes": ""
++    "notes": "The prompt makes the referral win over what the nurse saw today. Today's notes must win (shared guideline 1)."
+   },
+   {
+     "questionId": "wound-status",
diff --git a/evidence/parallel-llm-fix/prompt-lab-run/03-answer.txt b/evidence/parallel-llm-fix/prompt-lab-run/03-answer.txt
new file mode 100644
index 000000000..cc4b074d6
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-run/03-answer.txt
@@ -0,0 +1,4 @@
+$ node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-20 yes --by prompt-lab-reviewer
+ANSWERED 01M36CGX2HC59PAVGKV0SCFW6S human-20 yes
+Continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+exit=0
diff --git a/evidence/parallel-llm-fix/prompt-lab-run/04-resume.txt b/evidence/parallel-llm-fix/prompt-lab-run/04-resume.txt
new file mode 100644
index 000000000..df359df5a
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-run/04-resume.txt
@@ -0,0 +1,62 @@
+$ node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js resume --no-observer-link --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+○ run-1 (deterministic) 0.00s
+✓ run-1 (deterministic) 0.01s completionReason: success
+○ llm-2 (llm) 0.00s
+✓ llm-2 (llm) 1.84s completionReason: success
+○ llm-3 (llm) 0.00s
+✓ llm-3 (llm) 0.01s completionReason: success
+○ run-4 (deterministic) 0.00s
+✓ run-4 (deterministic) 0.01s completionReason: success
+○ llm-5 (llm) 0.00s
+✓ llm-5 (llm) 0.01s completionReason: success
+○ llm-6 (llm) 0.00s
+✓ llm-6 (llm) 0.01s completionReason: success
+○ run-7 (deterministic) 0.00s
+✓ run-7 (deterministic) 0.01s completionReason: success
+○ llm-8 (llm) 0.00s
+✓ llm-8 (llm) 0.01s completionReason: success
+○ run-9 (deterministic) 0.00s
+✓ run-9 (deterministic) 0.01s completionReason: success
+○ llm-10 (llm) 0.00s
+○ llm-11 (llm) 0.00s
+○ llm-12 (llm) 0.00s
+○ llm-13 (llm) 0.00s
+○ llm-14 (llm) 0.00s
+○ llm-15 (llm) 0.00s
+○ llm-16 (llm) 0.00s
+○ llm-17 (llm) 0.00s
+○ llm-18 (llm) 0.00s
+✓ llm-10 (llm) 0.04s completionReason: success
+✓ llm-11 (llm) 0.04s completionReason: success
+✓ llm-12 (llm) 0.05s completionReason: success
+✓ llm-13 (llm) 0.05s completionReason: success
+✓ llm-14 (llm) 0.09s completionReason: success
+✓ llm-15 (llm) 0.09s completionReason: success
+✓ llm-16 (llm) 0.10s completionReason: success
+✓ llm-17 (llm) 0.10s completionReason: success
+✓ llm-18 (llm) 0.11s completionReason: success
+○ run-19 (deterministic) 0.00s
+✓ run-19 (deterministic) 0.02s completionReason: success
+○ human-20 (deterministic) 0.00s
+✓ human-20 (deterministic) 0.03s completionReason: success
+○ run-21 (deterministic) 0.00s
+✓ run-21 (deterministic) 0.09s completionReason: success
+○ run-22 (deterministic) 0.00s
+✓ run-22 (deterministic) 0.09s completionReason: success
+○ llm-23 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CM1MRES6V92VT42Q0RZ7H" step "llm-23" (llm) is running under a worker lease until 1790142217166.
+↻ llm-23 (llm) 0.02s
+WAITING [worker_lease] Run "01M36CM1MRES6V92VT42Q0RZ7H" step "llm-23" (llm) is running under a worker lease until 1790142227169.
+↻ llm-23 (llm) 10.05s
+✓ llm-23 (llm) 10.37s completionReason: success
+○ run-24 (deterministic) 0.00s
+✓ run-24 (deterministic) 0.10s completionReason: success
+○ run-25 (deterministic) 0.00s
+✓ run-25 (deterministic) 0.10s completionReason: success
+○ human-26 (deterministic) 0.00s
+⏸ human-26 (human) 0.01s
+PARKED [run_parked] Run "01M36CGX2HC59PAVGKV0SCFW6S" is waiting for prompt-lab-reviewer to answer human-26: "Commit output · sunrise: agency-specific prompts ready to go live in Apricot: living-situation.\nSent to the question manager (frozen here): wound-status → config-sunrise-wound-status-7309e97c.\nHeld as drafts until a shelf patient covers them: ostomy-supplies.\nTo commit only some, edit /tmp/flows-prompt-lab-proof/lab/work/sunrise-soc/e4a1ccb3/commit.json: {\"mode\":\"only\"|\"except\",\"questions\":[...]}.\nyes = commit (Done = live, no promote step); no = leave them as drafts."
+Answer with: flows answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-26 yes|no
+Then continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+RUN 01M36CGX2HC59PAVGKV0SCFW6S parked
+exit=3
diff --git a/evidence/parallel-llm-fix/prompt-lab-run/05-answer.txt b/evidence/parallel-llm-fix/prompt-lab-run/05-answer.txt
new file mode 100644
index 000000000..3bf84768b
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-run/05-answer.txt
@@ -0,0 +1,4 @@
+$ node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-26 yes --by prompt-lab-reviewer
+ANSWERED 01M36CGX2HC59PAVGKV0SCFW6S human-26 yes
+Continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+exit=0
diff --git a/evidence/parallel-llm-fix/prompt-lab-run/06-resume.txt b/evidence/parallel-llm-fix/prompt-lab-run/06-resume.txt
new file mode 100644
index 000000000..9898998f3
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab-run/06-resume.txt
@@ -0,0 +1,60 @@
+$ node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js resume --no-observer-link --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+○ run-1 (deterministic) 0.00s
+✓ run-1 (deterministic) 0.08s completionReason: success
+○ llm-2 (llm) 0.00s
+✓ llm-2 (llm) 1.85s completionReason: success
+○ llm-3 (llm) 0.00s
+✓ llm-3 (llm) 0.02s completionReason: success
+○ run-4 (deterministic) 0.00s
+✓ run-4 (deterministic) 0.02s completionReason: success
+○ llm-5 (llm) 0.00s
+✓ llm-5 (llm) 0.02s completionReason: success
+○ llm-6 (llm) 0.00s
+✓ llm-6 (llm) 0.02s completionReason: success
+○ run-7 (deterministic) 0.00s
+✓ run-7 (deterministic) 0.02s completionReason: success
+○ llm-8 (llm) 0.00s
+✓ llm-8 (llm) 0.02s completionReason: success
+○ run-9 (deterministic) 0.00s
+✓ run-9 (deterministic) 0.02s completionReason: success
+○ llm-10 (llm) 0.00s
+○ llm-11 (llm) 0.00s
+○ llm-12 (llm) 0.00s
+○ llm-13 (llm) 0.00s
+○ llm-14 (llm) 0.00s
+○ llm-15 (llm) 0.00s
+○ llm-16 (llm) 0.00s
+○ llm-17 (llm) 0.00s
+○ llm-18 (llm) 0.00s
+✓ llm-10 (llm) 0.05s completionReason: success
+✓ llm-11 (llm) 0.06s completionReason: success
+✓ llm-12 (llm) 0.06s completionReason: success
+✓ llm-13 (llm) 0.07s completionReason: success
+✓ llm-14 (llm) 0.12s completionReason: success
+✓ llm-15 (llm) 0.12s completionReason: success
+✓ llm-16 (llm) 0.13s completionReason: success
+✓ llm-17 (llm) 0.14s completionReason: success
+✓ llm-18 (llm) 0.15s completionReason: success
+○ run-19 (deterministic) 0.00s
+✓ run-19 (deterministic) 0.02s completionReason: success
+○ human-20 (deterministic) 0.00s
+✓ human-20 (deterministic) 0.02s completionReason: success
+○ run-21 (deterministic) 0.00s
+✓ run-21 (deterministic) 0.02s completionReason: success
+○ run-22 (deterministic) 0.00s
+✓ run-22 (deterministic) 0.02s completionReason: success
+○ llm-23 (llm) 0.00s
+✓ llm-23 (llm) 0.02s completionReason: success
+○ run-24 (deterministic) 0.00s
+✓ run-24 (deterministic) 0.02s completionReason: success
+○ run-25 (deterministic) 0.00s
+✓ run-25 (deterministic) 0.02s completionReason: success
+○ human-26 (deterministic) 0.00s
+✓ human-26 (deterministic) 0.04s completionReason: success
+○ run-27 (deterministic) 0.00s
+✓ run-27 (deterministic) 0.10s completionReason: success
+○ run-28 (deterministic) 0.00s
+✓ run-28 (deterministic) 0.10s completionReason: success
+FAILED [protocol_error] relayflowd could not complete the resume request: unawaited_step: flow "prompt-lab" returned with unawaited steps: llm-10 (f.llm), llm-11 (f.llm), llm-13 (f.llm), llm-14 (f.llm)
+RUN 01M36CGX2HC59PAVGKV0SCFW6S unknown
+exit=1
diff --git a/evidence/parallel-llm-fix/prompt-lab.txt b/evidence/parallel-llm-fix/prompt-lab.txt
new file mode 100644
index 000000000..553cede0b
--- /dev/null
+++ b/evidence/parallel-llm-fix/prompt-lab.txt
@@ -0,0 +1,23 @@
+{"seeded":"/tmp/flows-prompt-lab-proof/lab"}
+== Job 1: new agency sunrise / soc
+Answer with: flows answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-20 yes|no
+Then continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+RUN 01M36CGX2HC59PAVGKV0SCFW6S parked
+exit=3
+$ node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-20 yes --by prompt-lab-reviewer
+ANSWERED 01M36CGX2HC59PAVGKV0SCFW6S human-20 yes
+Continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+exit=0
+Answer with: flows answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-26 yes|no
+Then continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+RUN 01M36CGX2HC59PAVGKV0SCFW6S parked
+exit=3
+$ node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js answer --data-dir /tmp/flows-prompt-lab-data 01M36CGX2HC59PAVGKV0SCFW6S human-26 yes --by prompt-lab-reviewer
+ANSWERED 01M36CGX2HC59PAVGKV0SCFW6S human-26 yes
+Continue with: flows resume --data-dir /tmp/flows-prompt-lab-data --local-agent 01M36CGX2HC59PAVGKV0SCFW6S
+exit=0
+✓ run-28 (deterministic) 0.10s completionReason: success
+FAILED [protocol_error] relayflowd could not complete the resume request: unawaited_step: flow "prompt-lab" returned with unawaited steps: llm-10 (f.llm), llm-11 (f.llm), llm-13 (f.llm), llm-14 (f.llm)
+RUN 01M36CGX2HC59PAVGKV0SCFW6S unknown
+exit=1
+prove: /tmp/flows-prompt-lab-proof/06-resume.txt exited 1, expected 0
diff --git a/evidence/parallel-llm-fix/prove-fixed.sh b/evidence/parallel-llm-fix/prove-fixed.sh
new file mode 100644
index 000000000..a37df7233
--- /dev/null
+++ b/evidence/parallel-llm-fix/prove-fixed.sh
@@ -0,0 +1,103 @@
+#!/usr/bin/env bash
+# End-to-end proof of the Prompt Lab relayflow, locally, with real model calls.
+#
+# Seeds a fresh lab from fixtures/, then drives all three jobs through the real
+# kernel with `flows run --local-agent`. At every parked `f.human` gate this
+# script plays the reviewer named in the input: it applies the documented edit
+# (captured as a diff), answers with `flows answer`, and `flows resume`s.
+# Every command is captured with its literal output and exit code, and the
+# script stops on the first exit it did not expect.
+#
+#   ./prove.sh [out-dir]        default: evidence/run
+set -uo pipefail
+cd "$(dirname "$0")"
+OUT=${1:-evidence/run}
+LAB=$OUT/lab
+DD=${FLOWS_DATA_DIR:-$(mktemp -d)}
+REVIEWER=prompt-lab-reviewer
+N=0
+STATUS=
+FILE=
+
+die() { echo "prove: $*" >&2; exit 1; }
+[ -e "$LAB" ] && die "refusing: $LAB exists"
+mkdir -p "$OUT"
+node --no-warnings --experimental-strip-types store.ts "$LAB" seed fixtures || die "seed failed"
+
+# capture <expected-exit...> -- <name> <command...>: run it, write "$ cmd", its
+# output and exit code to $OUT/NN-name.txt, and stop unless the exit is one of
+# the expected ones (0 completed, 3 parked on a gate).
+capture() {
+  local want=()
+  while [ "$1" != "--" ]; do want+=("$1"); shift; done
+  shift
+  local name=$1
+  shift
+  N=$((N + 1))
+  FILE=$(printf '%s/%02d-%s.txt' "$OUT" "$N" "$name")
+  { echo "\$ $*"; "$@" 2>&1; STATUS=$?; echo "exit=$STATUS"; } > "$FILE"
+  grep -v 'WAITING\|↻\|○' "$FILE" | tail -4 | cut -c1-240
+  local w
+  for w in "${want[@]}"; do [ "$STATUS" = "$w" ] && return 0; done
+  die "$FILE exited $STATUS, expected ${want[*]}"
+}
+run_id()  { grep -o 'RUN [0-9A-Z]\{26\}' "$FILE" | tail -1 | cut -d' ' -f2; }
+wait_id() { grep -o ' human-[0-9]* yes|no' "$FILE" | tail -1 | awk '{print $1}'; }
+parked_on() { grep -q "PARKED.*$1" "$FILE" || die "$FILE did not park on \"$1\""; }
+gate_path() { grep -o "$LAB/work/[^ ]*\.json" "$FILE" | head -1; }
+# edit <file> <node-expression over v>: the reviewer's change, captured as a diff
+edit() {
+  local file=$1 expr=$2 before
+  [ -f "$file" ] || die "no file to edit: $file"
+  before=$(mktemp)
+  cp "$file" "$before"
+  node -e "const fs=require('fs');const v=JSON.parse(fs.readFileSync('$file','utf8'));$expr;fs.writeFileSync('$file',JSON.stringify(v,null,2)+'\n')" \
+    || die "reviewer edit failed on $file"
+  N=$((N + 1))
+  { echo "# reviewer edit: $file"; diff -u "$before" "$file" | tail -n +3; } > "$(printf '%s/%02d-reviewer-edit.diff' "$OUT" "$N")"
+}
+# answer_and_resume <expected-exit>: 3 when another gate follows, 0 when the run should complete.
+answer_and_resume() {
+  local run wait
+  run=$(run_id)
+  wait=$(wait_id)
+  [ -n "$run" ] && [ -n "$wait" ] || die "no parked gate in $FILE"
+  capture 0 -- answer node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js answer --data-dir "$DD" "$run" "$wait" yes --by "$REVIEWER"
+  capture "$1" -- resume node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js resume --no-observer-link --data-dir "$DD" --local-agent "$run"
+}
+flow() { capture 3 -- "$1" node /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js run --no-observer-link --data-dir "$DD" --local-agent prompt-lab.flow.ts --input "$2"; }
+
+echo "== Job 1: new agency sunrise / soc"
+flow job1-run "{\"job\":\"new-agency\",\"reviewer\":\"$REVIEWER\",\"lab\":\"$LAB\",\"agency\":\"sunrise\",\"visitType\":\"soc\"}"
+parked_on "Config workbench"
+# The reviewer's first pass: Pat's wound is open today, whatever the discharge summary said.
+edit "$(gate_path)" 'const r=v.find(r=>r.questionId==="wound-status"&&r.patientId==="pat");r.target={answer:"Ongoing",confidence:"High",explanation:"Today'"'"'s visit notes describe an open 2.0 x 1.5 cm left heel wound with serous drainage and slough; the discharge summary'"'"'s closed status is outdated."};r.notes="The prompt makes the referral win over what the nurse saw today. Today'"'"'s notes must win (shared guideline 1)."'
+answer_and_resume 3
+parked_on "Commit output"
+# Only covered prompts are offered; ostomy-supplies (no shelf patient yet) is held as a draft.
+grep -q "Held as drafts until a shelf patient covers them: ostomy-supplies" "$FILE" || die "ostomy-supplies was offered for commit"
+answer_and_resume 0   # commit all offered
+
+echo "== Test patient creator: the ostomy gap brief Job 1 queued"
+flow patient-run "{\"job\":\"patient\",\"reviewer\":\"$REVIEWER\",\"lab\":\"$LAB\",\"briefId\":\"gap-ostomy-supplies-soc\"}"
+parked_on "Test patient creator"
+answer_and_resume 0   # kick generate; Patient QA locks it, nobody approves the chart
+
+echo "== Job 2: fix wound-status from the config-send issue"
+ISSUE=$(node -e "const q=require('./$LAB/queue/issues.json');console.log(q.find(i=>i.questionIds[0]==='wound-status').id)") \
+  || die "no wound-status issue was sent to the question manager"
+flow job2-run "{\"job\":\"fix\",\"reviewer\":\"$REVIEWER\",\"lab\":\"$LAB\",\"issueId\":\"$ISSUE\"}"
+parked_on "Question workbench"
+answer_and_resume 3   # gold as prefilled: Pat's config-level target travelled with the issue
+parked_on "Re-run and score"
+answer_and_resume 0   # mark done = live, for every agency using it
+
+echo "== Final lab state"
+capture 0 -- lab-state node -e "
+const b=require('./$LAB/bank.json');
+for (const [q,v] of Object.entries(b.questions)) console.log(q.padEnd(17),'live',String(v.livePromptId).padEnd(30),'draft',v.draftPromptId??'-');
+console.log('shelf', require('fs').readdirSync('$LAB/shelf').join(' '));
+console.log('issues', JSON.stringify(require('./$LAB/queue/issues.json').map(i=>[i.id,i.status])));
+console.log('patient briefs', JSON.stringify(require('./$LAB/queue/patient-briefs.json').map(i=>[i.id,i.status])));
+console.log('gold', Object.keys(require('./$LAB/gold.json')).join(' '));"
+echo "data dir: $DD"
diff --git a/evidence/parallel-llm-fix/regression-pass.txt b/evidence/parallel-llm-fix/regression-pass.txt
new file mode 100644
index 000000000..7b7a4429a
--- /dev/null
+++ b/evidence/parallel-llm-fix/regression-pass.txt
@@ -0,0 +1,16 @@
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/authored-parallel-llm.test.ts (6 tests) 114270ms
+   ✓ parallel llm capacity 1 > deduplicates concurrent preflight probes 3122ms
+   ✓ parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 8803ms
+   ✓ parallel llm capacity 1 > keeps the durable root lease alive across two cold models 47739ms
+   ✓ parallel llm capacity 4 > deduplicates concurrent preflight probes 1178ms
+   ✓ parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 7141ms
+   ✓ parallel llm capacity 4 > keeps the durable root lease alive across two cold models 46286ms
+
+ Test Files  1 passed (1)
+      Tests  6 passed (6)
+   Start at  05:40:14
+   Duration  115.45s (transform 576ms, setup 0ms, collect 1.00s, tests 114.27s, environment 0ms, prepare 50ms)
+
diff --git a/evidence/parallel-llm-fix/regression.txt b/evidence/parallel-llm-fix/regression.txt
new file mode 100644
index 000000000..106d88f2c
--- /dev/null
+++ b/evidence/parallel-llm-fix/regression.txt
@@ -0,0 +1,87 @@
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ❯ tests/authored-parallel-llm.test.ts (6 tests | 6 failed) 113509ms
+   × parallel llm capacity 1 > deduplicates concurrent preflight probes 2515ms
+     → expected [ …(10) ] to have a length of 9 but got 10
+   × parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 8547ms
+     → expected [ …(10) ] to have a length of 9 but got 10
+   × parallel llm capacity 1 > keeps the durable root lease alive across two cold models 47711ms
+     → expected [ …(11) ] to have a length of 10 but got 11
+   × parallel llm capacity 4 > deduplicates concurrent preflight probes 1189ms
+     → expected [ …(10) ] to have a length of 9 but got 10
+   × parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 7166ms
+     → expected [ …(10) ] to have a length of 9 but got 10
+   × parallel llm capacity 4 > keeps the durable root lease alive across two cold models 46380ms
+     → expected [ …(11) ] to have a length of 10 but got 11
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 6 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > deduplicates concurrent preflight probes
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 4 > deduplicates concurrent preflight probes
+AssertionError: expected [ …(10) ] to have a length of 9 but got 10
+
+- Expected
++ Received
+
+- 9
++ 10
+
+ ❯ Object.assertJournals tests/authored-parallel-llm.test.ts:51:17
+     49|   async function assertJournals(expected: number) {
+     50|     const ids = readdirSync(join(fixture.data, 'runs')).filter(name =>…
+     51|     expect(ids).toHaveLength(expected);
+       |                 ^
+     52|     const expired: unknown[] = [];
+     53|     for (const name of ids) {
+ ❯ tests/authored-parallel-llm.test.ts:88:16
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/6]⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight
+AssertionError: expected [ …(10) ] to have a length of 9 but got 10
+
+- Expected
++ Received
+
+- 9
++ 10
+
+ ❯ Object.assertJournals tests/authored-parallel-llm.test.ts:51:17
+     49|   async function assertJournals(expected: number) {
+     50|     const ids = readdirSync(join(fixture.data, 'runs')).filter(name =>…
+     51|     expect(ids).toHaveLength(expected);
+       |                 ^
+     52|     const expired: unknown[] = [];
+     53|     for (const name of ids) {
+ ❯ tests/authored-parallel-llm.test.ts:97:16
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/6]⎯
+
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 1 > keeps the durable root lease alive across two cold models
+ FAIL  tests/authored-parallel-llm.test.ts > parallel llm capacity 4 > keeps the durable root lease alive across two cold models
+AssertionError: expected [ …(11) ] to have a length of 10 but got 11
+
+- Expected
++ Received
+
+- 10
++ 11
+
+ ❯ Object.assertJournals tests/authored-parallel-llm.test.ts:51:17
+     49|   async function assertJournals(expected: number) {
+     50|     const ids = readdirSync(join(fixture.data, 'runs')).filter(name =>…
+     51|     expect(ids).toHaveLength(expected);
+       |                 ^
+     52|     const expired: unknown[] = [];
+     53|     for (const name of ids) {
+ ❯ tests/authored-parallel-llm.test.ts:119:16
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[3/6]⎯
+
+ Test Files  1 failed (1)
+      Tests  6 failed (6)
+   Start at  05:37:50
+   Duration  114.67s (transform 568ms, setup 0ms, collect 985ms, tests 113.51s, environment 0ms, prepare 50ms)
+
diff --git a/evidence/parallel-llm-fix/related-pass.txt b/evidence/parallel-llm-fix/related-pass.txt
new file mode 100644
index 000000000..4ba8caee0
--- /dev/null
+++ b/evidence/parallel-llm-fix/related-pass.txt
@@ -0,0 +1,24 @@
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/preflight.test.ts (67 tests) 117ms
+ ✓ tests/cli-probe.test.ts (6 tests) 858ms
+ ✓ tests/communication-environment-preflight.test.ts (6 tests) 4ms
+ ✓ tests/cli.test.ts (70 tests) 1884ms
+   ✓ flows check CLI > binds a checked relative wrapper to the flow directory for worker execution 440ms
+   ✓ flows check CLI > resolves a bare PATH-resolved claude with no declared model, in an isolated PATH 420ms
+ ✓ tests/authored-parallel-agents.test.ts (8 tests) 12632ms
+   ✓ authored steps under local workers with capacity > runs more concurrent f.llm calls than the worker holds side by side, never more than its capacity 1195ms
+   ✓ authored steps under local workers with capacity > completes more concurrent f.agent calls than the worker holds: the overflow waits for a slot instead of parking 2956ms
+   ✓ authored steps under local workers with capacity > runs agents in distinct working directories side by side (the kernel carries cwd) 637ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a symlink alias of the same directory 1926ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a directory nested inside the other 1046ms
+   ✓ authored steps under local workers with capacity > never starts queued agents once the body has failed 1915ms
+   ✓ authored steps under local workers with capacity > never starts a queued agent when the agent holding the only slot fails 1974ms
+   ✓ authored steps under local workers with capacity > parks the overflow when the body is not told the capacity (the defect this closes) 982ms
+
+ Test Files  5 passed (5)
+      Tests  157 passed (157)
+   Start at  05:39:44
+   Duration  13.97s (transform 1.47s, setup 0ms, collect 4.15s, tests 15.49s, environment 1ms, prepare 254ms)
+
diff --git a/evidence/parallel-llm-fix/related.txt b/evidence/parallel-llm-fix/related.txt
new file mode 100644
index 000000000..32ff28569
--- /dev/null
+++ b/evidence/parallel-llm-fix/related.txt
@@ -0,0 +1,50 @@
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/preflight.test.ts (67 tests) 104ms
+ ❯ tests/cli-probe.test.ts (6 tests | 1 failed) 812ms
+   × reports signal termination in both drivers 101ms
+     → expected Error: CLI probe failed { detail: 's…' } to match object { detail: 'signal:SIGTERM' }
+ ✓ tests/communication-environment-preflight.test.ts (6 tests) 5ms
+ ✓ tests/cli.test.ts (70 tests) 1710ms
+   ✓ flows check CLI > binds a checked relative wrapper to the flow directory for worker execution 330ms
+   ✓ flows check CLI > resolves a bare PATH-resolved claude with no declared model, in an isolated PATH 389ms
+ ✓ tests/authored-parallel-agents.test.ts (8 tests) 11090ms
+   ✓ authored steps under local workers with capacity > runs more concurrent f.llm calls than the worker holds side by side, never more than its capacity 1130ms
+   ✓ authored steps under local workers with capacity > completes more concurrent f.agent calls than the worker holds: the overflow waits for a slot instead of parking 2560ms
+   ✓ authored steps under local workers with capacity > runs agents in distinct working directories side by side (the kernel carries cwd) 626ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a symlink alias of the same directory 1038ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a directory nested inside the other 1033ms
+   ✓ authored steps under local workers with capacity > never starts queued agents once the body has failed 1914ms
+   ✓ authored steps under local workers with capacity > never starts a queued agent when the agent holding the only slot fails 1915ms
+   ✓ authored steps under local workers with capacity > parks the overflow when the body is not told the capacity (the defect this closes) 872ms
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/cli-probe.test.ts > reports signal termination in both drivers
+AssertionError: expected Error: CLI probe failed { detail: 's…' } to match object { detail: 'signal:SIGTERM' }
+
+- Expected
++ Received
+
+- Object {
+-   "detail": "signal:SIGTERM",
++ CliProbeError {
++   "detail": "spawn_failed",
+  }
+
+ ❯ tests/cli-probe.test.ts:61:3
+     59|   const { path, directory } = wrapper(identify + "process.kill(process…
+     60|   expect(() => probeCli(path, directory, 'test-model')).toThrow(expect…
+     61|   await expect(probeCliAsync(path, directory, 'test-model')).rejects.t…
+       |   ^
+     62| });
+     63| 
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯
+
+ Test Files  1 failed | 4 passed (5)
+      Tests  1 failed | 156 passed (157)
+   Start at  05:39:17
+   Duration  12.24s (transform 1.21s, setup 0ms, collect 3.62s, tests 13.72s, environment 1ms, prepare 202ms)
+
diff --git a/evidence/parallel-llm-fix/repro-capacity1.txt b/evidence/parallel-llm-fix/repro-capacity1.txt
new file mode 100644
index 000000000..9e486d9c2
--- /dev/null
+++ b/evidence/parallel-llm-fix/repro-capacity1.txt
@@ -0,0 +1,41 @@
+○ run-1 (deterministic) 0.00s
+✓ run-1 (deterministic) 0.02s completionReason: success
+○ llm-2 (llm) 0.00s
+○ llm-3 (llm) 0.00s
+○ llm-4 (llm) 0.00s
+○ llm-5 (llm) 0.00s
+○ llm-6 (llm) 0.00s
+○ llm-7 (llm) 0.00s
+○ llm-8 (llm) 0.00s
+○ llm-9 (llm) 0.00s
+○ llm-10 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CGEMJRJG2HRVZJ9C7CQB2" step "llm-2" (llm) is running under a worker lease until 1790142099406.
+↻ llm-2 (llm) 2.29s
+✓ llm-2 (llm) 4.25s completionReason: success
+WAITING [worker_lease] Run "01M36CGGJD3NJP04H66RJW8MDE" step "llm-3" (llm) is running under a worker lease until 1790142101380.
+↻ llm-3 (llm) 4.26s
+✓ llm-3 (llm) 6.57s completionReason: success
+WAITING [worker_lease] Run "01M36CGJV2FPXESM9MNPAHSMXA" step "llm-4" (llm) is running under a worker lease until 1790142103704.
+↻ llm-4 (llm) 6.59s
+✓ llm-4 (llm) 9.27s completionReason: success
+WAITING [worker_lease] Run "01M36CGNFHQ2EH4GH0TRR9XGYB" step "llm-5" (llm) is running under a worker lease until 1790142106407.
+↻ llm-5 (llm) 9.29s
+✓ llm-5 (llm) 11.70s completionReason: success
+WAITING [worker_lease] Run "01M36CGQV9QAK5Q8R2NT0Z83KE" step "llm-6" (llm) is running under a worker lease until 1790142108833.
+↻ llm-6 (llm) 11.72s
+✓ llm-6 (llm) 14.47s completionReason: success
+WAITING [worker_lease] Run "01M36CGTHWWRQRJHYXFHP8MCVR" step "llm-7" (llm) is running under a worker lease until 1790142111602.
+↻ llm-7 (llm) 14.48s
+✓ llm-7 (llm) 17.31s completionReason: success
+WAITING [worker_lease] Run "01M36CGXASCMH6YZDPW98WG2NN" step "llm-8" (llm) is running under a worker lease until 1790142114446.
+↻ llm-8 (llm) 17.33s
+✓ llm-8 (llm) 19.75s completionReason: success
+WAITING [worker_lease] Run "01M36CGZPRH7QYGD3ESRAFY544" step "llm-9" (llm) is running under a worker lease until 1790142116878.
+↻ llm-9 (llm) 19.76s
+✓ llm-9 (llm) 22.22s completionReason: success
+WAITING [worker_lease] Run "01M36CH241FY5H47NJ7JHT3ZBZ" step "llm-10" (llm) is running under a worker lease until 1790142119351.
+↻ llm-10 (llm) 22.23s
+✓ llm-10 (llm) 25.01s completionReason: success
+○ run-11 (deterministic) 0.00s
+✓ run-11 (deterministic) 0.02s completionReason: success
+RUN 01M36CGCC7T4X6P1XRRKM3BBSF completed (12 steps) completionReason: success
diff --git a/evidence/parallel-llm-fix/repro-capacity4.txt b/evidence/parallel-llm-fix/repro-capacity4.txt
new file mode 100644
index 000000000..e67a6265d
--- /dev/null
+++ b/evidence/parallel-llm-fix/repro-capacity4.txt
@@ -0,0 +1,41 @@
+○ run-1 (deterministic) 0.00s
+✓ run-1 (deterministic) 0.02s completionReason: success
+○ llm-2 (llm) 0.00s
+○ llm-3 (llm) 0.00s
+○ llm-4 (llm) 0.00s
+○ llm-5 (llm) 0.00s
+○ llm-6 (llm) 0.00s
+○ llm-7 (llm) 0.00s
+○ llm-8 (llm) 0.00s
+○ llm-9 (llm) 0.00s
+○ llm-10 (llm) 0.00s
+WAITING [worker_lease] Run "01M36CF7BTGJEMYX64R70QA85F" step "llm-2" (llm) is running under a worker lease until 1790142059197.
+↻ llm-2 (llm) 3.18s
+WAITING [worker_lease] Run "01M36CF7C24EJ27MX7Q0GRHB3R" step "llm-3" (llm) is running under a worker lease until 1790142059198.
+↻ llm-3 (llm) 3.18s
+WAITING [worker_lease] Run "01M36CF7CADB4QKY9MY3ANKYC2" step "llm-4" (llm) is running under a worker lease until 1790142059205.
+↻ llm-4 (llm) 3.18s
+WAITING [worker_lease] Run "01M36CF7CMQFZWYJGC0JQM0AMR" step "llm-5" (llm) is running under a worker lease until 1790142059210.
+↻ llm-5 (llm) 3.18s
+✓ llm-5 (llm) 5.60s completionReason: success
+WAITING [worker_lease] Run "01M36CF9RWTK7DTRCE76JMQYCX" step "llm-6" (llm) is running under a worker lease until 1790142061650.
+↻ llm-6 (llm) 5.61s
+✓ llm-2 (llm) 5.98s completionReason: success
+✓ llm-3 (llm) 5.99s completionReason: success
+WAITING [worker_lease] Run "01M36CFA4TRJSAMKXCNF073WH0" step "llm-7" (llm) is running under a worker lease until 1790142062033.
+↻ llm-7 (llm) 6.00s
+WAITING [worker_lease] Run "01M36CFA5CWS9RR3N5X7T7K7YS" step "llm-8" (llm) is running under a worker lease until 1790142062050.
+↻ llm-8 (llm) 6.01s
+✓ llm-4 (llm) 6.02s completionReason: success
+WAITING [worker_lease] Run "01M36CFA640PMA3JP60HNHXQ1Z" step "llm-9" (llm) is running under a worker lease until 1790142062075.
+↻ llm-9 (llm) 6.03s
+✓ llm-6 (llm) 7.71s completionReason: success
+WAITING [worker_lease] Run "01M36CFBV0WH9DNMJBQAW05J9D" step "llm-10" (llm) is running under a worker lease until 1790142063766.
+↻ llm-10 (llm) 7.72s
+✓ llm-9 (llm) 8.61s completionReason: success
+✓ llm-7 (llm) 8.98s completionReason: success
+✓ llm-8 (llm) 9.14s completionReason: success
+✓ llm-10 (llm) 10.13s completionReason: success
+○ run-11 (deterministic) 0.00s
+✓ run-11 (deterministic) 0.02s completionReason: success
+RUN 01M36CF44N8Q424XQZSGEA9EKJ completed (12 steps) completionReason: success
diff --git a/evidence/parallel-llm-fix/repro-final-capacity1.txt b/evidence/parallel-llm-fix/repro-final-capacity1.txt
new file mode 100644
index 000000000..024e93d1b
--- /dev/null
+++ b/evidence/parallel-llm-fix/repro-final-capacity1.txt
@@ -0,0 +1,41 @@
+○ run-1 (deterministic) 0.00s
+✓ run-1 (deterministic) 0.03s completionReason: success
+○ llm-2 (llm) 0.00s
+○ llm-3 (llm) 0.00s
+○ llm-4 (llm) 0.00s
+○ llm-5 (llm) 0.00s
+○ llm-6 (llm) 0.00s
+○ llm-7 (llm) 0.00s
+○ llm-8 (llm) 0.00s
+○ llm-9 (llm) 0.00s
+○ llm-10 (llm) 0.00s
+WAITING [worker_lease] Run "01M36DBY2Y33WH7GG9ZW4HN16K" step "llm-2" (llm) is running under a worker lease until 1790143000156.
+↻ llm-2 (llm) 2.36s
+✓ llm-2 (llm) 5.17s completionReason: success
+WAITING [worker_lease] Run "01M36DC12SFN3XRP3XZS5DZPMY" step "llm-3" (llm) is running under a worker lease until 1790143003027.
+↻ llm-3 (llm) 5.17s
+✓ llm-3 (llm) 7.56s completionReason: success
+WAITING [worker_lease] Run "01M36DC3EHT0GA8EC3RS4ZZZ2A" step "llm-4" (llm) is running under a worker lease until 1790143005448.
+↻ llm-4 (llm) 7.56s
+✓ llm-4 (llm) 10.55s completionReason: success
+WAITING [worker_lease] Run "01M36DC6CQDQ7BB1FWV1JCMEK6" step "llm-5" (llm) is running under a worker lease until 1790143008461.
+↻ llm-5 (llm) 10.55s
+✓ llm-5 (llm) 12.98s completionReason: success
+WAITING [worker_lease] Run "01M36DC8S3ARXF65D6ENCDH4HP" step "llm-6" (llm) is running under a worker lease until 1790143010905.
+↻ llm-6 (llm) 12.98s
+✓ llm-6 (llm) 15.31s completionReason: success
+WAITING [worker_lease] Run "01M36DCB2DMASR0GX519PGSW1P" step "llm-7" (llm) is running under a worker lease until 1790143013253.
+↻ llm-7 (llm) 15.31s
+✓ llm-7 (llm) 17.31s completionReason: success
+WAITING [worker_lease] Run "01M36DCD1DZGDCRES8X3G0JTWB" step "llm-8" (llm) is running under a worker lease until 1790143015267.
+↻ llm-8 (llm) 17.30s
+✓ llm-8 (llm) 19.45s completionReason: success
+WAITING [worker_lease] Run "01M36DCF55HRFS7V80RYS5TY2X" step "llm-9" (llm) is running under a worker lease until 1790143017435.
+↻ llm-9 (llm) 19.46s
+✓ llm-9 (llm) 21.72s completionReason: success
+WAITING [worker_lease] Run "01M36DCHCC4F42WYPBZPA7E8YY" step "llm-10" (llm) is running under a worker lease until 1790143019714.
+↻ llm-10 (llm) 21.72s
+✓ llm-10 (llm) 24.56s completionReason: success
+○ run-11 (deterministic) 0.00s
+✓ run-11 (deterministic) 0.02s completionReason: success
+RUN 01M36DBVZG7SWRRDC9VVXW7GV9 completed (12 steps) completionReason: success
diff --git a/evidence/parallel-llm-fix/repro-final-capacity4.txt b/evidence/parallel-llm-fix/repro-final-capacity4.txt
new file mode 100644
index 000000000..7857ecf50
--- /dev/null
+++ b/evidence/parallel-llm-fix/repro-final-capacity4.txt
@@ -0,0 +1,41 @@
+○ run-1 (deterministic) 0.00s
+✓ run-1 (deterministic) 0.03s completionReason: success
+○ llm-2 (llm) 0.00s
+○ llm-3 (llm) 0.00s
+○ llm-4 (llm) 0.00s
+○ llm-5 (llm) 0.00s
+○ llm-6 (llm) 0.00s
+○ llm-7 (llm) 0.00s
+○ llm-8 (llm) 0.00s
+○ llm-9 (llm) 0.00s
+○ llm-10 (llm) 0.00s
+WAITING [worker_lease] Run "01M36DBY9XFPVX5W815P9G0WSE" step "llm-2" (llm) is running under a worker lease until 1790143000261.
+↻ llm-2 (llm) 2.73s
+WAITING [worker_lease] Run "01M36DBYDBR8D7G9M1M73T7R0A" step "llm-3" (llm) is running under a worker lease until 1790143000349.
+↻ llm-3 (llm) 2.70s
+WAITING [worker_lease] Run "01M36DBYFKDZ5KAWKE539EHNXD" step "llm-4" (llm) is running under a worker lease until 1790143000440.
+↻ llm-4 (llm) 2.68s
+WAITING [worker_lease] Run "01M36DBYKEEQ3SBSQGQ8WP7JQK" step "llm-5" (llm) is running under a worker lease until 1790143000562.
+↻ llm-5 (llm) 2.66s
+✓ llm-3 (llm) 4.95s completionReason: success
+WAITING [worker_lease] Run "01M36DC0XBC820MN9ZCDVAT5MK" step "llm-6" (llm) is running under a worker lease until 1790143002850.
+↻ llm-6 (llm) 4.92s
+✓ llm-2 (llm) 5.13s completionReason: success
+WAITING [worker_lease] Run "01M36DC122A2T5FJG8ZA502F1Y" step "llm-7" (llm) is running under a worker lease until 1790143003003.
+↻ llm-7 (llm) 5.04s
+✓ llm-5 (llm) 5.59s completionReason: success
+WAITING [worker_lease] Run "01M36DC1JH4KG5NEZTJJMSNV74" step "llm-8" (llm) is running under a worker lease until 1790143003537.
+↻ llm-8 (llm) 5.56s
+✓ llm-4 (llm) 5.72s completionReason: success
+WAITING [worker_lease] Run "01M36DC1NY00KRCDVGTXP44KKQ" step "llm-9" (llm) is running under a worker lease until 1790143003638.
+↻ llm-9 (llm) 5.65s
+✓ llm-7 (llm) 7.46s completionReason: success
+WAITING [worker_lease] Run "01M36DC3E3QA97KQWBFKWEYKZQ" step "llm-10" (llm) is running under a worker lease until 1790143005434.
+↻ llm-10 (llm) 7.43s
+✓ llm-6 (llm) 7.59s completionReason: success
+✓ llm-8 (llm) 7.89s completionReason: success
+✓ llm-9 (llm) 8.24s completionReason: success
+✓ llm-10 (llm) 9.69s completionReason: success
+○ run-11 (deterministic) 0.00s
+✓ run-11 (deterministic) 0.02s completionReason: success
+RUN 01M36DBVZYQGPS3GKMPTV3X0Y2 completed (12 steps) completionReason: success
diff --git a/evidence/parallel-llm-fix/repro-final-journals.txt b/evidence/parallel-llm-fix/repro-final-journals.txt
new file mode 100644
index 000000000..a58e62555
--- /dev/null
+++ b/evidence/parallel-llm-fix/repro-final-journals.txt
@@ -0,0 +1,2 @@
+{"directory": "/tmp/flows-verified-capacity1", "journals": 13, "lease_expired": []}
+{"directory": "/tmp/flows-verified-capacity4", "journals": 13, "lease_expired": []}
diff --git a/evidence/parallel-llm-fix/repro-journals.txt b/evidence/parallel-llm-fix/repro-journals.txt
new file mode 100644
index 000000000..9edbe8ad0
--- /dev/null
+++ b/evidence/parallel-llm-fix/repro-journals.txt
@@ -0,0 +1,2 @@
+{"directory": "/tmp/flows-repro-capacity1", "journals": 13, "lease_expired": []}
+{"directory": "/tmp/flows-repro-capacity4", "journals": 13, "lease_expired": []}
diff --git a/evidence/parallel-llm-fix/scan-journals.py b/evidence/parallel-llm-fix/scan-journals.py
new file mode 100644
index 000000000..7ac321e39
--- /dev/null
+++ b/evidence/parallel-llm-fix/scan-journals.py
@@ -0,0 +1,15 @@
+import glob
+import json
+import sqlite3
+import sys
+
+for directory in sys.argv[1:]:
+    paths = glob.glob(directory + '/runs/*.sqlite3')
+    expired = []
+    for path in paths:
+        with sqlite3.connect(path) as db:
+            for step, payload in db.execute("select step_id, payload from entries where entry_type='step.completed'"):
+                if json.loads(payload).get('completionReason') == 'lease_expired':
+                    expired.append({'journal': path, 'step': step})
+    print(json.dumps({'directory': directory, 'journals': len(paths), 'lease_expired': expired}))
+    assert paths and not expired
diff --git a/evidence/parallel-llm-fix/sdk-suite.txt b/evidence/parallel-llm-fix/sdk-suite.txt
new file mode 100644
index 000000000..347f0f4a4
--- /dev/null
+++ b/evidence/parallel-llm-fix/sdk-suite.txt
@@ -0,0 +1,1691 @@
+
+> @relayflows/sdk@2.0.29 test
+> sh scripts/test.sh
+
+
+> @relayflows/sdk@2.0.29 test:prep
+> ( cd ../../kernel && sh ../ops/cargo.sh build ) && ( [ ! -d ../../testdata/preflight ] || find ../../testdata/preflight -name '*-cli' -type f -exec chmod +x {} + )
+
+    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.07s
+
+> @relayflows/sdk@2.0.29 typecheck
+> tsc --noEmit && tsc -p tsconfig.type-tests.json
+
+
+> @relayflows/sdk@2.0.29 build
+> tsc && node scripts/make-cli-executable.mjs
+
+
+> @relayflows/sdk@2.0.29 typecheck:tests
+> tsc -p tsconfig.tests.json
+
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+stdout | tests/live-kernel.test.ts
+LIVE_KERNEL relayflowd=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd
+LIVE_KERNEL flows=/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/cli.js
+
+ ✓ tests/preflight.test.ts (67 tests) 108ms
+ ✓ tests/cloud-read.test.ts (41 tests) 42ms
+ ❯ tests/hosted-extension-isolation.test.ts (22 tests | 15 failed) 308ms
+   × hosted extension capability isolation > executes the exact capability-only handler for a queued receipt 13ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > executes the exact capability-only handler for a duplicate receipt 4ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > launches through the captured process primitive after builtin export synchronization 12ms
+     → promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+   × hosted extension capability isolation > ignores inherited launcher overrides and decodes manifests with the captured Buffer intrinsic 4ms
+     → promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+   × hosted extension capability isolation > streams verified bytes when the live store is replaced and no writable staging path exists 202ms
+     → promise rejected "Error: Hosted extension sandbox exited wi… { code: '…' }" instead of resolving
+   × hosted extension capability isolation > mounts pinned private Surface bytes when the live package changes before launch 1ms
+     → ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-48daMa/dist/flow.js'
+   × hosted extension capability isolation > refuses Surface runtime bytes that differ from the reviewed pin before launch 1ms
+     → ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-nbb4yF/dist/flow.js'
+   × hosted extension capability isolation > refuses oversized Surface files through the bounded descriptor reader 1ms
+     → ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-0JhPEY/dist/flow.js'
+   × hosted extension capability isolation > shields verified Surface files before async settlement 1ms
+     → ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-SpYDeT/dist/flow.js'
+   × hosted extension capability isolation > preserves a typed host refusal while disclosing only a fixed marker to the child 4ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to be Error: private Cloud policy detail { code: '…' } // Object.is equality
+   × hosted extension capability isolation > denies ambient credentials, host files, writes, network, subprocesses, and undeclared context verbs 7ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > enforces OS address-space and data bounds on native Buffer allocation 4ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_unsupported', …(1) }
+   × hosted extension capability isolation > blocks extra handler fields and authority-bearing receipt fields at the parent port 2ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+   × hosted extension capability isolation > constructs adapter authority with the captured freeze intrinsic 3ms
+     → bubblewrap is unavailable
+   × hosted extension capability isolation > writes the Surface manifest and protocol without inherited toJSON behavior 3ms
+     → promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+ ✓ tests/cli.test.ts (70 tests) 1836ms
+   ✓ flows check CLI > binds a checked relative wrapper to the flow directory for worker execution 422ms
+   ✓ flows check CLI > resolves a bare PATH-resolved claude with no declared model, in an isolated PATH 415ms
+ ✓ tests/plugin-extension.test.ts (90 tests) 437ms
+ ✓ tests/cloud-transcript-codex.test.ts (39 tests) 20ms
+ ✓ tests/cloud-sync.test.ts (40 tests) 772ms
+ ✓ tests/observer-link.test.ts (39 tests) 144ms
+(node:24071) ExperimentalWarning: SQLite is an experimental feature and might change at any time
+(Use `node --trace-warnings ...` to show where the warning was created)
+ ✓ tests/authored-flow.test.ts (34 tests) 788ms
+ ✓ tests/agent-transcript.test.ts (29 tests) 283ms
+ ✓ tests/cli-status.test.ts (27 tests) 1011ms
+   ✓ flows status > resolves the run with no arguments from inside a worker-spawned agent 769ms
+ ✓ tests/cloud-run.test.ts (58 tests) 679ms
+ ✓ tests/babysitter-native-extension.test.ts (41 tests | 1 skipped) 1479ms
+ ✓ tests/relay-cli-surface.test.ts (75 tests) 39ms
+ ✓ tests/worker-cli.test.ts (18 tests) 24985ms
+   ✓ registered CLI model defaults > passes the same priced Claude default to the real provider invocation 396ms
+   ✓ step discovery environment > names the run, step, attempt and an absolute data dir for a direct agent spawn 419ms
+   ✓ step discovery environment > exports none of the four without a data dir, even when the worker inherited them 433ms
+   ✓ wrapper discovery environment > sets the four names from the dispatch and still refuses ambient values and other secrets 414ms
+   ✓ wrapper discovery environment > exports none of the four to a wrapper without a data dir, even when the worker inherited them 342ms
+   ✓ custom wrapper execution identity > passes an explicit safe environment at identification and execution 414ms
+   ✓ custom wrapper execution identity > refuses a wrapper symlink retarget before delivering private values 368ms
+   ✓ custom wrapper execution identity > bounds wrapper execution after acknowledgement 384ms
+   ✓ custom wrapper execution identity > bounds captured wrapper output 307ms
+   ✓ custom wrapper execution identity > refuses a duplicate execute protocol frame 337ms
+   ✓ custom wrapper execution bounds are reader-owned > resolves when a conforming wrapper leaks a stdio pipe to a background helper 1906ms
+   ✓ custom wrapper execution bounds are reader-owned > resolves when the leaked helper inherits stderr only 1907ms
+   ✓ custom wrapper execution bounds are reader-owned > resolves when a wrapper leaks a stdio pipe and exits before identifying 3597ms
+   ✓ custom wrapper execution bounds are reader-owned > journals a completionReason at the default bound when a wrapper leaks a stdio pipe 11569ms
+   ✓ custom wrapper execution bounds are reader-owned > accepts an execute token and an over-8KiB payload flushed in one write 349ms
+   ✓ custom wrapper execution bounds are reader-owned > accepts the same over-8KiB payload whether or not it coalesces with the execute token 1139ms
+   ✓ custom wrapper execution bounds are reader-owned > still bounds an un-terminated handshake buffer and names the bound 370ms
+   ✓ delivers the journaled memory pack to the real wrapper and excludes its charge from completion usage 332ms
+ ✓ tests/step-failure-diagnostic.test.ts (25 tests) 58ms
+ ✓ tests/daemon-lifecycle.test.ts (42 tests) 45ms
+ ✓ tests/stop-process-group.test.ts (9 tests) 12759ms
+   ✓ every stop reaches the process group, not just the direct child > exits the run after an execution-timeout stop 775ms
+   ✓ every stop reaches the process group, not just the direct child > exits the run after a protocol terminate stop 395ms
+   ✓ every stop reaches the process group, not just the direct child > kills a SIGTERM-deaf grandchild after a protocol terminate stop 1666ms
+   ✓ every stop reaches the process group, not just the direct child > kills a SIGTERM-deaf grandchild after an execution-timeout stop 2033ms
+   ✓ every stop reaches the process group, not just the direct child > holds the loop open long enough for the escalation to run 1092ms
+   ✓ a wrapper that exits with no execution deadline still drains > reports the wrapper result and reaps a grandchild holding its pipes 656ms
+   ✓ a wrapper that exits with no execution deadline still drains > reaps a SIGTERM-deaf grandchild holding its pipes 1621ms
+   ✓ a wrapper that exits with no execution deadline still drains > settles on its own deadline when an escaped holder withholds close 4268ms
+ ✓ tests/run-state.test.ts (21 tests) 12ms
+ ✓ tests/cloud-deploy.test.ts (40 tests) 932ms
+ ❯ tests/hosted-extension-protocol.test.ts (24 tests | 8 failed) 40093ms
+   × hosted extension hostile protocol > uses captured JSON intrinsics for the complete parent boundary 7ms
+     → bubblewrap is unavailable
+   × hosted extension hostile protocol > rejects an import-time different PR frame with zero adapter calls 8ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+   × hosted extension hostile protocol > rejects an import-time different delivery frame with zero adapter calls 3ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+   × hosted extension hostile protocol > rejects an import-time different event frame with zero adapter calls 2ms
+     → expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+   × hosted extension hostile protocol > rejects two forged calls after the authoritative first outcome settles 10003ms
+     → hostile child did not invoke the adapter
+   × hosted extension hostile protocol > waits for a pending adapter to reject after a forged child error 10005ms
+     → hostile child did not invoke the adapter
+   × hosted extension hostile protocol > waits for a pending adapter to resolve after a forged child error 10005ms
+     → hostile child did not invoke the adapter
+   × hosted extension hostile protocol > returns a typed adapter rejection even when the hostile child hangs 10006ms
+     → hostile child did not invoke the adapter
+ ✓ tests/hosted-base-snapshot.test.ts (18 tests) 1935ms
+   ✓ hosted base private snapshot > stops streaming project entries at the shared count limit 1854ms
+stdout | tests/live-kernel.test.ts > surface resume after a real daemon kill > resumes a three-step run with each successful completion exactly once
+LIVE_KERNEL kill -9 pid=25410 run=01M36CVPD1JG5S41C76P3C5D53 while step=two state=Running
+
+ ❯ tests/live-kernel.test.ts (31 tests | 9 failed) 52134ms
+   ✓ built flows CLI against live relayflowd > twenty-six-step reuses 25 durable completions after editing the failed final step 2204ms
+   ✓ built flows CLI against live relayflowd > runs rung (a), parks rung (b), and keeps JSON report-shaped 2659ms
+   ✓ built flows CLI against live relayflowd > allows a deterministic run to exceed the bounded request timeout 32477ms
+   ✓ built flows CLI against live relayflowd > follows a live worker dispatch through flows run 573ms
+   ✓ built flows CLI against live relayflowd > runs an agent CLI end to end through the SDK worker 323ms
+   ✓ built flows CLI against live relayflowd > f.agent lowers to a real agent step and dispatches through a live worker 473ms
+   ✓ built flows CLI against live relayflowd > can always get a parked run to a late-attaching worker 5593ms
+   ✓ built flows CLI against live relayflowd > reports a real manual-recovery NeedsHuman state as parked 429ms
+   × built flows CLI against live relayflowd > runs hn-monitor analyze-story end-to-end via a stub agent CLI (gate 2 clause 2 demo) 425ms
+     → expected { …(12) } to match object { output: { …(3) }, …(1) }
+(22 matching properties omitted from actual)
+   × built flows CLI against live relayflowd > hn-monitor analyze-story FAILS verification when the CLI omits required schema fields 385ms
+     → expected { …(12) } to match object { …(3) }
+(21 matching properties omitted from actual)
+   × built flows CLI against live relayflowd > agent step preserves the CliResult wrapper as output when the CLI emits non-JSON text 402ms
+     → expected null not to be null
+   × built flows CLI against live relayflowd > AgentWorker exposes wake_context to the CLI via RELAYFLOW_WAKE_CONTEXT env var (real analyzer prerequisite) 346ms
+     → Cannot read properties of null (reading 'story_title')
+   × built flows CLI against live relayflowd > AgentWorker leaves RELAYFLOW_WAKE_CONTEXT UNSET when the run has no wake_context (undefined-vs-null pin) 382ms
+     → Cannot read properties of null (reading 'env_present')
+   ✓ built flows CLI against live relayflowd > AgentWorker passes a declared model to an identified wrapper as RELAYFLOW_MODEL 494ms
+   ✓ built flows CLI against live relayflowd > AgentWorker refuses a nonconforming journal-submitted wrapper before exposing RELAYFLOW_MODEL 435ms
+   ✓ built flows CLI against live relayflowd > AgentWorker executes the raw claude adapter with its real model flag 388ms
+   ✓ built flows CLI against live relayflowd > AgentWorker executes the raw codex adapter with its real model flag 365ms
+   × built flows CLI against live relayflowd > AgentWorker leaves RELAYFLOW_MODEL UNSET when the step declares no model 380ms
+     → Cannot read properties of null (reading 'story_title')
+   × built flows CLI against live relayflowd > hn-monitor analyze-story reaches done through the real Claude analyzer CLI 27ms
+     → LIVE_ANALYZER_UNAVAILABLE: "/home/daytona/.relayflow-v2-supervisor/durable/repository/testdata/preflight/analyze-story-claude-cli" does not identify as relayflows-agent-cli-v1 — failing because gate-2 acceptance requires the real analyzer to execute. Set RELAYFLOWS_ALLOW_ANALYZER_SKIP=1 only if this run is not gate evidence.
+   ✓ built flows CLI against live relayflowd > preflights before journaling and names an unreachable socket 882ms
+   × built flows CLI against live relayflowd > starts exactly one daemon when two runs race for one empty data dir 527ms
+     → WARNING [unprovable_effects] Step "greet" command "echo" resolves, but its effects cannot be proven before execution.
+WARNING [unprovable_effects] Step "shout" command "echo" resolves, but its effects cannot be proven before execution.
+WARNING [editor_schema_missing] For editor validation, add this first line: # yaml-language-server: $schema=https://schema.relayflows.dev/v0.1/flows.schema.json
+REFUSED [relayflowd_not_found] No relayflowd binary could be found. Install the runtime package for this host (@relayflows/runtime-linux-x64), or set RELAYFLOWD_BIN to a relayflowd executable. Tried: /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/relayflowd.
+: expected 2 to be +0 // Object.is equality
+   ✓ surface resume after a real daemon kill > resumes a three-step run with each successful completion exactly once 881ms
+   × a relayflow can be scheduled: tick source against live relayflowd > a tick spawns a real run whose step reports the SCHEDULED instant 387ms
+     → expected null to deeply equal { schedule_id: 'heartbeat-1m', …(3) }
+ ✓ tests/journal-client.test.ts (17 tests) 82ms
+ ✓ tests/authored-root.test.ts (13 tests) 162ms
+ ✓ tests/cloud-connect.test.ts (24 tests) 2926ms
+   ✓ hosted verbs connect before they submit > flows run --cloud submits once the prompt connected the integration 2122ms
+ ✓ tests/flow-extension-compose.test.ts (22 tests) 3788ms
+   ✓ composing flow extensions onto a base flow > composes two extensions in declaration order, and the order is the lockfile order 412ms
+   ✓ composing flow extensions onto a base flow > flows check reports the composition and keeps the composed triggers deliverable 727ms
+ ❯ tests/authored-node-runtime.test.ts (14 tests | 14 skipped) 11ms
+ ❯ tests/mcp.test.ts (30 tests | 4 skipped) 9603ms
+   ✓ MCP preflight and transports > flows check refuses an undeclared server with exit 2 and no daemon 579ms
+   ✓ MCP preflight and transports > flows check reports a refusing server and leaves no PID 778ms
+   ✓ MCP preflight and transports > kills a SIGTERM-resistant silent child after a parent-owned handshake deadline 1313ms
+   ✓ MCP preflight and transports > reaps a SIGTERM-resistant descendant with inherit stdio before cleanup finishes 1110ms
+   ✓ MCP preflight and transports > reaps a SIGTERM-resistant descendant with ignore stdio before cleanup finishes 2061ms
+   ✓ MCP preflight and transports > reports malformed connection configuration as config_invalid 598ms
+ ✓ tests/bundle.test.ts (26 tests) 9311ms
+   ✓ immutable bundles > returns exit 2 naming a byte-flipped payload and refuses to reuse corruption 525ms
+   ✓ immutable bundles > verifies with --verify in any position and answers --json with one object 737ms
+   ✓ immutable bundles > refuses --out with --verify rather than ignoring the destination 367ms
+   ✓ immutable bundles > builds and verifies the canonical YAML fixture through the compiled CLI 1183ms
+   ✓ immutable bundles > emits the ephemeral warning on CLI stderr and uses the default output directory 746ms
+   ✓ immutable bundles > refuses build-provable CLI resolution errors without environment probes 386ms
+   ✓ immutable bundles > builds a standalone TS fixture twice with identical executable hashes 2442ms
+   ✓ immutable bundles > refuses to label installed dependency drift with lockfile pins 386ms
+   ✓ immutable bundles > refuses invalid CLI arguments %j 394ms
+   ✓ immutable bundles > refuses invalid CLI arguments "--out" 430ms
+   ✓ immutable bundles > refuses invalid CLI arguments "--verify" 578ms
+   ✓ immutable bundles > refuses invalid CLI arguments "--verify" 553ms
+   ✓ immutable bundles > refuses invalid CLI arguments "--out" 451ms
+ ✓ tests/validate.test.ts (68 tests) 25ms
+ ✓ tests/verb-field-lint.test.ts (96 tests) 320ms
+ ✓ tests/close-pr-flow.test.ts (28 tests) 10414ms
+   ✓ close-pr journaled repair loop > reads an existing Bugbot finding, repairs in the same worktree, pushes and re-verifies before merging 2239ms
+   ✓ close-pr journaled repair loop > opens a PR and feeds failed CI logs into the repair agent 1798ms
+   ✓ close-pr journaled repair loop > parks after exactly three nonconverging repairs, with accumulated blockers 1949ms
+   ✓ close-pr journaled repair loop > can converge on the third repair 2080ms
+   ✓ close-pr journaled repair loop > executes the deterministic commit and force-push steps against a local Git remote, including a no-op repair 2098ms
+ ✓ tests/tick-source.test.ts (33 tests) 25ms
+ ✓ tests/authored-step-graph.test.ts (25 tests) 815ms
+ ✓ tests/agent-relay-transport.test.ts (16 tests) 2236ms
+   ✓ Relay completion at the journal boundary > does not complete at readiness and journals exact output, receipt, and priced accounting 1006ms
+   ✓ Relay completion at the journal boundary > aborts polling on rejected renewal and never writes a stale completion 1002ms
+ ✓ tests/pr-review-post.test.ts (21 tests) 2129ms
+ ✓ tests/step-failure-excerpt.test.ts (42 tests) 193ms
+ ✓ tests/authored-flow-lifecycle-executor.test.ts (27 tests) 570ms
+ ✓ tests/authored-flow-slack.test.ts (7 tests) 1724ms
+   ✓ authored Slack helper effects > replays after SIGKILL before confirm with the same token and one successful completion 698ms
+   ✓ authored Slack helper effects > replays after SIGKILL before complete with the same token and one successful completion 481ms
+ ✓ tests/tick-runner.test.ts (22 tests) 2288ms
+   ✓ CLI argument parsing refuses coercion rather than accepting it > refuses --interval-ms fractional as an invocation error 389ms
+   ✓ CLI argument parsing refuses coercion rather than accepting it > refuses --interval-ms exponent notation as an invocation error 370ms
+   ✓ CLI argument parsing refuses coercion rather than accepting it > refuses --interval-ms hex as an invocation error 368ms
+   ✓ CLI argument parsing refuses coercion rather than accepting it > refuses --interval-ms trailing text as an invocation error 385ms
+   ✓ CLI argument parsing refuses coercion rather than accepting it > refuses --interval-ms empty as an invocation error 372ms
+   ✓ CLI argument parsing refuses coercion rather than accepting it > accepts an exact integer and proceeds past parsing 378ms
+ ✓ tests/authored-step-index.test.ts (17 tests) 20ms
+(node:29919) ExperimentalWarning: SQLite is an experimental feature and might change at any time
+(Use `node --trace-warnings ...` to show where the warning was created)
+ ✓ tests/flow-executor-chain.test.ts (14 tests) 10301ms
+   ✓ flow executor LLM and output-binding chain > runs f.llm -> f.agent -> f.run with schema-verified journal output and the exact allowed model 836ms
+   ✓ flow executor LLM and output-binding chain > runs a dollar-budgeted authored Claude agent with the same default used by preflight 474ms
+   ✓ flow executor LLM and output-binding chain > fails invalid LLM output before the next step: {"message":7} 360ms
+   ✓ flow executor LLM and output-binding chain > preserves JSON values without promoting them to process wrappers: null 510ms
+   ✓ flow executor LLM and output-binding chain > preserves JSON values without promoting them to process wrappers: "hello" 324ms
+   ✓ flow executor LLM and output-binding chain > runs the exact authored flagship f.llm -> f.agent -> f.run path through the durable CLI root 1492ms
+   ✓ flow executor LLM and output-binding chain > resumes an interrupted durable authored root without replaying completed flagship effects 3251ms
+   ✓ flow executor LLM and output-binding chain > passes a declarative verified value through an agent into a deterministic artifact 757ms
+   ✓ flow executor LLM and output-binding chain > flows run consumes YAML bindings and resume reuses the original journal output 1084ms
+ ✓ tests/cli-replay.test.ts (37 tests) 1302ms
+   ✓ flows replay > --json is byte-identical across two CLI invocations (diff) 792ms
+ ✓ tests/authored-node-result.test.ts (39 tests) 16ms
+ ✓ tests/gate-contract.test.ts (20 tests) 152ms
+ ✓ tests/authored-human.test.ts (13 tests) 84ms
+ ✓ tests/wrapper-execution-duration.test.ts (7 tests) 10889ms
+   ✓ keeps the handshake deadline independent of the removed execution deadline 10052ms
+   ✓ still lets a lease abort stop an unlimited wrapper before it produces output 510ms
+ ✓ tests/direct-input.test.ts (6 tests) 5651ms
+   ✓ direct .flow.ts input through the built CLI and live runtime > returns exit 3 for an authored human handoff and persists its outcome 569ms
+   ✓ direct .flow.ts input through the built CLI and live runtime > returns exit 1 for an authored step_failed verdict and persists its outcome 597ms
+   ✓ direct .flow.ts input through the built CLI and live runtime > executes inline and file JSON input through relayflowd 2033ms
+   ✓ direct .flow.ts input through the built CLI and live runtime > refuses missing and malformed input before contacting relayflowd 1533ms
+   ✓ direct .flow.ts input through the built CLI and live runtime > does not run the authored body before daemon availability 490ms
+   ✓ direct .flow.ts input through the built CLI and live runtime > refuses oversized file input before contacting relayflowd 429ms
+ ✓ tests/authored-run-failure-evidence.test.ts (9 tests) 1031ms
+   ✓ the child index after the process that wrote it is gone > still names every child, with its own run id, after a daemon restart 572ms
+ ✓ tests/cli-hn-monitor.test.ts (16 tests) 94ms
+ ✓ tests/cloud-schedule.test.ts (17 tests) 5614ms
+   ✓ schedule lowering > marks a non-grid cron as Cloud-only rather than approximating it, with a silence budget from its own cadence 3307ms
+   ✓ flows check prints declared schedules > shows the lowering for a fixed interval and the Cloud-only note for a real cron 1810ms
+ ✓ tests/daemon-lifecycle-live.test.ts (9 tests) 6432ms
+   ✓ flows run against a data dir with no daemon (§6 test 7) > cold start spawns exactly one daemon, the run succeeds, and the daemon outlives the CLI 473ms
+   ✓ flows run against a data dir with no daemon (§6 test 7) > polls, bounded, for a daemon that holds the lock before it binds 1365ms
+   ✓ flows run against a data dir with no daemon (§6 test 7) > attaches to a serving daemon that has not published a connection file 436ms
+   ✓ flows run against a data dir with no daemon (§6 test 7) > a second run attaches to the daemon the first one started, spawning nothing 840ms
+   ✓ flows run against a data dir with no daemon (§6 test 7) > detects a stale connection file left by a hard kill and starts a fresh daemon 935ms
+   ✓ concurrent invocations against one empty data dir (§6 test 15) > ends with exactly one daemon owning the socket, and both runs succeed 1080ms
+   ✓ refusals from a spawn that cannot produce a daemon > names relayflowd_not_found rather than falling through to PATH 393ms
+   ✓ refusals from a spawn that cannot produce a daemon > names daemon_start_failed and quotes the daemon log when startup dies 455ms
+   ✓ refusals from a spawn that cannot produce a daemon > refuses a daemon speaking another protocol version instead of binding over it 454ms
+(node:31388) Warning: Transcript tail for run-9/analyze attempt 1 (stdout) could not be written; the step continues without it: EACCES: permission denied, mkdir '/tmp/transcript-tail-t1B4pL/runs/run-9/steps'
+(Use `node --trace-warnings ...` to show where the warning was created)
+ ✓ tests/transcript-tail.test.ts (11 tests) 753ms
+   ✓ direct agent spawn > tees stdout and stderr into tail files that name the dispatch 322ms
+   ✓ direct agent spawn > completes the step when the tail directory cannot be created 352ms
+ ✓ tests/authored-agent-artifacts.test.ts (4 tests) 311ms
+ ✓ tests/authored-helpers.test.ts (6 tests) 2932ms
+   ✓ runs every available provider through the real kernel and resumes completed effects without a second write 1529ms
+   ✓ replays after SIGKILL before confirm with the same token and one successful completion 500ms
+   ✓ replays after SIGKILL before complete with the same token and one successful completion 499ms
+ ✓ tests/authored-flow-operation.test.ts (24 tests) 546ms
+ ✓ tests/flow-requirements.test.ts (14 tests) 466ms
+ ✓ tests/backlog-picker.test.ts (14 tests) 47ms
+ ✓ tests/plugin-store-bounds.test.ts (11 tests) 75ms
+ ✓ tests/backlog-picker-flow.test.ts (6 tests) 267ms
+ ✓ tests/hosted-extension-protocol-intrinsics.test.ts (6 tests) 13ms
+ ✓ tests/preflight-permissions-unenforced.test.ts (17 tests) 215ms
+ ✓ tests/wrapper-exit-drain.test.ts (8 tests) 2642ms
+   ✓ reports a signalled wrapper death while a pipe is held, with its output intact 383ms
+   ✓ refuses a duplicate execute frame found at finalization, despite a clean exit 310ms
+   ✓ lets a lease abort outrank a successful exit still being drained 527ms
+ ❯ tests/stuck-run-triage.test.ts (22 tests | 22 failed) 70ms
+   × stuck-run-triage input validation > refuses an 8-character run-id prefix: Cloud has no prefix lookup 5ms
+     → expected [Function] to throw error matching /not full Cloud run ids: c649fe14/ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage input validation > refuses the whole batch when any id is invalid, rather than dropping it 1ms
+     → expected [Function] to throw error matching /not full Cloud run ids: nope!/ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage input validation > refuses an empty batch 1ms
+     → expected [Function] to throw error matching /needs runIds/ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage input validation > refuses a batch too large for the edge step lease 1ms
+     → expected [Function] to throw error matching /exceeds the 8 that fit/ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage input validation > accepts eight ids — the incident batch is inside the bound 3ms
+     → promise rejected "TypeError: expected an @relayflows/surfac…" instead of resolving
+   × stuck-run-triage apiUrl > refuses to send the Cloud bearer token to an unapproved origin 0ms
+     → expected [Function] to throw error matching /refusing to send the Cloud bearer to…/\ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage apiUrl > refuses a non-URL apiUrl 0ms
+     → expected [Function] to throw error matching /is not a URL/ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage apiUrl > allows an approved origin and uses it in the curl 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage apiUrl > defaults to production Cloud 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage apiUrl > never publishes a run record the fetch did not produce 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage edge collection > names the Worker on every wrangler invocation 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage edge collection > accepts caller-supplied Workers and rejects option-shaped ones 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage edge collection > falls back when GNU timeout is absent, as it is on macOS 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage edge collection > runs the tails concurrently so wall time does not scale with the batch 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage edge collection > records wrangler's own exit status rather than head's 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage shell text > parses under both sh and bash 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage shell text > collects tails with no GNU timeout on PATH, as on a stock macOS 53ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage agents > declares read-only permissions on every agent 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage agents > tells the forensics agents their evidence is untrusted 0ms
+     → expected an @relayflows/surface flow handle
+   × stuck-run-triage fan-out > refuses a duplicate run id: two tails would share one evidence file 1ms
+     → expected [Function] to throw error matching /duplicate runIds: c649fe14-0c2e-4e51-…/ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage fan-out > refuses a duplicate Worker name for the same reason 0ms
+     → expected [Function] to throw error matching /duplicate workers: w-one/ but got 'expected an @relayflows/surface flow …'
+   × stuck-run-triage fan-out > bounds ids x workers, not just ids 0ms
+     → expected [Function] to throw error matching /24 concurrent tails, over the 16/ but got 'expected an @relayflows/surface flow …'
+ ✓ tests/worker-transcript.test.ts (5 tests) 180ms
+ ✓ tests/hosted-extension-routing.test.ts (7 tests) 7ms
+ ✓ tests/artifact-gates.test.ts (7 tests) 175ms
+ ✓ tests/webhook.test.ts (9 tests) 807ms
+   ✓ webhook ingress > checks TS declarations against flows.json without invoking handlers 709ms
+ ✓ tests/agent-artifacts-live.test.ts (5 tests) 43638ms
+   ✓ agent artifacts and gates through the built CLI, a real daemon and the local agent > journals the files the agent wrote, including under a dot-directory, and every artifact gate passes on that journal 1107ms
+   ✓ agent artifacts and gates through the built CLI, a real daemon and the local agent > fails the run when the artifact_exists gate names a file the agent did not write 13048ms
+   ✓ agent artifacts and gates through the built CLI, a real daemon and the local agent > fails the run with the author reason when a predicate gate returns false, journaling the verdict 14268ms
+   ✓ review follow-ups > applies a predicate gate on a helper step too, and journals its verdict 14180ms
+   ✓ review follow-ups > records predicate verdicts on the root run so a resume reuses them instead of re-running the closure 1033ms
+ ✓ tests/agent-transcript-live.test.ts (4 tests) 45166ms
+   ✓ the transcript digest through the built CLI, a real daemon and the local agent > preserves structured agent failure details and its completed root index 14668ms
+   ✓ the transcript digest through the built CLI, a real daemon and the local agent > preserves structured llm failure details and its completed root index 14652ms
+   ✓ the transcript digest through the built CLI, a real daemon and the local agent > journals the digest in trajectory_tail on a successful agent step and writes the file it points at 773ms
+   ✓ the transcript digest through the built CLI, a real daemon and the local agent > on a failed agent step, names the failure and the transcript in the terminal diagnostic, redacted 15073ms
+ ✓ tests/human-live.test.ts (3 tests) 6241ms
+   ✓ f.human against a real daemon > parks with the question, refuses wrong answers, records one, and resumes to success 3714ms
+   ✓ f.human against a real daemon > a "no" is a value the body branches on: declined, exit 0, no effect 1561ms
+   ✓ f.human against a real daemon > refuses to answer a run the daemon does not know 964ms
+ ✓ tests/authored-step-failed.test.ts (10 tests) 37ms
+ ✓ tests/cli-watch.test.ts (10 tests) 15881ms
+   ✓ flows check --watch > rechecks syntax errors, clears once, and returns the last refusal on Ctrl-C 1293ms
+   ✓ flows check --watch > streams JSON lines without ANSI, recovers after atomic saves, and exits zero after repair 1948ms
+   ✓ flows check --watch > coalesces 20 concurrent saves into at most two rechecks 1857ms
+   ✓ flows check --watch > watches transitive relative use imports, cycles, and nearest config changes 2361ms
+   ✓ flows check --watch > refreshes the import graph and notices missing imports being created 2334ms
+   ✓ flows check --watch > reloads authored TypeScript instead of reusing the first imported definition 1553ms
+   ✓ flows check --watch > detects a nearer config appearing and falls back after it is deleted 1910ms
+   ✓ flows check --watch > keeps watching after the target is deleted and recreated 1851ms
+   ✓ flows check --watch > queues changes during a slow check without overlapping checks 773ms
+ ✓ tests/budget-preflight.test.ts (25 tests) 16ms
+ ✓ tests/authored-parallel-agents.test.ts (8 tests) 11090ms
+   ✓ authored steps under local workers with capacity > runs more concurrent f.llm calls than the worker holds side by side, never more than its capacity 1127ms
+   ✓ authored steps under local workers with capacity > completes more concurrent f.agent calls than the worker holds: the overflow waits for a slot instead of parking 2399ms
+   ✓ authored steps under local workers with capacity > runs agents in distinct working directories side by side (the kernel carries cwd) 632ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a symlink alias of the same directory 1109ms
+   ✓ authored steps under local workers with capacity > serializes agents whose cwd is a directory nested inside the other 1050ms
+   ✓ authored steps under local workers with capacity > never starts queued agents once the body has failed 1925ms
+   ✓ authored steps under local workers with capacity > never starts a queued agent when the agent holding the only slot fails 1916ms
+   ✓ authored steps under local workers with capacity > parks the overflow when the body is not told the capacity (the defect this closes) 930ms
+ ✓ tests/budget-unmetered-live.test.ts (3 tests) 1343ms
+   ✓ unmetered budget spend through the live kernel > runs an unpriced step under a dollar budget without tripping it, journaling unknown dollars 496ms
+   ✓ unmetered budget spend through the live kernel > still counts an unpriced step toward a token budget 568ms
+ ✓ tests/provider-trigger-contract.test.ts (7 tests) 526ms
+   ✓ provider trigger contract > fails `flows check` before deployment and passes once the event is real 325ms
+ ✓ tests/work-package-consumer.test.ts (13 tests) 107ms
+ ✓ tests/spec-parity.test.ts (31 tests) 361ms
+ ✓ tests/helpers-fanout.test.ts (96 tests) 227ms
+(node:35350) ExperimentalWarning: SQLite is an experimental feature and might change at any time
+(Use `node --trace-warnings ...` to show where the warning was created)
+ ✓ tests/authored-step-graph-live.test.ts (1 test) 694ms
+   ✓ the authored step DAG through the live kernel > carries labels and predecessors on every index record and journal step, ids unchanged 694ms
+ ✓ tests/generate-triggers.test.ts (7 tests) 1076ms
+   ✓ discovers new adapters, preserves exact event names, and prefers adapter-local mappings 346ms
+ ❯ tests/webhook-live.test.ts (6 tests | 6 failed) 62434ms
+   × executes and deduplicates 'app_mention' only for its provider and matching payload 10413ms
+     → webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+   × executes and deduplicates 'reaction_added' only for its provider and matching payload 10409ms
+     → webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+   × executes and deduplicates 'pull_request' only for its provider and matching payload 10411ms
+     → webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+   × flows serve-webhook writes JSON before the daemon starts, then journals and archives exactly once 10400ms
+     → webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+   × replays a dropped file after SIGKILL before spawn 10408ms
+     → webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+   × resumes the same journal after SIGKILL after spawn and before acknowledgement 10392ms
+     → webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+ ✓ tests/webhook-hardening.test.ts (11 tests) 62ms
+ ✓ tests/human-to.test.ts (8 tests) 10ms
+ ✓ tests/pty-sidechannel.test.ts (11 tests) 5715ms
+   ✓ view attach preserves worker completion and marks only drive 782ms
+   ✓ drive attach preserves worker completion and marks only drive 371ms
+   ✓ passthrough attach preserves worker completion and marks only drive 799ms
+   ✓ none attach preserves worker completion and marks only drive 778ms
+   ✓ none subscriber lets an unattended CLI read EOF 412ms
+   ✓ view subscriber lets an unattended CLI read EOF 369ms
+   ✓ passthrough subscriber lets an unattended CLI read EOF 409ms
+   ✓ incomplete subscriber lets an unattended CLI read EOF 412ms
+   ✓ rejects drive after EOF without marking human intervention 730ms
+   ✓ delivers all drive bytes in order across child stdin backpressure 650ms
+ ✓ tests/plugin-loader.test.ts (9 tests) 192ms
+ ✓ tests/worker-cli-result-exit.test.ts (5 tests) 32887ms
+   ✓ a Claude agent step completes on its result, not only on process exit > settles a hung, successful run within the grace and stops its whole tree 31615ms
+   ✓ a Claude agent step completes on its result, not only on process exit > maps an error result on a hung run to a failed exit 31616ms
+   ✓ a Claude agent step completes on its result, not only on process exit > leaves a hang before any result to the existing stops 32010ms
+   ✓ an agent tree does not outlive the process that spawned it > kills the agent group when the run process is terminated by SIGTERM 797ms
+ ✓ tests/worker-lease.test.ts (7 tests) 12ms
+ ✓ tests/yaml-helpers.test.ts (33 tests) 75ms
+ ✓ tests/authored-agent-permissions.test.ts (26 tests) 780ms
+ ✓ tests/babysitter-catalog-export.test.ts (14 tests) 784ms
+   ✓ Babysitter catalog artifact export > CLI refuses an existing output and leaves no file on validation failure 624ms
+ ✓ tests/redact.test.ts (35 tests) 11ms
+ ✓ tests/communication.test.ts (10 tests) 14ms
+ ✓ tests/typed-output.test.ts (14 tests) 212ms
+ ❯ tests/canonical-software-factory.test.ts (3 tests | 3 failed) 19ms
+   × canonical software-factory metadata contract > opens the actual catalog flow with the ticket title and exactly one GitHub closing line 12ms
+     → expected an @relayflows/surface flow handle
+   × canonical software-factory metadata contract > normalizes whitespace and caps the title at 240 Unicode code points 6ms
+     → expected an @relayflows/surface flow handle
+   × canonical software-factory metadata contract > fails closed before push when GitHub identity is missing or the final body duplicates its closing line 0ms
+     → expected an @relayflows/surface flow handle
+ ✓ tests/budget-attribution.test.ts (5 tests) 7ms
+ ✓ tests/deploy.test.ts (11 tests) 5282ms
+   ✓ flows deploy file buckets > publishes the full signed layout byte-for-byte and redeploys as a noop 845ms
+   ✓ flows deploy file buckets > answers --json with one object per outcome 802ms
+   ✓ flows deploy file buckets > reports a refusal as JSON under --json 405ms
+   ✓ flows deploy file buckets > refuses a missing local bundle before creating the bucket 396ms
+   ✓ flows deploy file buckets > refuses an unreachable bucket before copying 417ms
+   ✓ flows deploy file buckets > refuses an unwritable bucket 386ms
+   ✓ flows deploy file buckets > refuses local tampering of spec.canonical.json 385ms
+   ✓ flows deploy file buckets > refuses local tampering of identity.json 395ms
+   ✓ flows deploy file buckets > refuses asset bundles instead of using daemon-relative files 414ms
+   ✓ flows deploy file buckets > never labels a corrupt existing deployment as a noop 812ms
+ ✓ tests/effect-channel.test.ts (5 tests) 349ms
+ ✓ tests/json-schema-bound.test.ts (71 tests) 2452ms
+   ✓ JSON Schema termination bound > walks a deep schema with an explicit stack rather than recursion 2010ms
+ ✓ tests/mcp-lifecycle.test.ts (4 tests) 12ms
+ ✓ tests/model-selection.test.ts (10 tests) 17ms
+ ✓ tests/f-memory.test.ts (7 tests) 828ms
+ ✓ tests/relayflowd-path.test.ts (10 tests) 5ms
+ ✓ tests/agent-artifacts.test.ts (9 tests) 19ms
+ ✓ tests/step-lease.test.ts (36 tests) 66546ms
+   ✓ f.run leases against the live kernel > enforces 10000 ms for 'sleep 5; printf ok' 5093ms
+   ✓ f.run leases against the live kernel > enforces 40000 ms for 'sleep 31; printf ok' 31084ms
+   ✓ f.run leases against the live kernel > enforces 30000 ms for 'sleep 31; printf ok' 30123ms
+ ✓ tests/authored-plugin-effect.test.ts (6 tests) 55ms
+ ✓ tests/yaml-local-agent-live.test.ts (7 tests) 4580ms
+   ✓ YAML --local-agent through the built CLI and real daemon > runs with the checked step CLI and model and journals done 641ms
+   ✓ YAML --local-agent through the built CLI and real daemon > runs with the checked named CLI and model and journals done 677ms
+   ✓ YAML --local-agent through the built CLI and real daemon > runs with the checked flow CLI and model and journals done 915ms
+   ✓ YAML --local-agent through the built CLI and real daemon > runs with the checked project CLI and model and journals done 623ms
+   ✓ YAML --local-agent through the built CLI and real daemon > still parks without --local-agent 534ms
+   ✓ YAML --local-agent through the built CLI and real daemon > reports the agent process failure 622ms
+   ✓ YAML --local-agent through the built CLI and real daemon > preserves declared workspace surfaces that the local worker cannot pin 568ms
+ ✓ tests/worker-slots.test.ts (7 tests) 5ms
+ ✓ tests/local-dev-ux.test.ts (8 tests) 63ms
+ ↓ tests/relay-cli-surface-live.test.ts (3 tests | 3 skipped)
+ ✓ tests/authored-declined.test.ts (13 tests) 48ms
+ ✓ tests/resume-failure.test.ts (2 tests) 5ms
+ ✓ tests/dependency-validation.test.ts (6 tests) 800ms
+   ✓ dependency validation > accepts a valid 10,000-step reverse chain through every direct public boundary 430ms
+ ✓ tests/authored-hooks.test.ts (5 tests) 7ms
+ ✓ tests/local-agent-live.test.ts (5 tests) 66377ms
+   ✓ built CLI local agent against a real daemon > dispatches through the wrapper and keeps --json stdout report-shaped 863ms
+   ✓ built CLI local agent against a real daemon > runs beyond the initial 30-second lease without a second invocation 36934ms
+   ✓ built CLI local agent against a real daemon > renders actual agent completion in text output 1091ms
+   ✓ built CLI local agent against a real daemon > returns a failed run when the agent process fails 12868ms
+   ✓ built CLI local agent against a real daemon > refuses a workspace it cannot pin before invoking the agent 14621ms
+ ✓ tests/input-binding.test.ts (12 tests) 419ms
+ ✓ tests/communication-review.test.ts (5 tests) 354ms
+ ✓ tests/yaml-helper-effect.test.ts (4 tests) 143ms
+ ✓ tests/deterministic-llm.test.ts (5 tests) 96ms
+ ✓ tests/scope-preflight.test.ts (6 tests) 8ms
+ ✓ tests/build-gate.test.ts (3 tests) 1683ms
+   ✓ flows build gates on flows check green (#318) > refuses a flow with an unresolvable named-agent CLI and leaves no artifacts 675ms
+   ✓ flows build gates on flows check green (#318) > --json emits one CheckReport object on stdout on refusal, exits 2, no artifacts 432ms
+   ✓ flows build gates on flows check green (#318) > builds the bundle on success (regression: gate must not block valid flows) 576ms
+ ✓ tests/scope-compiler.test.ts (25 tests) 11ms
+ ✓ tests/bin.test.ts (7 tests) 3653ms
+   ✓ built flows binary > refuses through a symlink to the built artifact 816ms
+   ✓ built flows binary > refuses through a symlinked directory component 671ms
+   ✓ built flows binary > classifies a signal-terminated auth probe as probe_failed 675ms
+   ✓ built flows binary > classifies an unavailable PATH resolver as probe_failed 592ms
+   ✓ built flows binary > does not describe a present non-executable CLI as missing 488ms
+   ✓ built flows binary > runs one auth probe for three steps sharing a flow CLI 408ms
+ ✓ tests/communication-worker.test.ts (15 tests) 1493ms
+ ✓ tests/hn-poller.test.ts (6 tests) 7ms
+ ✓ tests/plugin-add.test.ts (7 tests) 1883ms
+   ✓ installs a real offline npm fixture and includes declarations 476ms
+   ✓ typechecks the augmented verb and rejects unknown namespaces 1394ms
+ ✓ tests/authored-step-failed-exit.test.ts (3 tests) 8ms
+ ✓ tests/direct-run-failure.test.ts (8 tests) 11ms
+ ✓ tests/run-from-digest.test.ts (6 tests) 5598ms
+   ✓ flows run digest input > submits the sealed canonical spec through the normal journal path without checkout 512ms
+   ✓ flows run digest input > uses a verified cache hit even after the bucket is removed 437ms
+   ✓ flows run digest input > resolves deploy.bucket from flows.json and honors explicit override 1496ms
+   ✓ flows run digest input > refuses an unconfigured bucket 1215ms
+   ✓ flows run digest input > refuses tampered spec.canonical.json before creating run data 1095ms
+   ✓ flows run digest input > refuses tampered identity.json before creating run data 842ms
+ ✓ tests/dir-watcher-poller.test.ts (6 tests) 4ms
+ ✓ tests/model-pricing.test.ts (10 tests) 5ms
+ ❯ tests/provider-trigger-executor.test.ts (4 tests | 3 failed) 18ms
+   × the kernel executes compiled 'app_mention' subscriptions with provider isolation and durable dedupe 10ms
+     → spawnSync /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+   × the kernel executes compiled 'reaction_added' subscriptions with provider isolation and durable dedupe 3ms
+     → spawnSync /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+   × the kernel executes compiled 'pull_request' subscriptions with provider isolation and durable dedupe 3ms
+     → spawnSync /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+ ✓ tests/transcript-tail-close.test.ts (2 tests) 913ms
+   ✓ a stalled transcript-tail close > does not hold the spawn open past its bounded window 476ms
+   ✓ a stalled tail close beside a transcript that finished > still journals the transcript pointer 437ms
+ ✓ tests/wrapper-artifacts-cwd.test.ts (2 tests) 77ms
+ ✓ tests/yaml-helper-live.test.ts (1 test) 1735ms
+   ✓ runs compiled YAML helpers through the built CLI and kernel effect journal 1735ms
+ ✓ tests/hello-deterministic.test.ts (5 tests) 17ms
+ ✓ tests/cli-adapter.test.ts (4 tests) 5ms
+ ✓ tests/transcript-exclusion-timeout.test.ts (1 test) 187ms
+ ✓ tests/work-package-validator.test.ts (7 tests) 6ms
+ ❯ tests/communication-mixed-resume.test.ts (1 test | 1 failed) 13ms
+   × resumes mixed ordinary and linked agents through the real daemon without stealing peer capacity 13ms
+     → ENOENT: no such file or directory, open '/tmp/communication-resume-vfA6bi/data/connection.json'
+ ✓ tests/authored-use-loader.test.ts (5 tests) 709ms
+ ✓ tests/cli-probe.test.ts (6 tests) 983ms
+ ✓ tests/cli-answer.test.ts (15 tests) 9ms
+ ✓ tests/authored-declined-live.test.ts (1 test) 1800ms
+   ✓ runs an input guard and resumes its completed declined root without repeated effects 1800ms
+ ✓ tests/agent-relay-hardening.test.ts (12 tests) 12ms
+ ✓ tests/bundle-preflight.test.ts (4 tests) 892ms
+   ✓ bundle execution preflight > ignores surrounding cache configuration on a verified cache hit 447ms
+   ✓ bundle execution preflight > uses the built alias for a nameless flow even in a digest-only cache directory 409ms
+ ✓ tests/communication-preflight.test.ts (13 tests) 34ms
+ ↓ tests/real-cli-adapters.test.ts (3 tests | 3 skipped)
+ ✓ tests/memoization.test.ts (57 tests) 51ms
+ ✓ tests/fs-descriptor.test.ts (1 test) 6ms
+ ✓ tests/classify-outcome.test.ts (2 tests) 2161ms
+   ✓ classifyOutcome > gives up and reports when a running run never becomes classifiable 2007ms
+ ✓ tests/authored-preflight.test.ts (3 tests) 111ms
+ ✓ tests/parse-json-output.test.ts (7 tests) 5ms
+ ✓ tests/journal-client-completion.test.ts (4 tests) 102ms
+ ✓ tests/communication-environment-preflight.test.ts (6 tests) 8ms
+ ✓ tests/budget-authored-live.test.ts (2 tests) 307ms
+ ✓ tests/slack-writeback.test.ts (1 test) 258ms
+ ✓ tests/authored-surface-authority.test.ts (2 tests) 225ms
+ ✓ tests/adapters/claude.test.ts (7 tests) 5ms
+ ✓ tests/worker-cli-abort.test.ts (2 tests) 2649ms
+   ✓ stops claude and its process group when lease ownership is lost 1301ms
+   ✓ stops wrapper.mjs and its process group when lease ownership is lost 1347ms
+ ✓ tests/worker-cli-cwd.test.ts (2 tests) 265ms
+ ✓ tests/adapters/codex.test.ts (7 tests) 4ms
+ ✓ tests/communication-history.test.ts (1 test) 3ms
+ ✓ tests/adapters/registry.test.ts (4 tests) 4ms
+ ✓ tests/slack-block-kit.test.ts (5 tests) 13ms
+ ✓ tests/promise-ancestry.test.ts (2 tests) 311ms
+   ✓ reachableTargets > handles more than 32 targets and a chain too deep for recursion 308ms
+ ✓ tests/authored-declined-report.test.ts (6 tests) 8ms
+ ✓ tests/communication-refusal.test.ts (1 test) 15ms
+ ✓ tests/agent-cwd-validation.test.ts (2 tests) 424ms
+   ✓ declarative agent cwd > is refused by `flows check` on a YAML flow before anything runs 421ms
+ ✓ tests/catalog-plugins.test.ts (2 tests) 3ms
+ ✓ tests/check-command-cwd.test.ts (1 test) 13ms
+ ✓ tests/communication-lazy.test.ts (1 test) 4ms
+ ✓ tests/cli-progress-wait.test.ts (2 tests) 3ms
+ ↓ tests/run-digest-live.test.ts (1 test | 1 skipped)
+ ✓ tests/placement.test.ts (54 tests) 16ms
+ ✓ tests/bundle-transport.test.ts (20 tests) 2599ms
+   ✓ digest references > accepts and deploys the build output for hello 430ms
+   ✓ digest references > accepts and deploys the build output for Hello 420ms
+   ✓ digest references > accepts and deploys the build output for hello.world 415ms
+   ✓ digest references > accepts and deploys the build output for hello_world 418ms
+   ✓ digest references > accepts and deploys the build output for 123 492ms
+   ✓ digest references > accepts and deploys the build output for A_b.c-1 422ms
+ ✓ tests/canonical-tree.test.ts (1 test) 2ms
+ ✓ tests/communication-tools.test.ts (1 test) 77ms
+ ✓ tests/authored-admission.test.ts (2 tests) 3ms
+ ✓ tests/memory.test.ts (18 tests) 9ms
+ ✓ tests/worker-platform.test.ts (1 test) 3ms
+ ✓ tests/run-digest.test.ts (4 tests) 1610ms
+   ✓ digest run configuration refusals > reports config_invalid before fetching or starting a run for {invalid json 405ms
+   ✓ digest run configuration refusals > reports config_invalid before fetching or starting a run for {"deploy":{}} 397ms
+   ✓ digest run configuration refusals > reports config_invalid before fetching or starting a run for {"deploy":{"bucket":123}} 406ms
+   ✓ digest run configuration refusals > reports config_invalid before fetching or starting a run for {"deploy":{"bucket":""}} 401ms
+ ✓ tests/authored-parallel-llm.test.ts (6 tests) 117134ms
+   ✓ parallel llm capacity 1 > deduplicates concurrent preflight probes 3035ms
+   ✓ parallel llm capacity 1 > completes nine calls without expired child leases during slow preflight 9012ms
+   ✓ parallel llm capacity 1 > keeps the durable root lease alive across two cold models 48242ms
+   ✓ parallel llm capacity 4 > deduplicates concurrent preflight probes 1998ms
+   ✓ parallel llm capacity 4 > completes nine calls without expired child leases during slow preflight 8082ms
+   ✓ parallel llm capacity 4 > keeps the durable root lease alive across two cold models 46754ms
+
+⎯⎯⎯⎯⎯⎯ Failed Suites 2 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-node-runtime.test.ts [ tests/authored-node-runtime.test.ts ]
+AssertionError: expected '1.3.6' to be '1.4.0' // Object.is equality
+
+Expected: "1.4.0"
+Received: "1.3.6"
+
+ ❯ tests/authored-node-runtime.test.ts:18:77
+     16| 
+     17| beforeAll(() => {
+     18|   expect(spawnSync(bun, ['--version'], { encoding: 'utf8' }).stdout.tr…
+       |                                                                             ^
+     19|   expect(existsSync(daemon), 'build the current kernel or set RELAYFLO…
+     20|   stage = mkdtempSync(join(tmpdir(), 'authored-standalone-build-'));
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/69]⎯
+
+ FAIL  tests/mcp.test.ts > authored MCP effects against the real kernel
+Error: journal client: connect failed: connect ENOENT /tmp/relayflowd-1c26badc8d89.sock
+ ❯ Socket.onError src/journal-client.ts:102:16
+    100|         socket.removeAllListeners();
+    101|         this.failAll(err);
+    102|         reject(new Error(`journal client: connect failed: ${err.messag…
+       |                ^
+    103|       };
+    104|       socket.once('error', onError);
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/69]⎯
+
+⎯⎯⎯⎯⎯⎯ Failed Tests 67 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/canonical-software-factory.test.ts > canonical software-factory metadata contract > opens the actual catalog flow with the ticket title and exactly one GitHub closing line
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ runCanonical tests/canonical-software-factory.test.ts:69:22
+     67|   };
+     68| 
+     69|   const definition = getFlowDefinition(softwareFactory);
+       |                      ^
+     70|   return definition.body(context as never, { issue, approver: 'khaliq'…
+     71|     root,
+ ❯ tests/canonical-software-factory.test.ts:82:26
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[3/69]⎯
+
+ FAIL  tests/canonical-software-factory.test.ts > canonical software-factory metadata contract > normalizes whitespace and caps the title at 240 Unicode code points
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ runCanonical tests/canonical-software-factory.test.ts:69:22
+     67|   };
+     68| 
+     69|   const definition = getFlowDefinition(softwareFactory);
+       |                      ^
+     70|   return definition.body(context as never, { issue, approver: 'khaliq'…
+     71|     root,
+ ❯ tests/canonical-software-factory.test.ts:99:26
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[4/69]⎯
+
+ FAIL  tests/canonical-software-factory.test.ts > canonical software-factory metadata contract > fails closed before push when GitHub identity is missing or the final body duplicates its closing line
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ tests/canonical-software-factory.test.ts:108:24
+    106| 
+    107|   it('fails closed before push when GitHub identity is missing or the …
+    108|     const definition = getFlowDefinition(softwareFactory);
+       |                        ^
+    109|     const commands: string[] = [];
+    110|     let completionReason = '';
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[5/69]⎯
+
+ FAIL  tests/communication-mixed-resume.test.ts > resumes mixed ordinary and linked agents through the real daemon without stealing peer capacity
+Error: ENOENT: no such file or directory, open '/tmp/communication-resume-vfA6bi/data/connection.json'
+ ❯ tests/communication-mixed-resume.test.ts:54:35
+     52|   } finally {
+     53|     clearTimeout(timeout); state.release(); client.close();
+     54|     try { process.kill(JSON.parse(readFileSync(join(dataDir, 'connecti…
+       |                                   ^
+     55|     finally { rmSync(root, { recursive: true, force: true }); }
+     56|   }
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[6/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > executes the exact capability-only handler for a queued receipt
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > executes the exact capability-only handler for a duplicate receipt
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:221:26
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[7/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > launches through the captured process primitive after builtin export synchronization
+AssertionError: promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:283:11
+    281|           input: descriptor(),
+    282|           babysitterTurn: { queue: async () => ({ receiptId: 'receipt-…
+    283|         })).resolves.toEqual({ completionReason: 'success', capability…
+       |           ^
+    284|       } finally {
+    285|         process.execPath = originalExecPath;
+
+Caused by: Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:277:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[8/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > ignores inherited launcher overrides and decodes manifests with the captured Buffer intrinsic
+AssertionError: promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:373:11
+    371|           input: descriptor('delivery-options'),
+    372|           babysitterTurn: { queue: async () => ({ receiptId: 'receipt-…
+    373|         })).resolves.toEqual({ completionReason: 'success', capability…
+       |           ^
+    374|       } finally {
+    375|         Buffer.prototype.toString = bufferToString;
+
+Caused by: Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:367:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[9/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > streams verified bytes when the live store is replaced and no writable staging path exists
+AssertionError: promise rejected "Error: Hosted extension sandbox exited wi… { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:431:7
+    429|         return { receiptId: 'receipt-1', status: 'queued' };
+    430|       } },
+    431|     })).resolves.toEqual({ completionReason: 'success', capabilityCall…
+       |       ^
+    432|     expect(calls).toBe(1);
+    433|     expect(readFileSync(join(installed.directory, 'babysitter.flow.ts'…
+
+Caused by: Error: Hosted extension sandbox exited without a valid completion (exit 1): /tmp/hosted-bwrap-wrapper-VmLyve/bwrap-wrapper:20
+if (child.error) throw child.error;
+                 ^
+
+Error: spawnSync /usr/bin/bwrap ENOENT
+    at Object.spawnSync (node:internal/child_process:1103:20)
+    at spawnSync (node:child_process:911:24)
+    at Object.<anonymous> (/tmp/hosted-bwrap-wrapper-VmLyve/bwrap-wrapper:19:15)
+    at Module._compile (node:internal/modules/cjs/loader:1809:14)
+    at Object..js (node:internal/modules/cjs/loader:1940:10)
+    at Module.load (node:internal/modules/cjs/loader:1530:32)
+    at Module._load (node:internal/modules/cjs/loader:1332:12)
+    at wrapModuleLoad (node:internal/modules/cjs/loader:255:19)
+    at Module.executeUserEntryPoint [as runMain] (node:internal/modules/run_main:154:5)
+    at node:internal/main/run_main_module:33:47 {
+  errno: -2,
+  code: 'ENOENT',
+  syscall: 'spawnSync /usr/bin/bwrap',
+  path: '/usr/bin/bwrap',
+  spawnargs: [
+    '--unshare-all',
+    '--die-with-parent',
+    '--new-session',
+    '--clearenv',
+    '--cap-drop',
+    'ALL',
+    '--dir',
+    '/usr',
+    '--ro-bind',
+    '/usr/lib',
+    '/usr/lib',
+    '--ro-bind',
+    '/usr/lib64',
+    '/usr/lib64',
+    '--ro-bind',
+    '/usr/lib',
+    '/lib',
+    '--ro-bind',
+    '/usr/lib64',
+    '/lib64',
+    '--proc',
+    '/proc',
+    '--dev',
+    '/dev',
+    '--tmpfs',
+    '/tmp',
+    '--dir',
+    '/runtime',
+    '--dir',
+    '/extension',
+    '--dir',
+    '/extension/node_modules',
+    '--dir',
+    '/extension/node_modules/@relayflows',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface/dist',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface/dist/helpers',
+    '--dir',
+    '/extension/node_modules/@relayflows/surface/dist/triggers',
+    '--dir',
+    '/extension/src',
+    '--perms',
+    '0500',
+    '--ro-bind-data',
+    '4',
+    '/runtime/node',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '5',
+    '/runtime/runner.mjs',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '6',
+    '/extension/node_modules/@relayflows/surface/package.json',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '7',
+    '/extension/node_modules/@relayflows/surface/index.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '8',
+    '/extension/node_modules/@relayflows/surface/runtime.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '9',
+    '/extension/node_modules/@relayflows/surface/dist/flow.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '10',
+    '/extension/node_modules/@relayflows/surface/dist/helpers/providers.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '11',
+    '/extension/node_modules/@relayflows/surface/dist/provider-trigger.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '12',
+    '/extension/node_modules/@relayflows/surface/dist/schedule.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '13',
+    '/extension/node_modules/@relayflows/surface/dist/triggers.js',
+    '--perms',
+    '0400',
+    '--ro-bind-data',
+    '14',
+    '/extension/node_modules/@relayflows/surface/dist/triggers/github.js',
+    '--perms',
+    ... 27 more items
+  ]
+}
+
+Node.js v25.6.0
+
+ ❯ Object.<anonymous> ../../../../../../../tmp/hosted-bwrap-wrapper-VmLyve/bwrap-wrapper:19:15
+ ❯ refuse src/hosted-extension-protocol.ts:135:21
+ ❯ ChildProcess.<anonymous> src/hosted-extension-protocol.ts:234:21
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[10/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > mounts pinned private Surface bytes when the live package changes before launch
+Error: ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-48daMa/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:437:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[11/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > refuses Surface runtime bytes that differ from the reviewed pin before launch
+Error: ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-nbb4yF/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:462:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[12/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > refuses oversized Surface files through the bounded descriptor reader
+Error: ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-0JhPEY/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:483:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[13/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > shields verified Surface files before async settlement
+Error: ENOENT: no such file or directory, copyfile '/home/daytona/.relayflow-v2-supervisor/durable/repository/packages/surface/dist/flow.js' -> '/tmp/hosted-surface-test-SpYDeT/dist/flow.js'
+ ❯ surfaceFixture tests/hosted-extension-isolation.test.ts:109:5
+    107|     const target = join(surfaceRoot, 'dist', file);
+    108|     mkdirSync(resolve(target, '..'), { recursive: true });
+    109|     copyFileSync(join(sourceRoot, 'dist', file), target);
+       |     ^
+    110|   }
+    111|   return surfaceRoot;
+ ❯ tests/hosted-extension-isolation.test.ts:507:25
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[14/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > preserves a typed host refusal while disclosing only a fixed marker to the child
+AssertionError: expected Error: bubblewrap is unavailable { code: '…' } to be Error: private Cloud policy detail { code: '…' } // Object.is equality
+
+- Expected
++ Received
+
+- [Error: private Cloud policy detail]
++ [Error: bubblewrap is unavailable]
+
+ ❯ tests/hosted-extension-isolation.test.ts:560:5
+    558|       provider: 'github', eventType: 'pull_request.labeled', deliveryI…
+    559|     });
+    560|     await expect(runVerifiedNativeExtensionSandbox({
+       |     ^
+    561|       artifact: await artifact(source), manifest: validateFlowExtensio…
+    562|       babysitterTurn: { queue: async () => { throw refusal; } },
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[15/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > denies ambient credentials, host files, writes, network, subprocesses, and undeclared context verbs
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:624:13
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[16/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > enforces OS address-space and data bounds on native Buffer allocation
+AssertionError: expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_unsupported', …(1) }
+
+- Expected
++ Received
+
+- Object {
++ PluginError {
+    "code": "plugin_unsupported",
+-   "message": StringMatching /(?:Failed to allocate memory|Array buffer allocation failed)/u,
+  }
+
+ ❯ tests/hosted-extension-isolation.test.ts:646:5
+    644|     });
+    645|     let calls = 0;
+    646|     await expect(runVerifiedNativeExtensionSandbox({
+       |     ^
+    647|       artifact: installed,
+    648|       manifest: validateFlowExtensionManifest(manifest()),
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[17/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > blocks extra handler fields and authority-bearing receipt fields at the parent port
+AssertionError: expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+
+- Expected
++ Received
+
+- Object {
+-   "code": "plugin_event_unroutable",
++ PluginError {
++   "code": "plugin_unsupported",
+  }
+
+ ❯ tests/hosted-extension-isolation.test.ts:675:5
+    673|     });
+    674|     let calls = 0;
+    675|     await expect(runVerifiedNativeExtensionSandbox({
+       |     ^
+    676|       artifact: await artifact(source), manifest: validateFlowExtensio…
+    677|       babysitterTurn: { queue: async () => { calls += 1; return { rece…
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[18/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > constructs adapter authority with the captured freeze intrinsic
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:745:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[19/69]⎯
+
+ FAIL  tests/hosted-extension-isolation.test.ts > hosted extension capability isolation > writes the Surface manifest and protocol without inherited toJSON behavior
+AssertionError: promise rejected "Error: bubblewrap is unavailable { code: '…' }" instead of resolving
+ ❯ tests/hosted-extension-isolation.test.ts:788:11
+    786|           babysitterTurn: { queue: async () => ({ receiptId: 'receipt-…
+    787|           timeoutMs: 3_000,
+    788|         })).resolves.toEqual({ completionReason: 'success', capability…
+       |           ^
+    789|       } finally {
+    790|         if (previous === undefined) delete (Object.prototype as { toJS…
+
+Caused by: Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-isolation.test.ts:781:22
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[20/69]⎯
+
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > uses captured JSON intrinsics for the complete parent boundary
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-protocol.test.ts:309:24
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[21/69]⎯
+
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > rejects an import-time different PR frame with zero adapter calls
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > rejects an import-time different delivery frame with zero adapter calls
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > rejects an import-time different event frame with zero adapter calls
+AssertionError: expected Error: bubblewrap is unavailable { code: '…' } to match object { code: 'plugin_event_unroutable' }
+
+- Expected
++ Received
+
+- Object {
+-   "code": "plugin_event_unroutable",
++ PluginError {
++   "code": "plugin_unsupported",
+  }
+
+ ❯ tests/hosted-extension-protocol.test.ts:430:5
+    428|   ])('rejects an import-time %s frame with zero adapter calls', async …
+    429|     let calls = 0;
+    430|     await expect(runVerifiedNativeExtensionSandbox({
+       |     ^
+    431|       artifact: await artifact(hostileImport([frame, { type: 'error', …
+    432|       manifest: validateFlowExtensionManifest(manifest()), dispatch: d…
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[22/69]⎯
+
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > rejects two forged calls after the authoritative first outcome settles
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > waits for a pending adapter to reject after a forged child error
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > waits for a pending adapter to resolve after a forged child error
+ FAIL  tests/hosted-extension-protocol.test.ts > hosted extension hostile protocol > returns a typed adapter rejection even when the hostile child hangs
+Error: hostile child did not invoke the adapter
+ ❯ Timeout._onTimeout tests/hosted-extension-protocol.test.ts:118:45
+    116| async function waitForInvocation(invoked: Promise<void>): Promise<void…
+    117|   await new Promise<void>((resolve, reject) => {
+    118|     const timeout = setTimeout(() => reject(new Error('hostile child d…
+       |                                             ^
+    119|     void invoked.then(() => { clearTimeout(timeout); resolve(); }, rej…
+    120|   });
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[23/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > runs hn-monitor analyze-story end-to-end via a stub agent CLI (gate 2 clause 2 demo)
+AssertionError: expected { …(12) } to match object { output: { …(3) }, …(1) }
+(22 matching properties omitted from actual)
+
+- Expected
++ Received
+
+  Object {
+-   "output": Object {
+-     "reasoning": "stub agent runtime — deterministic output for gate-2 clause-2 demo",
+-     "relevance_score": 5,
+-     "story_title": "stub",
+-   },
++   "output": null,
+    "verification": Object {
+-     "gate": "json_schema",
+-     "verdict": "pass",
++     "gate": "execution",
++     "verdict": "fail",
+    },
+  }
+
+ ❯ tests/live-kernel.test.ts:657:36
+    655|         && (entry as { step_id?: string }).step_id === 'analyze-story',
+    656|     ) as { payload: { output: unknown; verification: unknown } } | und…
+    657|     expect(stepCompleted?.payload).toMatchObject({
+       |                                    ^
+    658|       output: {
+    659|         story_title: 'stub',
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[24/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > hn-monitor analyze-story FAILS verification when the CLI omits required schema fields
+AssertionError: expected { …(12) } to match object { …(3) }
+(21 matching properties omitted from actual)
+
+- Expected
++ Received
+
+  Object {
+-   "completionReason": "retries_exhausted",
++   "completionReason": "worker_error",
+    "output": null,
+    "verification": Object {
+-     "gate": "json_schema",
++     "gate": "execution",
+      "verdict": "fail",
+    },
+  }
+
+ ❯ tests/live-kernel.test.ts:752:36
+    750|     // its verification record names the json_schema rejection. The re…
+    751|     // parsed value is nulled before the completion is persisted.
+    752|     expect(stepCompleted?.payload).toMatchObject({
+       |                                    ^
+    753|       completionReason: 'retries_exhausted',
+    754|       output: null,
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[25/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > agent step preserves the CliResult wrapper as output when the CLI emits non-JSON text
+AssertionError: expected null not to be null
+ ❯ tests/live-kernel.test.ts:823:24
+    821|     // here (parseJsonOutput returned null on non-JSON stdout) and
+    822|     // these assertions would all fail.
+    823|     expect(output).not.toBeNull();
+       |                        ^
+    824|     expect(output.exit_code).toBe(0);
+    825|     expect(output.stdout_tail).toContain('looked at the story');
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[26/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > AgentWorker exposes wake_context to the CLI via RELAYFLOW_WAKE_CONTEXT env var (real analyzer prerequisite)
+TypeError: Cannot read properties of null (reading 'story_title')
+ ❯ tests/live-kernel.test.ts:891:42
+    889|     ) as { payload: { output: { story_title: string; reasoning: string…
+    890|     expect(stepCompleted).toBeDefined();
+    891|     expect(stepCompleted!.payload.output.story_title).toBe(`echoed:${s…
+       |                                          ^
+    892|     expect(stepCompleted!.payload.output.reasoning).toContain(String(s…
+    893| 
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[27/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > AgentWorker leaves RELAYFLOW_WAKE_CONTEXT UNSET when the run has no wake_context (undefined-vs-null pin)
+TypeError: Cannot read properties of null (reading 'env_present')
+ ❯ tests/live-kernel.test.ts:958:38
+    956|     ) as { payload: { output: { env_present: boolean } } } | undefined;
+    957|     expect(completed).toBeDefined();
+    958|     expect(completed!.payload.output.env_present).toBe(false);
+       |                                      ^
+    959| 
+    960|     delete process.env.RELAYFLOW_WAKE_CONTEXT;
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[28/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > AgentWorker leaves RELAYFLOW_MODEL UNSET when the step declares no model
+TypeError: Cannot read properties of null (reading 'story_title')
+ ❯ tests/live-kernel.test.ts:1194:38
+    1192|     expect(completed).toBeDefined();
+    1193|     // UNSET, not EMPTY and not the leaked parent value.
+    1194|     expect(completed!.payload.output.story_title).toBe('model:UNSET');
+       |                                      ^
+    1195| 
+    1196|     delete process.env.RELAYFLOW_MODEL;
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[29/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > hn-monitor analyze-story reaches done through the real Claude analyzer CLI
+Error: LIVE_ANALYZER_UNAVAILABLE: "/home/daytona/.relayflow-v2-supervisor/durable/repository/testdata/preflight/analyze-story-claude-cli" does not identify as relayflows-agent-cli-v1 — failing because gate-2 acceptance requires the real analyzer to execute. Set RELAYFLOWS_ALLOW_ANALYZER_SKIP=1 only if this run is not gate evidence.
+ ❯ tests/live-kernel.test.ts:1223:15
+    1221|       const notice = `LIVE_ANALYZER_UNAVAILABLE: ${readiness.detail}`;
+    1222|       if (process.env['RELAYFLOWS_ALLOW_ANALYZER_SKIP'] !== '1') {
+    1223|         throw new Error(
+       |               ^
+    1224|           `${notice} — failing because gate-2 acceptance requires the …
+    1225|           + 'Set RELAYFLOWS_ALLOW_ANALYZER_SKIP=1 only if this run is …
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[30/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > starts exactly one daemon when two runs race for one empty data dir
+AssertionError: WARNING [unprovable_effects] Step "greet" command "echo" resolves, but its effects cannot be proven before execution.
+WARNING [unprovable_effects] Step "shout" command "echo" resolves, but its effects cannot be proven before execution.
+WARNING [editor_schema_missing] For editor validation, add this first line: # yaml-language-server: $schema=https://schema.relayflows.dev/v0.1/flows.schema.json
+REFUSED [relayflowd_not_found] No relayflowd binary could be found. Install the runtime package for this host (@relayflows/runtime-linux-x64), or set RELAYFLOWD_BIN to a relayflowd executable. Tried: /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/dist/relayflowd.
+: expected 2 to be +0 // Object.is equality
+
+- Expected
++ Received
+
+- 0
++ 2
+
+ ❯ tests/live-kernel.test.ts:1388:40
+    1386|     ]);
+    1387| 
+    1388|     expect(first.status, first.stderr).toBe(0);
+       |                                        ^
+    1389|     expect(second.status, second.stderr).toBe(0);
+    1390|     expect(first.stdout).toContain('completionReason: success');
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[31/69]⎯
+
+ FAIL  tests/live-kernel.test.ts > a relayflow can be scheduled: tick source against live relayflowd > a tick spawns a real run whose step reports the SCHEDULED instant
+AssertionError: expected null to deeply equal { schedule_id: 'heartbeat-1m', …(3) }
+
+- Expected: 
+Object {
+  "lag_ms": 43000,
+  "schedule_id": "heartbeat-1m",
+  "scheduled_for_ms": 1764000000000,
+  "slot": 29400000,
+}
+
++ Received: 
+null
+
+ ❯ tests/live-kernel.test.ts:1665:39
+    1663|     // The bound: the run reports the grid instant and its own lag, so…
+    1664|     // backfilled run can tell it is running for a slot from the past.
+    1665|     expect(completed!.payload.output).toEqual({
+       |                                       ^
+    1666|       schedule_id: 'heartbeat-1m',
+    1667|       slot: 29_400_000,
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[32/69]⎯
+
+ FAIL  tests/provider-trigger-executor.test.ts > the kernel executes compiled 'app_mention' subscriptions with provider isolation and durable dedupe
+ FAIL  tests/provider-trigger-executor.test.ts > the kernel executes compiled 'reaction_added' subscriptions with provider isolation and durable dedupe
+ FAIL  tests/provider-trigger-executor.test.ts > the kernel executes compiled 'pull_request' subscriptions with provider isolation and durable dedupe
+Error: spawnSync /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+ ❯ submit tests/provider-trigger-executor.test.ts:43:89
+     41|     steps: [{ id: 'effect', type: 'deterministic', command: `printf ac…
+     42|   }))));
+     43|   const submit = (envelope: unknown, key: string, executor = source.na…
+       |                                                                                         ^
+     44|     '--data-dir', dir, 'run', spec, '--event', JSON.stringify({ type: …
+     45|   ], { encoding: 'utf8', stdio: 'pipe' })) as { matched: boolean; dedu…
+ ❯ tests/provider-trigger-executor.test.ts:50:12
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[33/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage input validation > refuses an 8-character run-id prefix: Cloud has no prefix lookup
+AssertionError: expected [Function] to throw error matching /not full Cloud run ids: c649fe14/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/not full Cloud run ids: c649fe14/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[34/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage input validation > refuses the whole batch when any id is invalid, rather than dropping it
+AssertionError: expected [Function] to throw error matching /not full Cloud run ids: nope!/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/not full Cloud run ids: nope!/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[35/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage input validation > refuses an empty batch
+AssertionError: expected [Function] to throw error matching /needs runIds/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/needs runIds/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[36/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage input validation > refuses a batch too large for the edge step lease
+AssertionError: expected [Function] to throw error matching /exceeds the 8 that fit/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/exceeds the 8 that fit/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[37/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage input validation > accepts eight ids — the incident batch is inside the bound
+AssertionError: promise rejected "TypeError: expected an @relayflows/surfac…" instead of resolving
+ ❯ tests/stuck-run-triage.test.ts:62:40
+     60|   it('accepts eight ids — the incident batch is inside the bound', asy…
+     61|     const ids = Array.from({ length: 8 }, (_, i) => `${ID_A.slice(0, -…
+     62|     await expect(drive({ runIds: ids })).resolves.toBeDefined();
+       |                                        ^
+     63|   });
+     64| });
+
+Caused by: TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+ ❯ tests/stuck-run-triage.test.ts:62:18
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[38/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage apiUrl > refuses to send the Cloud bearer token to an unapproved origin
+AssertionError: expected [Function] to throw error matching /refusing to send the Cloud bearer to…/\ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/refusing to send the Cloud bearer token to https:\/\/evil\.example/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[39/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage apiUrl > refuses a non-URL apiUrl
+AssertionError: expected [Function] to throw error matching /is not a URL/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/is not a URL/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[40/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage apiUrl > allows an approved origin and uses it in the curl
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:77:23
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[41/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage apiUrl > defaults to production Cloud
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:82:23
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[42/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage apiUrl > never publishes a run record the fetch did not produce
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:89:34
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[43/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage edge collection > names the Worker on every wrangler invocation
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:98:33
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[44/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage edge collection > accepts caller-supplied Workers and rejects option-shaped ones
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:107:33
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[45/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage edge collection > falls back when GNU timeout is absent, as it is on macOS
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:115:33
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[46/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage edge collection > runs the tails concurrently so wall time does not scale with the batch
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:123:33
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[47/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage edge collection > records wrangler's own exit status rather than head's
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:129:33
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[48/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage shell text > parses under both sh and bash
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:137:23
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[49/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage shell text > collects tails with no GNU timeout on PATH, as on a stock macOS
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:157:33
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[50/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage agents > declares read-only permissions on every agent
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:176:23
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[51/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage agents > tells the forensics agents their evidence is untrusted
+TypeError: expected an @relayflows/surface flow handle
+ ❯ Module.getFlowDefinition node_modules/@relayflows/surface/src/flow.ts:149:11
+ ❯ drive tests/stuck-run-triage.test.ts:34:9
+     32|     done: () => {},
+     33|   };
+     34|   await getFlowDefinition<StuckRunTriageInput>(triage).body(f as never…
+       |         ^
+     35|   return rec;
+     36| }
+ ❯ tests/stuck-run-triage.test.ts:182:23
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[52/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage fan-out > refuses a duplicate run id: two tails would share one evidence file
+AssertionError: expected [Function] to throw error matching /duplicate runIds: c649fe14-0c2e-4e51-…/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/duplicate runIds: c649fe14-0c2e-4e51-9a6a-4f0d1b0f77aa/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[53/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage fan-out > refuses a duplicate Worker name for the same reason
+AssertionError: expected [Function] to throw error matching /duplicate workers: w-one/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/duplicate workers: w-one/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[54/69]⎯
+
+ FAIL  tests/stuck-run-triage.test.ts > stuck-run-triage fan-out > bounds ids x workers, not just ids
+AssertionError: expected [Function] to throw error matching /24 concurrent tails, over the 16/ but got 'expected an @relayflows/surface flow …'
+
+- Expected: 
+/24 concurrent tails, over the 16/
+
++ Received: 
+"expected an @relayflows/surface flow handle"
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[55/69]⎯
+
+ FAIL  tests/webhook-live.test.ts > executes and deduplicates 'app_mention' only for its provider and matching payload
+ FAIL  tests/webhook-live.test.ts > executes and deduplicates 'reaction_added' only for its provider and matching payload
+ FAIL  tests/webhook-live.test.ts > executes and deduplicates 'pull_request' only for its provider and matching payload
+Error: webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+ ❯ until tests/webhook-live.test.ts:39:9
+     37|   const deadline = Date.now() + 10_000;
+     38|   while (Date.now() < deadline) { if (await predicate()) return; await…
+     39|   throw new Error(`webhook integration timed out: ${detail()}`);
+       |         ^
+     40| }
+     41| async function daemon(dir: string): Promise<ChildProcess> {
+ ❯ daemon tests/webhook-live.test.ts:43:3
+ ❯ tests/webhook-live.test.ts:100:3
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[56/69]⎯
+
+ FAIL  tests/webhook-live.test.ts > flows serve-webhook writes JSON before the daemon starts, then journals and archives exactly once
+Error: webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+ ❯ until tests/webhook-live.test.ts:39:9
+     37|   const deadline = Date.now() + 10_000;
+     38|   while (Date.now() < deadline) { if (await predicate()) return; await…
+     39|   throw new Error(`webhook integration timed out: ${detail()}`);
+       |         ^
+     40| }
+     41| async function daemon(dir: string): Promise<ChildProcess> {
+ ❯ daemon tests/webhook-live.test.ts:43:3
+ ❯ tests/webhook-live.test.ts:121:3
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[57/69]⎯
+
+ FAIL  tests/webhook-live.test.ts > replays a dropped file after SIGKILL before spawn
+Error: webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+ ❯ until tests/webhook-live.test.ts:39:9
+     37|   const deadline = Date.now() + 10_000;
+     38|   while (Date.now() < deadline) { if (await predicate()) return; await…
+     39|   throw new Error(`webhook integration timed out: ${detail()}`);
+       |         ^
+     40| }
+     41| async function daemon(dir: string): Promise<ChildProcess> {
+ ❯ daemon tests/webhook-live.test.ts:43:3
+ ❯ tests/webhook-live.test.ts:137:17
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[58/69]⎯
+
+ FAIL  tests/webhook-live.test.ts > resumes the same journal after SIGKILL after spawn and before acknowledgement
+Error: webhook integration timed out: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/debug/relayflowd ENOENT
+ ❯ until tests/webhook-live.test.ts:39:9
+     37|   const deadline = Date.now() + 10_000;
+     38|   while (Date.now() < deadline) { if (await predicate()) return; await…
+     39|   throw new Error(`webhook integration timed out: ${detail()}`);
+       |         ^
+     40| }
+     41| async function daemon(dir: string): Promise<ChildProcess> {
+ ❯ daemon tests/webhook-live.test.ts:43:3
+ ❯ tests/webhook-live.test.ts:150:17
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[59/69]⎯
+
+⎯⎯⎯⎯⎯⎯ Unhandled Errors ⎯⎯⎯⎯⎯⎯
+
+Vitest caught 2 unhandled errors during the test run.
+This might cause false positive tests. Resolve unhandled errors to make sure your tests are not affected.
+
+⎯⎯⎯⎯ Unhandled Rejection ⎯⎯⎯⎯⎯
+Error: bubblewrap is unavailable
+ ❯ unsupported src/hosted-extension-sandbox.ts:488:9
+    486| 
+    487| function unsupported(message: string): never {
+    488|   throw new PluginError('plugin_unsupported', message);
+       |         ^
+    489| }
+    490| 
+ ❯ executable src/hosted-extension-sandbox.ts:469:18
+ ❯ Module.runHostedExtensionSandbox src/hosted-extension-sandbox.ts:160:17
+ ❯ Module.runVerifiedNativeExtensionSandbox src/hosted-extension-isolation.ts:173:16
+ ❯ tests/hosted-extension-protocol.test.ts:454:17
+ ❯ node_modules/@vitest/runner/dist/index.js:533:5
+ ❯ runTest node_modules/@vitest/runner/dist/index.js:1056:11
+ ❯ runSuite node_modules/@vitest/runner/dist/index.js:1205:15
+ ❯ runSuite node_modules/@vitest/runner/dist/index.js:1205:15
+ ❯ runFiles node_modules/@vitest/runner/dist/index.js:1262:5
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { code: 'plugin_unsupported' }
+This error originated in "tests/hosted-extension-protocol.test.ts" test file. It doesn't mean the error was thrown inside the file itself, but while it was running.
+The latest test that might've caused the error is "rejects two forged calls after the authoritative first outcome settles". It might mean one of the following:
+- The error was thrown, while Vitest was running this test.
+- If the error occurred after the test had been completed, this was the last documented test before it was thrown.
+
+⎯⎯⎯⎯⎯ Uncaught Exception ⎯⎯⎯⎯⎯
+Error: spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/release/relayflowd ENOENT
+ ❯ Process.ChildProcess._handle.onexit node:internal/child_process:285:19
+ ❯ onErrorNT node:internal/child_process:483:16
+ ❯ processTicksAndRejections node:internal/process/task_queues:90:21
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+Serialized Error: { errno: -2, code: 'ENOENT', syscall: 'spawn /home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/release/relayflowd', path: '/home/daytona/.relayflow-v2-supervisor/durable/repository/kernel/target/release/relayflowd', spawnargs: [ '--data-dir', '/tmp/flows-mcp-daemon-xHMNd5', 'serve' ] }
+This error originated in "tests/mcp.test.ts" test file. It doesn't mean the error was thrown inside the file itself, but while it was running.
+The latest test that might've caused the error is "authored MCP effects against the real kernel". It might mean one of the following:
+- The error was thrown, while Vitest was running this test.
+- If the error occurred after the test had been completed, this was the last documented test before it was thrown.
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+
+ Test Files  10 failed | 172 passed | 3 skipped (185)
+      Tests  67 failed | 2810 passed | 26 skipped (2903)
+     Errors  2 errors
+   Start at  05:46:25
+   Duration  294.12s (transform 3.04s, setup 0ms, collect 47.25s, tests 766.88s, environment 25ms, prepare 8.05s)
+
diff --git a/evidence/parallel-llm-fix/standalone-baseline-build.txt b/evidence/parallel-llm-fix/standalone-baseline-build.txt
new file mode 100644
index 000000000..1001208c9
--- /dev/null
+++ b/evidence/parallel-llm-fix/standalone-baseline-build.txt
@@ -0,0 +1,6 @@
+Bundled 757 modules in 52ms
+
+  authored-node.mjs  3.1 MB  (entry point)
+
+ [116ms]  bundle  690 modules
+   [3ms] compile  /tmp/flows-parallel-base
diff --git a/evidence/parallel-llm-fix/standalone-baseline-smoke.txt b/evidence/parallel-llm-fix/standalone-baseline-smoke.txt
new file mode 100644
index 000000000..99aacb98b
--- /dev/null
+++ b/evidence/parallel-llm-fix/standalone-baseline-smoke.txt
@@ -0,0 +1 @@
+REFUSED [invalid_spec] Flow "/tmp/flows-prompt-lab/examples/prompt-lab/evidence/runtime-findings/runtime-parallel-llm-repro.flow.ts" could not be imported: unknown authored-flow error
diff --git a/evidence/parallel-llm-fix/standalone-build-final.txt b/evidence/parallel-llm-fix/standalone-build-final.txt
new file mode 100644
index 000000000..1936c935f
--- /dev/null
+++ b/evidence/parallel-llm-fix/standalone-build-final.txt
@@ -0,0 +1,6 @@
+Bundled 759 modules in 54ms
+
+  authored-node.mjs  3.0 MB  (entry point)
+
+ [133ms]  bundle  692 modules
+   [8ms] compile  /tmp/flows-parallel-fixed
diff --git a/evidence/parallel-llm-fix/standalone-build.txt b/evidence/parallel-llm-fix/standalone-build.txt
new file mode 100644
index 000000000..f0c247cf8
--- /dev/null
+++ b/evidence/parallel-llm-fix/standalone-build.txt
@@ -0,0 +1,6 @@
+Bundled 759 modules in 51ms
+
+  authored-node.mjs  3.0 MB  (entry point)
+
+  [72ms]  bundle  692 modules
+   [3ms] compile  /tmp/flows-parallel-fixed
diff --git a/evidence/parallel-llm-fix/standalone-smoke.txt b/evidence/parallel-llm-fix/standalone-smoke.txt
new file mode 100644
index 000000000..99aacb98b
--- /dev/null
+++ b/evidence/parallel-llm-fix/standalone-smoke.txt
@@ -0,0 +1 @@
+REFUSED [invalid_spec] Flow "/tmp/flows-prompt-lab/examples/prompt-lab/evidence/runtime-findings/runtime-parallel-llm-repro.flow.ts" could not be imported: unknown authored-flow error
diff --git a/evidence/parallel-llm-fix/static-refusals.txt b/evidence/parallel-llm-fix/static-refusals.txt
new file mode 100644
index 000000000..309cc56f0
--- /dev/null
+++ b/evidence/parallel-llm-fix/static-refusals.txt
@@ -0,0 +1,12 @@
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+ ✓ tests/authored-preflight.test.ts (3 tests) 120ms
+ ✓ tests/cli-probe.test.ts (6 tests) 866ms
+ ✓ tests/preflight.test.ts (67 tests) 108ms
+
+ Test Files  3 passed (3)
+      Tests  76 passed (76)
+   Start at  05:46:01
+   Duration  1.68s (transform 1.10s, setup 0ms, collect 2.60s, tests 1.09s, environment 0ms, prepare 141ms)
+
diff --git a/evidence/parallel-llm-fix/test-types-final.txt b/evidence/parallel-llm-fix/test-types-final.txt
new file mode 100644
index 000000000..b349a0934
--- /dev/null
+++ b/evidence/parallel-llm-fix/test-types-final.txt
@@ -0,0 +1,4 @@
+
+> @relayflows/sdk@2.0.29 typecheck:tests
+> tsc -p tsconfig.tests.json
+
diff --git a/evidence/parallel-llm-fix/test-types.txt b/evidence/parallel-llm-fix/test-types.txt
new file mode 100644
index 000000000..b349a0934
--- /dev/null
+++ b/evidence/parallel-llm-fix/test-types.txt
@@ -0,0 +1,4 @@
+
+> @relayflows/sdk@2.0.29 typecheck:tests
+> tsc -p tsconfig.tests.json
+
diff --git a/evidence/parallel-llm-fix/typecheck-final.txt b/evidence/parallel-llm-fix/typecheck-final.txt
new file mode 100644
index 000000000..302fb33bd
--- /dev/null
+++ b/evidence/parallel-llm-fix/typecheck-final.txt
@@ -0,0 +1,4 @@
+
+> @relayflows/sdk@2.0.29 typecheck
+> tsc --noEmit && tsc -p tsconfig.type-tests.json
+
diff --git a/evidence/parallel-llm-fix/typecheck.txt b/evidence/parallel-llm-fix/typecheck.txt
new file mode 100644
index 000000000..302fb33bd
--- /dev/null
+++ b/evidence/parallel-llm-fix/typecheck.txt
@@ -0,0 +1,4 @@
+
+> @relayflows/sdk@2.0.29 typecheck
+> tsc --noEmit && tsc -p tsconfig.type-tests.json
+
diff --git a/evidence/parallel-llm-fix/worktree-merge.txt b/evidence/parallel-llm-fix/worktree-merge.txt
new file mode 100644
index 000000000..7993e7d08
--- /dev/null
+++ b/evidence/parallel-llm-fix/worktree-merge.txt
@@ -0,0 +1 @@
+fatal: refusing to merge unrelated histories
diff --git a/evidence/parallel-llm-fix/worktree.txt b/evidence/parallel-llm-fix/worktree.txt
new file mode 100644
index 000000000..46a66689d
--- /dev/null
+++ b/evidence/parallel-llm-fix/worktree.txt
@@ -0,0 +1,2 @@
+Preparing worktree (detached HEAD 735f2e0)
+HEAD is now at 735f2e0 fix(examples): address prompt-lab review — store lock, menus, visit type, commit candidates
diff --git a/evidence/parallel-llm-lease/README.md b/evidence/parallel-llm-lease/README.md
new file mode 100644
index 000000000..c67f6d115
--- /dev/null
+++ b/evidence/parallel-llm-lease/README.md
@@ -0,0 +1,62 @@
+# Diagnosis evidence: concurrent `f.llm` dispatched after its lease expired
+
+Backs `plan.md` at the repo root. This directory records how the reported
+failure was **reproduced and localized before the fix**, not how the fix was
+verified. Post-fix commands and captured outputs live in
+[../parallel-llm-fix/README.md](../parallel-llm-fix/README.md).
+
+## Harness
+
+`diagnosis-harness.test.ts` is a throwaway vitest file that was run from
+`packages/sdk/tests/` and then removed. It is kept here verbatim so the
+numbers below can be reproduced: copy it back to
+`packages/sdk/tests/zz-scratch-repro.test.ts` and run the commands recorded in
+each transcript.
+
+It builds on the existing `chainFixture` (`packages/sdk/tests/flow-chain-fixture.ts`),
+which starts a real `relayflowd`, attaches a real `AgentWorker` + `LlmWorker`,
+and points `flows.json` at a fake wrapper CLI. The only change is that the fake
+CLI's **preflight probe branches** (`--relayflows-adapter-v1` identification and
+`auth status`) log every invocation with a timestamp, and the `auth` branch
+sleeps `SCRATCH_PROBE_MS`. Step *execution* is untouched and fast.
+
+Preconditions: `cd kernel && sh ../ops/cargo.sh build` and `npm run build` in
+`packages/sdk`.
+
+## What each transcript shows
+
+### `probe-count-capacity1.txt` — the repeated work
+
+`SCRATCH_PROBE_MS=0`, 9 concurrent `f.llm`, capacity 1. The run succeeds, and
+the probe log holds:
+
+    "probeCount": 27,
+    "byKind": { "identify": 18, "auth": 9 }
+
+18 of those 27 are preflight (9 identification + 9 `auth status`); the other 9
+identifications are the steps actually executing. One preflight probe round per
+`f.llm` call, all of it `spawnSync`.
+
+### `lease-expired-capacity1.txt` — the consequence
+
+Same flow, `SCRATCH_PROBE_MS=4000`. The run fails:
+
+    "elapsedMs": 36694,
+    "failure": "step_failed: journal step \"llm-1\" completed with lease_expired"
+
+    Error: Agent lease is already expired for 01M36BDH5G5CAFP4YVVW6BZ2MY/llm-1.
+     ❯ armExpiry src/worker-lease.ts:20:13
+     ❯ Module.withWorkerLease src/worker-lease.ts:47:5
+     ❯ LlmWorker.execute src/llm-worker.ts:60:46
+     ❯ JournalClient.onDispatch src/llm-worker.ts:49:26
+
+Same error string and same stack as the issue report. Elapsed ≈ 9 × 4 s: the
+event loop spent the whole run inside `spawnSync`. `llm-1` — the *first* call,
+the only one that ever held a lease at capacity 1 — is the attempt that expired.
+
+### `lease-expired-capacity4.txt` — capacity is not the variable
+
+Identical run at the default capacity 4: same failure, same step, elapsed
+36801 ms. The probe train runs *before* `WorkerSlots.run` is ever reached, so
+admission capacity cannot bound it. This is why the issue reproduces at
+`--agent-capacity 4` and `1` alike.
diff --git a/evidence/parallel-llm-lease/diagnosis-harness.test.ts b/evidence/parallel-llm-lease/diagnosis-harness.test.ts
new file mode 100644
index 000000000..cf53cdb6a
--- /dev/null
+++ b/evidence/parallel-llm-lease/diagnosis-harness.test.ts
@@ -0,0 +1,94 @@
+import { readFileSync, writeFileSync, existsSync } from 'node:fs';
+import { join, resolve } from 'node:path';
+import { afterEach, describe, expect, it } from 'vitest';
+import { flow } from '@relayflows/surface';
+import { executeAuthoredFlow } from '../src/authored-flow-executor.js';
+import { attachLocalAgent } from '../src/local-agent.js';
+import { JournalClient } from '../src/journal-client.js';
+import { LlmWorker } from '../src/llm-worker.js';
+import { socketPathFor } from '../src/daemon-connection.js';
+import { chainFixture } from './flow-chain-fixture.js';
+
+const closes: Array<() => Promise<void>> = [];
+afterEach(async () => { for (const close of closes.splice(0).reverse()) await close(); });
+
+const PROBE_MS = Number(process.env['SCRATCH_PROBE_MS'] ?? '4000');
+const N = Number(process.env['SCRATCH_N'] ?? '9');
+
+async function fixtureWithSlowProbe(capacity: number) {
+  const fixture = chainFixture();
+  closes.push(() => fixture.close());
+  const probes = join(fixture.root, 'probes.jsonl');
+  writeFileSync(fixture.wrapper, `#!/usr/bin/env node
+import { receiveWrapperRequest } from ${JSON.stringify(resolve('../../testdata/preflight/wrapper-session.mjs'))};
+import { appendFileSync } from 'node:fs';
+if (process.argv[2] === 'auth') {
+  appendFileSync(${JSON.stringify(probes)}, JSON.stringify({ kind: 'auth', at: Date.now() }) + '\\n');
+  await new Promise(r => setTimeout(r, ${PROBE_MS}));
+  process.exit(0);
+}
+if (process.argv[2] === '--relayflows-adapter-v1') {
+  appendFileSync(${JSON.stringify(probes)}, JSON.stringify({ kind: 'identify', at: Date.now() }) + '\\n');
+}
+const request = await receiveWrapperRequest();
+if (request) process.stdout.write('ok');
+`);
+  const client = await fixture.connect();
+  const agent = await attachLocalAgent(client, undefined, undefined, undefined, capacity);
+  closes.push(() => agent.close());
+  const llmClient = new JournalClient(socketPathFor(fixture.data));
+  await llmClient.connect();
+  await llmClient.hello('scratch-llm-worker');
+  closes.push(async () => { llmClient.close(); });
+  const llm = new LlmWorker(llmClient, `${agent.stream}-llm`, capacity);
+  await llm.attach();
+  closes.push(() => llm.close());
+  const readProbes = () => (existsSync(probes) ? readFileSync(probes, 'utf8').trim().split('\n')
+    .filter(Boolean).map(l => JSON.parse(l) as { kind: string; at: number }) : []);
+  return { fixture, client, agent, readProbes };
+}
+
+const many = flow('many-summaries', async f => {
+  await Promise.all(Array.from({ length: N }, (_, i) => f.llm`Summarize ${String(i)}`));
+  f.done('success');
+});
+
+describe('scratch: repeated per-call preflight probes', () => {
+  it(`runs ${N} concurrent f.llm at capacity 1`, async () => {
+    const CAP = Number(process.env['SCRATCH_CAP'] ?? '1');
+    const { fixture, client, agent, readProbes } = await fixtureWithSlowProbe(CAP);
+    let failure: unknown;
+    const started = Date.now();
+    let result: Awaited<ReturnType<typeof executeAuthoredFlow>> | undefined;
+    try {
+      result = await executeAuthoredFlow(many, client, undefined, {
+        flowPath: fixture.flowPath, localAgentStream: agent.stream, workerCapacity: CAP,
+      });
+    } catch (error) { failure = error; }
+    const elapsed = Date.now() - started;
+    const probes = readProbes();
+    // eslint-disable-next-line no-console
+    console.log(JSON.stringify({
+      elapsedMs: elapsed,
+      probeCount: probes.length,
+      byKind: probes.reduce((acc, p) => ({ ...acc, [p.kind]: (acc[p.kind] ?? 0) + 1 }), {} as Record<string, number>),
+      failure: failure instanceof Error ? failure.message : failure,
+      completionReason: result?.completionReason,
+    }, null, 2));
+
+    // Scan every child run journal for a lease_expired attempt.
+    const runIds = (result?.journalSteps ?? []).map(s => s.runId);
+    const expired: string[] = [];
+    for (const runId of runIds) {
+      const { entries } = await client.journalRead(runId, 1, 500);
+      for (const raw of entries as Array<{ entry_type: string; step_id?: string; payload?: { completionReason?: string } }>) {
+        if (raw.entry_type === 'step.completed' && raw.payload?.completionReason === 'lease_expired') {
+          expired.push(`${runId}/${raw.step_id}`);
+        }
+      }
+    }
+    // eslint-disable-next-line no-console
+    console.log('lease_expired attempts: ' + JSON.stringify(expired));
+    expect(true).toBe(true);
+  }, 300_000);
+});
diff --git a/evidence/parallel-llm-lease/lease-expired-capacity1.txt b/evidence/parallel-llm-lease/lease-expired-capacity1.txt
new file mode 100644
index 000000000..20c5d607d
--- /dev/null
+++ b/evidence/parallel-llm-lease/lease-expired-capacity1.txt
@@ -0,0 +1,55 @@
+$ SCRATCH_PROBE_MS=4000 SCRATCH_N=9 SCRATCH_CAP=1 npx vitest run tests/zz-scratch-repro.test.ts
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+stdout | tests/zz-scratch-repro.test.ts > scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1
+{
+  "elapsedMs": 36694,
+  "probeCount": 19,
+  "byKind": {
+    "identify": 10,
+    "auth": 9
+  },
+  "failure": "step_failed: journal step \"llm-1\" completed with lease_expired\nInspect: flows replay 01M36BDH5G5CAFP4YVVW6BZ2MY --at llm-1"
+}
+lease_expired attempts: []
+
+ ✓ tests/zz-scratch-repro.test.ts (1 test) 36750ms
+   ✓ scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1 36749ms
+
+⎯⎯⎯⎯⎯⎯ Unhandled Errors ⎯⎯⎯⎯⎯⎯
+
+Vitest caught 1 unhandled error during the test run.
+This might cause false positive tests. Resolve unhandled errors to make sure your tests are not affected.
+
+⎯⎯⎯⎯ Unhandled Rejection ⎯⎯⎯⎯⎯
+Error: Agent lease is already expired for 01M36BDH5G5CAFP4YVVW6BZ2MY/llm-1.
+ ❯ armExpiry src/worker-lease.ts:20:13
+     18|     const remaining = deadline - Date.now();
+     19|     if (!Number.isFinite(remaining) || remaining <= 0) {
+     20|       throw new Error(`Agent lease is already expired for ${dispatch.r…
+       |             ^
+     21|     }
+     22|     latestDeadline = deadline;
+ ❯ Module.withWorkerLease src/worker-lease.ts:47:5
+ ❯ LlmWorker.execute src/llm-worker.ts:60:46
+ ❯ JournalClient.onDispatch src/llm-worker.ts:49:26
+ ❯ JournalClient.emit node:events:508:20
+ ❯ JournalClient.onLine src/journal-client.ts:168:12
+ ❯ JournalClient.onData src/journal-client.ts:144:33
+ ❯ Socket.<anonymous> src/journal-client.ts:109:43
+ ❯ Socket.emit node:events:508:20
+ ❯ addChunk node:internal/streams/readable:564:12
+
+This error originated in "tests/zz-scratch-repro.test.ts" test file. It doesn't mean the error was thrown inside the file itself, but while it was running.
+The latest test that might've caused the error is "runs 9 concurrent f.llm at capacity 1". It might mean one of the following:
+- The error was thrown, while Vitest was running this test.
+- If the error occurred after the test had been completed, this was the last documented test before it was thrown.
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+
+ Test Files  1 passed (1)
+      Tests  1 passed (1)
+     Errors  1 error
+   Start at  05:21:59
+   Duration  37.90s (transform 545ms, setup 0ms, collect 971ms, tests 36.75s, environment 0ms, prepare 44ms)
+
diff --git a/evidence/parallel-llm-lease/lease-expired-capacity4.txt b/evidence/parallel-llm-lease/lease-expired-capacity4.txt
new file mode 100644
index 000000000..42343a538
--- /dev/null
+++ b/evidence/parallel-llm-lease/lease-expired-capacity4.txt
@@ -0,0 +1,55 @@
+$ SCRATCH_PROBE_MS=4000 SCRATCH_N=9 SCRATCH_CAP=4 npx vitest run tests/zz-scratch-repro.test.ts
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+stdout | tests/zz-scratch-repro.test.ts > scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1
+{
+  "elapsedMs": 36801,
+  "probeCount": 25,
+  "byKind": {
+    "identify": 16,
+    "auth": 9
+  },
+  "failure": "step_failed: journal step \"llm-1\" completed with lease_expired\nInspect: flows replay 01M36BETK0AMZ3DXPJW2MVCV51 --at llm-1"
+}
+lease_expired attempts: []
+
+ ✓ tests/zz-scratch-repro.test.ts (1 test) 36866ms
+   ✓ scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1 36865ms
+
+⎯⎯⎯⎯⎯⎯ Unhandled Errors ⎯⎯⎯⎯⎯⎯
+
+Vitest caught 1 unhandled error during the test run.
+This might cause false positive tests. Resolve unhandled errors to make sure your tests are not affected.
+
+⎯⎯⎯⎯ Unhandled Rejection ⎯⎯⎯⎯⎯
+Error: Agent lease is already expired for 01M36BETK0AMZ3DXPJW2MVCV51/llm-1.
+ ❯ armExpiry src/worker-lease.ts:20:13
+     18|     const remaining = deadline - Date.now();
+     19|     if (!Number.isFinite(remaining) || remaining <= 0) {
+     20|       throw new Error(`Agent lease is already expired for ${dispatch.r…
+       |             ^
+     21|     }
+     22|     latestDeadline = deadline;
+ ❯ Module.withWorkerLease src/worker-lease.ts:47:5
+ ❯ LlmWorker.execute src/llm-worker.ts:60:46
+ ❯ JournalClient.onDispatch src/llm-worker.ts:49:26
+ ❯ JournalClient.emit node:events:508:20
+ ❯ JournalClient.onLine src/journal-client.ts:168:12
+ ❯ JournalClient.onData src/journal-client.ts:144:33
+ ❯ Socket.<anonymous> src/journal-client.ts:109:43
+ ❯ Socket.emit node:events:508:20
+ ❯ addChunk node:internal/streams/readable:564:12
+
+This error originated in "tests/zz-scratch-repro.test.ts" test file. It doesn't mean the error was thrown inside the file itself, but while it was running.
+The latest test that might've caused the error is "runs 9 concurrent f.llm at capacity 1". It might mean one of the following:
+- The error was thrown, while Vitest was running this test.
+- If the error occurred after the test had been completed, this was the last documented test before it was thrown.
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
+
+ Test Files  1 passed (1)
+      Tests  1 passed (1)
+     Errors  1 error
+   Start at  05:22:42
+   Duration  38.06s (transform 576ms, setup 0ms, collect 1.02s, tests 36.87s, environment 0ms, prepare 53ms)
+
diff --git a/evidence/parallel-llm-lease/probe-count-capacity1.txt b/evidence/parallel-llm-lease/probe-count-capacity1.txt
new file mode 100644
index 000000000..b3d5e1170
--- /dev/null
+++ b/evidence/parallel-llm-lease/probe-count-capacity1.txt
@@ -0,0 +1,26 @@
+$ SCRATCH_PROBE_MS=0 SCRATCH_N=9 SCRATCH_CAP=1 npx vitest run tests/zz-scratch-repro.test.ts
+
+ RUN  v2.1.9 /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk
+
+stdout | tests/zz-scratch-repro.test.ts > scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1
+{
+  "elapsedMs": 1182,
+  "probeCount": 27,
+  "byKind": {
+    "identify": 18,
+    "auth": 9
+  },
+  "completionReason": "success"
+}
+
+stdout | tests/zz-scratch-repro.test.ts > scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1
+lease_expired attempts: []
+
+ ✓ tests/zz-scratch-repro.test.ts (1 test) 1253ms
+   ✓ scratch: repeated per-call preflight probes > runs 9 concurrent f.llm at capacity 1 1253ms
+
+ Test Files  1 passed (1)
+      Tests  1 passed (1)
+   Start at  05:21:53
+   Duration  2.40s (transform 547ms, setup 0ms, collect 976ms, tests 1.25s, environment 0ms, prepare 50ms)
+
diff --git a/packages/sdk/src/authored-preflight.ts b/packages/sdk/src/authored-preflight.ts
new file mode 100644
index 000000000..93d3ef43f
--- /dev/null
+++ b/packages/sdk/src/authored-preflight.ts
@@ -0,0 +1,51 @@
+import { dirname, resolve } from 'node:path';
+import { checkAuthoredFlow, readProjectConfig, type ProjectConfig } from './cli/check.js';
+import { probeCliAsync } from './cli/cli-probe.js';
+import { communicationInstruction } from './communication/spec.js';
+import { cliProbeKey, CliProbeError, resolvePreflight, type CliProbeOutcome } from './preflight.js';
+import type { FlowSpec } from './spec.js';
+
+/** Auth/model facts belong to one execution, including concurrent cold callers. */
+export function authoredPreflight(path: string) {
+  const probeCache = new Map<string, CliProbeOutcome>();
+  const pending = new Map<string, Promise<void>>();
+  const directory = dirname(resolve(path));
+  let communicationChecked = false;
+  return async (flow: FlowSpec) => {
+    let config: ProjectConfig;
+    try { config = readProjectConfig(directory); }
+    catch { return checkAuthoredFlow(flow, path); }
+    const resolved = resolvePreflight(flow, {
+      projectCli: config.cli, projectConfigPath: config.path, projectSearchStart: directory,
+      models: config.models, modelRegistryPath: config.modelRegistryPath,
+    });
+    if (!resolved.ok) return checkAuthoredFlow(flow, path, { projectConfig: config });
+    await Promise.all(resolved.resolutions.map(async resolution => {
+      const step = resolved.compiled!.steps.find(step => step.id === resolution.stepId)!;
+      const managed = step.type === 'agent' && communicationInstruction(step.instruction) !== undefined;
+      const key = cliProbeKey(resolution, managed);
+      if (probeCache.has(key)) return;
+      let probing = pending.get(key);
+      if (probing === undefined) {
+        probing = (async () => {
+          try {
+            probeCache.set(key, { result: await probeCliAsync(resolution.cli,
+              resolution.source === 'project' ? config.directory : directory,
+              resolution.model, managed ? 'managed' : undefined) });
+          } catch (error) {
+            probeCache.set(key, { failure: error instanceof CliProbeError ? error.detail : null });
+          }
+        })();
+        pending.set(key, probing);
+      }
+      await probing;
+      pending.delete(key);
+    }));
+    const result = checkAuthoredFlow(flow, path, { projectConfig: config, probeCache, communicationChecked });
+    if (result.report.ok && flow.steps.some(step =>
+      step.type === 'agent' && communicationInstruction(step.instruction) !== undefined)) {
+      communicationChecked = true;
+    }
+    return result;
+  };
+}
diff --git a/packages/sdk/src/authored-worker-step.ts b/packages/sdk/src/authored-worker-step.ts
index a42244ec6..8008bce56 100644
--- a/packages/sdk/src/authored-worker-step.ts
+++ b/packages/sdk/src/authored-worker-step.ts
@@ -3,7 +3,7 @@ import type { AuthoredBudget } from './authored-budget.js';
 import { parseBudget } from './budget.js';
 import type { AgentOptions, AgentResult, LlmOptions, NamedGate } from '@relayflows/surface';
 import { compileSpec, toKernelSpec } from './compile.js';
-import { checkAuthoredFlow } from './cli/check.js';
+import { authoredPreflight } from './authored-preflight.js';
 import { classifyOutcome, type RunLifecycleOptions, type RunReport } from './cli/run.js';
 import type { PreflightDiagnostic } from './preflight.js';
 import { AuthoredFlowExecutionError } from './authored-flow-error.js';
@@ -30,6 +30,7 @@ export function authoredWorkerRunner(
   // slot instead of being admitted and parked for want of a free worker.
   const slots = workerCapacity === undefined ? undefined
     : { agent: new WorkerSlots(workerCapacity), llm: new WorkerSlots(workerCapacity) };
+  const check = authoredPreflight(flowPath);
   const context: AuthoredStepContext = {
     ...(rootRunId === undefined ? {} : { rootRunId }),
     ...(waitOptions.dataDir === undefined ? {} : { dataDir: waitOptions.dataDir }),
@@ -43,7 +44,7 @@ export function authoredWorkerRunner(
     // (cli/check.ts), searching for the nearest flows.json from `flowPath`
     // and real-probing auth/model readiness. An authored agent step gets
     // nothing for free just because it was declared in TS instead of YAML.
-    const { report, flow: resolved } = checkAuthoredFlow(authoring, flowPath);
+    const { report, flow: resolved } = await check(authoring);
     if (!report.ok || resolved === undefined) {
       const refusal = report.diagnostics.find(
         (diagnostic): diagnostic is PreflightDiagnostic & { severity: 'refusal' } =>
diff --git a/packages/sdk/src/cli/bundle-preflight.ts b/packages/sdk/src/cli/bundle-preflight.ts
index bc0e98a36..f10112eb3 100644
--- a/packages/sdk/src/cli/bundle-preflight.ts
+++ b/packages/sdk/src/cli/bundle-preflight.ts
@@ -23,6 +23,6 @@ export async function checkRunnableBundle(directory: string, name: string): Prom
   // This deterministic-only slice declares no project models or executors.
   // Never discover flows.json in the deployment/cache tree or its ancestors.
   return checkAuthoredFlow(flow, join(directory, 'spec.canonical.json'), {
-    directory, models: [], executors: [],
+    projectConfig: { directory, models: [], executors: [] },
   });
 }
diff --git a/packages/sdk/src/cli/check.ts b/packages/sdk/src/cli/check.ts
index 6a91c4107..a4ff7f3a3 100644
--- a/packages/sdk/src/cli/check.ts
+++ b/packages/sdk/src/cli/check.ts
@@ -1,22 +1,12 @@
 import { communicationInstruction } from '../communication/spec.js';
 import { checkCommunicationEnvironment } from '../communication/preflight.js';
-import { agentEnvironment, brokerEnvironment } from '../communication/environment.js';
 import { accessSync, constants, readFileSync } from 'node:fs';
 import { dirname, isAbsolute, join, parse as parsePath, resolve } from 'node:path';
-import { spawnSync } from 'node:child_process';
 import { parse as parseYaml } from 'yaml';
 import { CompileError, compileSpec, kernelToAuthoring } from '../compile.js';
 import { helperReady } from '../yaml-helper-effect.js';
 import { flowRequirements, type FlowRequirements } from '../flow-requirements.js';
-import {
-  adapterIdentification,
-  authenticationProbe,
-  cliAdapterKind,
-  displayInvocation,
-  modelReadinessProbe,
-  type CliInvocation,
-} from '../cli-adapter.js';
-import { MODEL_ENV } from '../worker-cli.js';
+import { probeCli, resolveExecutable } from './cli-probe.js';
 import { modelNameError } from '../model-name.js';
 import type { FlowSpec } from '../spec.js';
 import type { McpServerConfig } from '../spec.js';
@@ -25,9 +15,7 @@ import type { StepGateInspection } from '../gate-contract.js';
 import type { CheckFailureKind, CheckWarningKind } from '../failure-kinds.js';
 import {
   preflight,
-  CliProbeError,
   type CliResolution,
-  type CliProbeResult,
   type PreflightDiagnostic,
   type PreflightProbes,
 } from '../preflight.js';
@@ -163,11 +151,17 @@ function safeRequirements(authoring: FlowSpec, projectCli: string | undefined):
   }
 }
 
+export interface AuthoredCheckOptions {
+  projectConfig?: ProjectConfig;
+  probeCache?: Map<string, import('../preflight.js').CliProbeOutcome>;
+  communicationChecked?: boolean;
+}
+
 /** Preflight a validated authored flow through the same path as YAML/JSON. */
-export function checkAuthoredFlow(authoring: FlowSpec, path: string, projectConfig?: ProjectConfig): CheckExecution {
+export function checkAuthoredFlow(authoring: FlowSpec, path: string, options: AuthoredCheckOptions = {}): CheckExecution {
   const absolutePath = resolve(path);
   try {
-    const config = projectConfig ?? readProjectConfig(dirname(absolutePath));
+    const config = options.projectConfig ?? readProjectConfig(dirname(absolutePath));
     const probes = systemProbes(dirname(absolutePath), config);
     const result = preflight(authoring, {
       projectCli: config.cli,
@@ -176,6 +170,7 @@ export function checkAuthoredFlow(authoring: FlowSpec, path: string, projectConf
       models: config.models,
       ...(config.modelRegistryPath !== undefined ? { modelRegistryPath: config.modelRegistryPath } : {}),
       probes,
+      probeCache: options.probeCache,
     });
     const flow = result.ok
       ? bindResolvedCliPaths(
@@ -186,7 +181,7 @@ export function checkAuthoredFlow(authoring: FlowSpec, path: string, projectConf
         )
       : undefined;
     if (flow?.steps.some(step => step.type === 'agent' && communicationInstruction(step.instruction))) {
-      try { checkCommunicationEnvironment(flow); }
+      try { if (!options.communicationChecked) checkCommunicationEnvironment(flow); }
       catch (error) {
         result.ok = false;
         result.diagnostics.push({ severity: 'refusal', kind: 'probe_failed',
@@ -448,155 +443,6 @@ function canonicalCli(cli: string, directory: string): string {
   return resolve(directory, cli);
 }
 
-function probeCli(
-  cli: string,
-  directory: string,
-  model?: string,
-  execution?: 'managed',
-): CliProbeResult {
-  const executable = resolveExecutable(cli, directory);
-  if (executable === undefined) return { exists: false, authenticated: false };
-  const kind = cliAdapterKind(executable);
-  // Relay owns interactive CLI launch/injection. Its generic PTY path is not
-  // the headless wrapper protocol; do not demand that protocol from Gemini,
-  // Cursor, OpenCode, or other interactive tools. Never invent an auth pass.
-  if (execution === 'managed' && kind === 'relayflows-wrapper-v1') {
-    return { exists: true, supported: true, authenticated: 'unverified' };
-  }
-  const environment = execution === 'managed'
-    ? { ...brokerEnvironment(process.env), ...agentEnvironment(executable) } : process.env;
-  const probe = (invocation: CliInvocation) => runProbe(executable, directory, invocation, environment);
-  const identification = adapterIdentification(kind);
-  const identified = probe(identification.invocation);
-  if (
-    identified.status !== 0
-    || (identification.expectedStdout !== undefined
-      && identified.stdout.trim() !== identification.expectedStdout)
-  ) {
-    return { exists: true, supported: false, authenticated: false };
-  }
-  const auth = authenticationProbe(kind);
-  const authCommand = displayInvocation(cli, auth);
-  if (model === undefined) {
-    return {
-      exists: true,
-      supported: true,
-      authenticated: probe(auth).status === 0,
-      authCommand,
-    };
-  }
-
-  const scoped = modelReadinessProbe(kind, model);
-  const modelCommand = displayInvocation(cli, scoped);
-  // A successful real provider round trip (or identified wrapper probe)
-  // proves both auth and exact-model access. On failure, run the adapter's
-  // actual auth command solely to classify auth vs model access truthfully.
-  if (probe(scoped).status === 0) {
-    return {
-      exists: true,
-      supported: true,
-      authenticated: true,
-      modelAvailable: true,
-      authCommand,
-      modelCommand,
-    };
-  }
-  const authProbe = probe(auth);
-  const authenticated = authProbe.status === 0;
-  return {
-    exists: true,
-    supported: true,
-    authenticated,
-    modelAvailable: false,
-    authCommand,
-    modelCommand,
-    // Only on failure: on success there is nothing to explain, and the output
-    // is the most identity-bearing thing this function touches.
-    ...(authenticated
-      ? {}
-      : {
-        authExitCode: authProbe.status,
-        authFailureDetail: redactProbeOutput(
-          `${authProbe.stderr}${authProbe.stdout}`,
-        ).trim().slice(0, 500),
-      }),
-  };
-}
-
-/**
- * Redact anything that looks like a credential or an account identifier.
- *
- * `auth status` output is diagnostic, but it is also the one place an account
- * email, org id or token fragment can appear. The point of surfacing it is to
- * say WHY a probe failed, which survives redaction; leaking an identity into a
- * refusal message that gets pasted into issues does not.
- */
-function redactProbeOutput(text: string): string {
-  return text
-    .replace(/[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}/g, '<redacted-email>')
-    .replace(/\b(sk|pk|oat|rt)[-_][A-Za-z0-9._-]{8,}/gi, '<redacted-token>')
-    .replace(/\b[A-Fa-f0-9]{32,}\b/g, '<redacted-hex>');
-}
-
-function runProbe(
-  executable: string,
-  directory: string,
-  invocation: CliInvocation,
-  environment: NodeJS.ProcessEnv = process.env,
-): { status: number | null; stdout: string; stderr: string } {
-  const env = { ...environment };
-  delete env[MODEL_ENV];
-  if (invocation.modelEnv !== undefined) env[MODEL_ENV] = invocation.modelEnv;
-  const result = spawnSync(executable, invocation.args, {
-    cwd: directory,
-    encoding: 'utf8',
-    // stderr was 'ignore'. A failing `auth status` writes its reason there, so
-    // discarding it made every authentication refusal structurally
-    // undiagnosable: the refusal could say a probe exited non-zero and never
-    // what it said. Captured, then redacted at the point of use.
-    stdio: ['ignore', 'pipe', 'pipe'],
-    timeout: invocation.timeoutMs,
-    env,
-  });
-  const failure = classifySpawnFailure(result.error, result.signal, invocation.timeoutMs);
-  if (failure !== undefined) throw failure;
-  return {
-    status: result.status,
-    stdout: result.stdout,
-    stderr: result.stderr ?? '',
-  };
-}
-
-function resolveExecutable(command: string, directory: string): string | undefined {
-  if (command.includes('/') || isAbsolute(command)) {
-    const path = isAbsolute(command) ? command : resolve(directory, command);
-    try {
-      accessSync(path, constants.X_OK);
-      return path;
-    } catch {
-      return undefined;
-    }
-  }
-  const result = spawnSync('which', [command], { encoding: 'utf8', timeout: 5_000 });
-  const failure = classifySpawnFailure(result.error, result.signal, 5_000);
-  if (failure !== undefined) throw failure;
-  return result.status === 0 ? result.stdout.trim() : undefined;
-}
-
-function classifySpawnFailure(
-  error: Error | undefined,
-  signal: NodeJS.Signals | null,
-  timeoutMs: number,
-): CliProbeError | undefined {
-  if (error !== undefined) {
-    const detail = (error as NodeJS.ErrnoException).code === 'ETIMEDOUT'
-      ? `timeout:${timeoutMs}ms` as const
-      : 'spawn_failed' as const;
-    return new CliProbeError(detail);
-  }
-  return signal === null ? undefined : new CliProbeError(`signal:${signal}`);
-}
-
 function executableExists(command: string, directory: string): boolean {
   return resolveExecutable(command, directory) !== undefined;
 }
diff --git a/packages/sdk/src/cli/cli-probe.ts b/packages/sdk/src/cli/cli-probe.ts
new file mode 100644
index 000000000..2268f0ad7
--- /dev/null
+++ b/packages/sdk/src/cli/cli-probe.ts
@@ -0,0 +1,189 @@
+import { accessSync, constants } from 'node:fs';
+import { isAbsolute, resolve } from 'node:path';
+import { execFile, spawnSync } from 'node:child_process';
+import { agentEnvironment, brokerEnvironment } from '../communication/environment.js';
+import { adapterIdentification, authenticationProbe, cliAdapterKind, displayInvocation,
+  modelReadinessProbe, type CliInvocation } from '../cli-adapter.js';
+import { MODEL_ENV } from '../worker-cli.js';
+import { CliProbeError, type CliProbeResult } from '../preflight.js';
+
+interface ProbeRequest {
+  executable: string;
+  directory: string;
+  invocation: CliInvocation;
+  environment: NodeJS.ProcessEnv;
+}
+interface ProbeOutput { status: number | null; stdout: string; stderr: string }
+
+/** One decision sequence; checking uses synchronous I/O, live flows yield the loop. */
+export function probeCli(...args: Parameters<typeof probeSequence>): CliProbeResult {
+  return driveSync(probeSequence(...args));
+}
+
+export async function probeCliAsync(...args: Parameters<typeof probeSequence>): Promise<CliProbeResult> {
+  const sequence = probeSequence(...args);
+  let next = sequence.next();
+  while (!next.done) next = sequence.next(await runProbeAsync(next.value));
+  return next.value;
+}
+
+export function resolveExecutable(command: string, directory: string): string | undefined {
+  return driveSync(executableSequence(command, directory));
+}
+
+function driveSync<T>(sequence: Generator<ProbeRequest, T, ProbeOutput>): T {
+  let next = sequence.next();
+  while (!next.done) {
+    const request = next.value;
+    const result = spawnSync(request.executable, request.invocation.args, probeOptions(request));
+    const failure = classifySpawnFailure(result.error, result.signal, request.invocation.timeoutMs);
+    if (failure !== undefined) throw failure;
+    next = sequence.next({ status: result.status, stdout: result.stdout ?? '', stderr: result.stderr ?? '' });
+  }
+  return next.value;
+}
+
+function probeOptions({ directory, invocation, environment }: ProbeRequest) {
+  const env = { ...environment };
+  delete env[MODEL_ENV];
+  if (invocation.modelEnv !== undefined) env[MODEL_ENV] = invocation.modelEnv;
+  return { cwd: directory, encoding: 'utf8' as const, timeout: invocation.timeoutMs,
+    maxBuffer: 1024 * 1024, env };
+}
+
+function runProbeAsync(request: ProbeRequest): Promise<ProbeOutput> {
+  return new Promise((resolve, reject) => {
+    const child = execFile(request.executable, request.invocation.args, probeOptions(request),
+      (error, stdout, stderr) => {
+        // Numeric exit codes are probe results; launch, timeout, signal and buffer
+        // errors are failures to collect a fact, just as in the synchronous driver.
+        if (error?.killed && typeof error.code !== 'string') return reject(new CliProbeError(`timeout:${request.invocation.timeoutMs}ms`));
+        const failure = classifySpawnFailure(
+          error !== null && typeof error.code !== 'number' && error.signal == null ? error : undefined,
+          error?.signal ?? null, request.invocation.timeoutMs);
+        if (failure !== undefined) return reject(failure);
+        resolve({ status: error === null ? 0 : typeof error.code === 'number' ? error.code : null,
+          stdout, stderr });
+      });
+    child.stdin?.end();
+  });
+}
+
+function* probeSequence(
+  cli: string,
+  directory: string,
+  model?: string,
+  execution?: 'managed',
+): Generator<ProbeRequest, CliProbeResult, ProbeOutput> {
+  const executable = yield* executableSequence(cli, directory);
+  if (executable === undefined) return { exists: false, authenticated: false };
+  const kind = cliAdapterKind(executable);
+  // Relay owns interactive CLI launch/injection. Its generic PTY path is not
+  // the headless wrapper protocol; do not demand that protocol from Gemini,
+  // Cursor, OpenCode, or other interactive tools. Never invent an auth pass.
+  if (execution === 'managed' && kind === 'relayflows-wrapper-v1') {
+    return { exists: true, supported: true, authenticated: 'unverified' };
+  }
+  const environment = execution === 'managed'
+    ? { ...brokerEnvironment(process.env), ...agentEnvironment(executable) } : process.env;
+  const probe = (invocation: CliInvocation): ProbeRequest => ({ executable, directory, invocation, environment });
+  const identification = adapterIdentification(kind);
+  const identified = yield probe(identification.invocation);
+  if (
+    identified.status !== 0
+    || (identification.expectedStdout !== undefined
+      && identified.stdout.trim() !== identification.expectedStdout)
+  ) {
+    return { exists: true, supported: false, authenticated: false };
+  }
+  const auth = authenticationProbe(kind);
+  const authCommand = displayInvocation(cli, auth);
+  if (model === undefined) {
+    return {
+      exists: true,
+      supported: true,
+      authenticated: (yield probe(auth)).status === 0,
+      authCommand,
+    };
+  }
+
+  const scoped = modelReadinessProbe(kind, model);
+  const modelCommand = displayInvocation(cli, scoped);
+  // A successful real provider round trip (or identified wrapper probe)
+  // proves both auth and exact-model access. On failure, run the adapter's
+  // actual auth command solely to classify auth vs model access truthfully.
+  if ((yield probe(scoped)).status === 0) {
+    return {
+      exists: true,
+      supported: true,
+      authenticated: true,
+      modelAvailable: true,
+      authCommand,
+      modelCommand,
+    };
+  }
+  const authProbe = yield probe(auth);
+  const authenticated = authProbe.status === 0;
+  return {
+    exists: true,
+    supported: true,
+    authenticated,
+    modelAvailable: false,
+    authCommand,
+    modelCommand,
+    // Only on failure: on success there is nothing to explain, and the output
+    // is the most identity-bearing thing this function touches.
+    ...(authenticated
+      ? {}
+      : {
+        authExitCode: authProbe.status,
+        authFailureDetail: redactProbeOutput(
+          `${authProbe.stderr}${authProbe.stdout}`,
+        ).trim().slice(0, 500),
+      }),
+  };
+}
+
+/**
+ * Redact anything that looks like a credential or an account identifier.
+ *
+ * `auth status` output is diagnostic, but it is also the one place an account
+ * email, org id or token fragment can appear. The point of surfacing it is to
+ * say WHY a probe failed, which survives redaction; leaking an identity into a
+ * refusal message that gets pasted into issues does not.
+ */
+function redactProbeOutput(text: string): string {
+  return text
+    .replace(/[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}/g, '<redacted-email>')
+    .replace(/\b(sk|pk|oat|rt)[-_][A-Za-z0-9._-]{8,}/gi, '<redacted-token>')
+    .replace(/\b[A-Fa-f0-9]{32,}\b/g, '<redacted-hex>');
+}
+
+function* executableSequence(command: string, directory: string): Generator<ProbeRequest, string | undefined, ProbeOutput> {
+  if (command.includes('/') || isAbsolute(command)) {
+    const path = isAbsolute(command) ? command : resolve(directory, command);
+    try {
+      accessSync(path, constants.X_OK);
+      return path;
+    } catch {
+      return undefined;
+    }
+  }
+  const result = yield { executable: 'which', directory: process.cwd(),
+    invocation: { args: [command], timeoutMs: 5_000 }, environment: process.env };
+  return result.status === 0 ? result.stdout.trim() : undefined;
+}
+
+function classifySpawnFailure(
+  error: Error | undefined,
+  signal: NodeJS.Signals | null,
+  timeoutMs: number,
+): CliProbeError | undefined {
+  if (error !== undefined) {
+    const detail = (error as NodeJS.ErrnoException).code === 'ETIMEDOUT'
+      ? `timeout:${timeoutMs}ms` as const
+      : 'spawn_failed' as const;
+    return new CliProbeError(detail);
+  }
+  return signal === null ? undefined : new CliProbeError(`signal:${signal}`);
+}
diff --git a/packages/sdk/src/preflight.ts b/packages/sdk/src/preflight.ts
index 8cc27caf0..9cb717cf6 100644
--- a/packages/sdk/src/preflight.ts
+++ b/packages/sdk/src/preflight.ts
@@ -64,7 +64,7 @@ export class CliProbeError extends Error {
   }
 }
 
-type CliProbeOutcome =
+export type CliProbeOutcome =
   | { result: CliProbeResult }
   | { failure: CliProbeFailureDetail | null };
 
@@ -98,6 +98,7 @@ export interface PreflightOptions {
   /** Exact, project-owned model allowlist from the nearest flows.json. */
   models?: readonly string[];
   modelRegistryPath?: string;
+  probeCache?: Map<string, CliProbeOutcome>;
   probes: PreflightProbes;
 }
 
@@ -193,7 +194,12 @@ async function probeMcp(result: PreflightResult, options: PreflightOptions): Pro
   return { ...result, ok: !result.diagnostics.some(d => d.severity === 'refusal'), mcpTools: Object.freeze(inventory) };
 }
 
-function preflightSync(flow: unknown, options: PreflightOptions): PreflightResult {
+type ResolutionOptions = Omit<PreflightOptions, 'probes' | 'probeCache'>;
+
+/** Resolve static declarations before any CLI, model, command or trigger probe. */
+export function resolvePreflight(flow: unknown, options: ResolutionOptions): PreflightResult & {
+  compiled?: import('./compile.js').CompiledFlowSpec;
+} {
   // Compile before touching any environment fact. `compileSpec` snapshots raw
   // input into inert data, validates it against the closed authoring schema,
   // and lowers `output` sugar into its json_schema gate — so the gate plan
@@ -234,7 +240,6 @@ function preflightSync(flow: unknown, options: PreflightOptions): PreflightResul
   const cliResolutionDiagnostics: PreflightDiagnostic[] = [];
   const resolutions: CliResolution[] = [];
   const resolutionByStep = new Map<string, CliResolution>();
-  const cliProbeResults = new Map<string, CliProbeOutcome>();
 
   // A pure fact about the compiled snapshot, collected before anything that
   // can return early. An unresolved CLI, an unknown model or a bad scope all
@@ -282,6 +287,16 @@ function preflightSync(flow: unknown, options: PreflightOptions): PreflightResul
     return { ok: false, gates: compiled.steps.map(inspectStepGate), resolutions, diagnostics };
   }
 
+  return { ok: true, gates: compiled.steps.map(inspectStepGate), resolutions, diagnostics, compiled };
+}
+
+function preflightSync(flow: unknown, options: PreflightOptions): PreflightResult {
+  const { compiled, ...result } = resolvePreflight(flow, options);
+  if (!result.ok || compiled === undefined) return result;
+  const { diagnostics, resolutions } = result;
+  const resolutionByStep = new Map(resolutions.map(resolution => [resolution.stepId, resolution]));
+  const cliProbeResults = options.probeCache ?? new Map<string, CliProbeOutcome>();
+
   for (const step of compiled.steps) {
     probeNamedGate(step, options.probes, diagnostics);
     warnOnVacuousGate(step, diagnostics);
@@ -326,7 +341,7 @@ function preflightSync(flow: unknown, options: PreflightOptions): PreflightResul
  * fact about the filesystem. Missing manifest means no known mounts — a grant
  * still parses but refuses as `mount_unknown` in that case.
  */
-function scopeDiagnostics(flow: FlowSpec, options: PreflightOptions): PreflightRefusal[] {
+function scopeDiagnostics(flow: FlowSpec, options: ResolutionOptions): PreflightRefusal[] {
   const workspace = flow.workspace;
   const toolsFs = flow.tools?.fs;
   if (workspace === undefined && toolsFs === undefined) return [];
@@ -350,7 +365,7 @@ function scopeDiagnostics(flow: FlowSpec, options: PreflightOptions): PreflightR
 /** Pure authoring validation: no executable, command, trigger, or daemon probe. */
 function unknownModelDiagnostics(
   flow: FlowSpec,
-  options: PreflightOptions,
+  options: ResolutionOptions,
   resolutionByStep: ReadonlyMap<string, CliResolution> = new Map(),
 ): PreflightRefusal[] {
   const diagnostics: PreflightRefusal[] = [];
@@ -464,7 +479,7 @@ function unknownModelMessage(
   return `Step "${stepId}" declares model "${model}"${cliContext}, but it is not listed in ${source}; add the exact model only after verifying that project is allowed to use it.`;
 }
 
-function unresolvedCliMessage(stepId: string, options: PreflightOptions): string {
+function unresolvedCliMessage(stepId: string, options: ResolutionOptions): string {
   const context = options.projectConfigPath !== undefined
     ? ` Nearest project config "${options.projectConfigPath}" declares no cli; outer configs are shadowed.`
     : options.projectSearchStart !== undefined
@@ -494,6 +509,10 @@ function resolveCli(
   return undefined;
 }
 
+export function cliProbeKey(resolution: CliResolution, managed = false): string {
+  return JSON.stringify([resolution.cli, resolution.source, resolution.model ?? null, managed]);
+}
+
 function probeResolvedCli(
   resolution: CliResolution,
   probes: PreflightProbes,
@@ -507,7 +526,7 @@ function probeResolvedCli(
   // Model is part of the key: the same CLI probed with two different models
   // is two different questions, and caching on the CLI alone would let a
   // model that the CLI cannot resolve inherit an earlier model's pass.
-  const cacheKey = JSON.stringify([resolution.cli, resolution.source, resolution.model ?? null, managed]);
+  const cacheKey = cliProbeKey(resolution, managed);
   let outcome = cache.get(cacheKey);
   if (outcome === undefined) {
     try {
diff --git a/packages/sdk/tests/authored-parallel-llm.test.ts b/packages/sdk/tests/authored-parallel-llm.test.ts
new file mode 100644
index 000000000..2e60623c0
--- /dev/null
+++ b/packages/sdk/tests/authored-parallel-llm.test.ts
@@ -0,0 +1,124 @@
+import { readFileSync, readdirSync, writeFileSync } from 'node:fs';
+import { join, resolve } from 'node:path';
+import { afterEach, describe, expect, it } from 'vitest';
+import { flow } from '@relayflows/surface';
+import { executeAuthoredFlow } from '../src/authored-flow-executor.js';
+import { executeDurableAuthoredFlow } from '../src/authored-root.js';
+import { loadAuthoredFlow } from '../src/authored-flow-loader.js';
+import { LlmWorker } from '../src/llm-worker.js';
+import { DEFAULT_LOCAL_AGENT_CAPACITY } from '../src/worker-slots.js';
+import { chainFixture } from './flow-chain-fixture.js';
+
+const closes: Array<() => Promise<void>> = [];
+afterEach(async () => { for (const close of closes.splice(0).reverse()) await close(); });
+
+async function setup(capacity: number, probeDelayMs: number) {
+  const fixture = chainFixture();
+  closes.push(() => fixture.close());
+  const probes = join(fixture.root, 'probes.jsonl');
+  const spans = join(fixture.root, 'spans.jsonl');
+  writeFileSync(probes, '');
+  writeFileSync(spans, '');
+  writeFileSync(join(fixture.root, 'flows.json'),
+    JSON.stringify({ cli: fixture.wrapper, models: ['test-model', 'second-model'] }));
+  writeFileSync(fixture.wrapper, `#!/usr/bin/env node
+import { receiveWrapperRequest } from ${JSON.stringify(resolve('../../testdata/preflight/wrapper-session.mjs'))};
+import { appendFileSync } from 'node:fs';
+if (process.argv[2] === 'auth') {
+  appendFileSync(${JSON.stringify(probes)}, JSON.stringify(process.env.RELAYFLOW_MODEL) + '\\n');
+  await new Promise(done => setTimeout(done, ${probeDelayMs}));
+  process.exit(0);
+}
+const request = await receiveWrapperRequest();
+if (request) {
+  const start = Date.now();
+  await new Promise(done => setTimeout(done, 150));
+  appendFileSync(${JSON.stringify(spans)}, JSON.stringify({ start, end: Date.now() }) + '\\n');
+  process.stdout.write('{"x":1}');
+}
+`);
+  const client = await fixture.connect();
+  const peer = client.createPeer();
+  await peer.connect();
+  await peer.hello('parallel-llm-regression');
+  closes.push(async () => { peer.close(); });
+  const worker = new LlmWorker(peer, 'parallel-llm', capacity);
+  await worker.attach();
+  closes.push(() => worker.close());
+  const lines = (path: string) => readFileSync(path, 'utf8').trim().split('\n').filter(Boolean).map(line => JSON.parse(line));
+  async function assertJournals(expected: number) {
+    const ids = readdirSync(join(fixture.data, 'runs')).filter(name => name.endsWith('.sqlite3'));
+    expect(ids).toHaveLength(expected);
+    const expired: unknown[] = [];
+    for (const name of ids) {
+      const { entries } = await client.journalRead(name.slice(0, -8), 1, 500);
+      expired.push(...entries.filter(entry => entry.entry_type === 'step.completed'
+        && (entry.payload as { completionReason?: string })?.completionReason === 'lease_expired'));
+    }
+    expect(expired).toEqual([]);
+  }
+  return { fixture, client, assertJournals, assertProbes(count: number) {
+    expect(lines(probes)).toHaveLength(count);
+  }, assertCapacity() {
+    const intervals = lines(spans) as Array<{ start: number; end: number }>;
+    expect(intervals).toHaveLength(9);
+    const peak = Math.max(...intervals.map(({ start }) =>
+      intervals.filter(other => other.start <= start && start < other.end).length));
+    expect(peak).toBeLessThanOrEqual(capacity);
+    expect(peak).toBeGreaterThan(0);
+  } };
+}
+
+const output = { type: 'object', properties: { x: { type: 'number' } }, required: ['x'], additionalProperties: false };
+const many = flow('parallel-llm', async f => {
+  await Promise.all(Array.from({ length: 9 }, (_, i) =>
+    f.llm(`Return {"x": ${i}+1} as JSON only.`, { output, model: 'test-model' })));
+  f.done('success');
+});
+
+describe.each([1, DEFAULT_LOCAL_AGENT_CAPACITY])('parallel llm capacity %i', capacity => {
+  it('deduplicates concurrent preflight probes', async () => {
+    const test = await setup(capacity, 0);
+    const result = await executeAuthoredFlow(many, test.client, undefined, {
+      flowPath: test.fixture.flowPath, workerCapacity: capacity,
+    });
+    expect(result.completionReason).toBe('success');
+    test.assertProbes(1);
+    test.assertCapacity();
+    await test.assertJournals(10);
+  }, 120_000);
+
+  it('completes nine calls without expired child leases during slow preflight', async () => {
+    // Before the fix, eight remaining probes block dispatch for 48 s (>30 s).
+    const test = await setup(capacity, 6_000);
+    const result = await executeAuthoredFlow(many, test.client, undefined, {
+      flowPath: test.fixture.flowPath, workerCapacity: capacity,
+    });
+    await test.assertJournals(10);
+    expect(result.completionReason).toBe('success');
+    test.assertProbes(1);
+    test.assertCapacity();
+  }, 120_000);
+
+  it('keeps the durable root lease alive across two cold models', async () => {
+    // A single cold probe exceeds the root's 30 s lease. Cache alone cannot
+    // rescue this; the process must keep handling heartbeats while probing.
+    const test = await setup(capacity, 45_000);
+    writeFileSync(test.fixture.flowPath, `
+import { flow } from '@relayflows/surface';
+export default flow('two-models', async f => {
+  await Promise.all(Array.from({ length: 9 }, (_, i) =>
+    f.llm('Return {"x":1}', { output: ${JSON.stringify(output)}, model: i % 2 ? 'second-model' : 'test-model' })));
+  f.done('success');
+});
+`);
+    const loaded = await loadAuthoredFlow(test.fixture.flowPath);
+    const result = await executeDurableAuthoredFlow(loaded, test.client, undefined, {
+      dataDir: test.fixture.data, workerCapacity: capacity,
+    });
+    await test.assertJournals(11);
+    expect(result.completionReason).toBe('success');
+    test.assertProbes(2);
+    test.assertCapacity();
+  }, 120_000);
+});
diff --git a/packages/sdk/tests/authored-preflight.test.ts b/packages/sdk/tests/authored-preflight.test.ts
new file mode 100644
index 000000000..a577999f7
--- /dev/null
+++ b/packages/sdk/tests/authored-preflight.test.ts
@@ -0,0 +1,54 @@
+import { chmodSync, existsSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { afterEach, expect, it } from 'vitest';
+import { authoredPreflight } from '../src/authored-preflight.js';
+import { SPEC_SCHEMA_VERSION, type FlowSpec } from '../src/spec.js';
+
+const directories: string[] = [];
+afterEach(() => { for (const directory of directories.splice(0)) rmSync(directory, { recursive: true, force: true }); });
+function setup() {
+  const directory = mkdtempSync(join(tmpdir(), 'authored-preflight-'));
+  directories.push(directory);
+  const calls = join(directory, 'calls');
+  const cli = join(directory, 'wrapper');
+  writeFileSync(cli, `#!/usr/bin/env node
+import { appendFileSync } from 'node:fs';
+appendFileSync(${JSON.stringify(calls)}, 'probe\\n');
+if (process.argv[2] === '--relayflows-adapter-v1') console.log('relayflows-agent-cli-v1');
+else process.exit(1);
+`);
+  chmodSync(cli, 0o755);
+  writeFileSync(join(directory, 'flows.json'), JSON.stringify({ cli, models: ['allowed'] }));
+  return { calls, check: authoredPreflight(join(directory, 'test.flow.ts')) };
+}
+function spec(id: string, model = 'allowed'): FlowSpec {
+  return { version: SPEC_SCHEMA_VERSION, name: 'test', steps: [{ id, type: 'llm', prompt: 'hello', model }] };
+}
+
+it('refuses unknown models before launching any provider probe', async () => {
+  const { check, calls } = setup();
+  const result = await check(spec('one', 'forbidden'));
+  expect(result.report.diagnostics).toContainEqual(expect.objectContaining({ kind: 'model_unknown' }));
+  expect(existsSync(calls)).toBe(false);
+});
+
+it('refuses malformed specs before launching any provider probe', async () => {
+  const { check, calls } = setup();
+  const invalid = spec('one');
+  (invalid.steps[0] as { prompt: unknown }).prompt = 42;
+  expect((await check(invalid)).report.ok).toBe(false);
+  expect(existsSync(calls)).toBe(false);
+});
+
+it('shares failed facts across callers but retains each step identity', async () => {
+  const { check, calls } = setup();
+  const results = await Promise.all(['one', 'two'].map(id => check(spec(id))));
+  for (const [index, result] of results.entries()) {
+    expect(result.report.diagnostics).toContainEqual(expect.objectContaining({
+      kind: 'cli_unauthenticated', stepId: index === 0 ? 'one' : 'two',
+    }));
+  }
+  // One identification, one exact-model probe, one auth classification.
+  expect(readFileSync(calls, 'utf8').trim().split('\n')).toHaveLength(3);
+});
diff --git a/packages/sdk/tests/cli-probe.test.ts b/packages/sdk/tests/cli-probe.test.ts
new file mode 100644
index 000000000..3d8934241
--- /dev/null
+++ b/packages/sdk/tests/cli-probe.test.ts
@@ -0,0 +1,62 @@
+import { chmodSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { afterEach, expect, it, vi } from 'vitest';
+import { probeCli, probeCliAsync } from '../src/cli/cli-probe.js';
+import * as adapters from '../src/cli-adapter.js';
+
+const directories: string[] = [];
+afterEach(() => {
+  vi.restoreAllMocks();
+  for (const directory of directories.splice(0)) rmSync(directory, { recursive: true, force: true });
+});
+function wrapper(body: string) {
+  const directory = mkdtempSync(join(tmpdir(), 'probe-drivers-'));
+  directories.push(directory);
+  const path = join(directory, 'wrapper');
+  writeFileSync(path, '#!/usr/bin/env node\n' + body);
+  chmodSync(path, 0o755);
+  return { path, directory };
+}
+const identify = `if (process.argv[2] === '--relayflows-adapter-v1') {
+  console.log('relayflows-agent-cli-v1'); process.exit(0);
+}`;
+
+it.each([
+  ['success', 'process.exit(0)', { authenticated: true, modelAvailable: true }],
+  ['model denied', "process.exit(process.env.RELAYFLOW_MODEL ? 1 : 0)", { authenticated: true, modelAvailable: false }],
+  ['auth denied', "console.error('person@example.com sk-123456789abcdef'); process.exit(1)",
+    { authenticated: false, modelAvailable: false, authExitCode: 1,
+      authFailureDetail: '<redacted-email> <redacted-token>' }],
+] as const)('keeps synchronous and asynchronous classification equal: %s', async (_name, body, expected) => {
+  const { path, directory } = wrapper(identify + body);
+  const sync = probeCli(path, directory, 'test-model');
+  expect(sync).toMatchObject(expected);
+  expect(await probeCliAsync(path, directory, 'test-model')).toEqual(sync);
+});
+
+it('keeps missing executables and unsupported identification fail closed', async () => {
+  const { path, directory } = wrapper("console.log('not a wrapper')");
+  expect(await probeCliAsync(path, directory)).toEqual(probeCli(path, directory));
+  expect(await probeCliAsync('./missing', directory)).toEqual({ exists: false, authenticated: false });
+  expect(await probeCliAsync('relayflows-no-such-executable', directory)).toEqual({ exists: false, authenticated: false });
+});
+
+it('reports timeout in both drivers while the async driver leaves the loop free', async () => {
+  const { path, directory } = wrapper(identify + 'setTimeout(() => {}, 10_000);');
+  const original = adapters.modelReadinessProbe;
+  vi.spyOn(adapters, 'modelReadinessProbe').mockImplementation((kind, model) =>
+    ({ ...original(kind, model), timeoutMs: 100 }));
+  expect(() => probeCli(path, directory, 'test-model')).toThrow(expect.objectContaining({ detail: 'timeout:100ms' }));
+  let ticked = false;
+  const timer = setTimeout(() => { ticked = true; }, 20);
+  await expect(probeCliAsync(path, directory, 'test-model')).rejects.toMatchObject({ detail: 'timeout:100ms' });
+  clearTimeout(timer);
+  expect(ticked).toBe(true);
+});
+
+it('reports signal termination in both drivers', async () => {
+  const { path, directory } = wrapper(identify + "process.kill(process.pid, 'SIGTERM');");
+  expect(() => probeCli(path, directory, 'test-model')).toThrow(expect.objectContaining({ detail: 'signal:SIGTERM' }));
+  await expect(probeCliAsync(path, directory, 'test-model')).rejects.toMatchObject({ detail: 'signal:SIGTERM' });
+});
diff --git a/packages/sdk/tests/communication-preflight.test.ts b/packages/sdk/tests/communication-preflight.test.ts
index ddb9cdb8f..7d8507b63 100644
--- a/packages/sdk/tests/communication-preflight.test.ts
+++ b/packages/sdk/tests/communication-preflight.test.ts
@@ -16,7 +16,7 @@ function fixture(cli: string) {
   const spec: FlowSpec = { version: '0.1.0', cli: executable, steps: [
     { id: 'a', type: 'agent', instruction: 'send' }, { id: 'b', type: 'agent', instruction: 'receive' },
   ], communication: { links: [{ from: 'a', to: 'b' }] } };
-  return { spec, check: (flow = spec) => checkAuthoredFlow(flow, join(root, 'flow.yaml'), config) };
+  return { spec, check: (flow = spec) => checkAuthoredFlow(flow, join(root, 'flow.yaml'), { projectConfig: config }) };
 }
 describe('managed CLI preflight', () => {
   it('probes known CLIs with the same filtered credentials as execution', () => {