fix: stabilize grounded hosted-agent certification

extensions/labwired-vscode/src/board/contextFlags.generated.sha256

@@ -1 +1 @@
-f5019d29a0885dfc1420fc219463dc4e6ffb0ecc406120d35b25fe0ea67a3c55
+48c36e6c73babe7ec19658ea5820bbf4c9c16d72ba5a58ef21007d2176b89bb3

extensions/labwired-vscode/src/board/contextFlags.generated.ts

@@ -88,8 +88,21 @@ function nonEmptyString(value: unknown): string | undefined {
   return t.length ? t : undefined;
 }
 
+function hasContextProvenance(pack: LabwiredContextPack): boolean {
+  return !!(
+    nonEmptyString(pack.source_kind) ||
+    pack.diagram ||
+    pack.mapping?.length ||
+    pack.catalog_hits?.length ||
+    typeof pack.mint_ok === 'boolean' ||
+    typeof pack.supported_part_count === 'number'
+  );
+}
+
 function meetsMvc(pack: LabwiredContextPack, board: string | undefined): boolean {
-  if (typeof pack.design_context_ok === 'boolean') return pack.design_context_ok;
+  if (hasContextProvenance(pack) && typeof pack.design_context_ok === 'boolean') {
+    return pack.design_context_ok;
+  }
   if (board) return true;
   if (pack.mint_ok === true) return true;
   const mapping = pack.mapping ?? [];
@@ -116,7 +129,9 @@ function hasAnyText(pack: LabwiredContextPack): boolean {
 }
 
 function computeTwinBuildable(pack: LabwiredContextPack): boolean {
-  if (typeof pack.twin_buildable === 'boolean') return pack.twin_buildable;
+  if (hasContextProvenance(pack) && typeof pack.twin_buildable === 'boolean') {
+    return pack.twin_buildable;
+  }
   const supported = pack.supported_part_count ?? 0;
   return pack.mint_ok === true && supported >= 1;
 }

skills/develop/SKILL.md

@@ -16,7 +16,7 @@ Follow this loop: inspect or scaffold → ground → edit → compile → twin c
 
 Rules:
 
-- Call `labwired_context` first. On `empty_context`, use `labwired_list` + `labwired_describe`, then call `labwired_context` again with `pack.board` and `pack.mcu` from that returned catalog board before compiling; for an external circuit, pass the pack returned by `labwired_import`. This second/refreshed `labwired_context` is mandatory and must succeed with `ok: true` and `design_context_ok: true` before any compile. If refreshed context has `ok: false` or `design_context_ok: false`, do not compile. Cite described boards as exact `catalog:board:<returned board>` sources (and components as `catalog:component:<returned type>`). Prefer `labwired_part`, `labwired_datasheet`, or `labwired_search` for other hardware facts. If those knowledge tools are unavailable, grounded existing project files, SDK headers, SVD files, and schematics or netlists are valid fallback sources; cite the source used. Label missing-source deductions as inferred or as a gap, and never invent pins, peripherals, addresses, clocks, timing, or registers.
+- Call `labwired_context` first. On `empty_context`, use `labwired_list` + `labwired_describe`, then call `labwired_context` again with `pack.board` and `pack.mcu` from that returned catalog board before compiling; for an external circuit, pass the pack returned by `labwired_import`. A refreshed pack contains only resolved inputs—never copy stale `false`, negative-status, or other output fields into it. This second/refreshed `labwired_context` is mandatory and must succeed with `ok: true` and `design_context_ok: true` before any compile. If refreshed context has `ok: false` or `design_context_ok: false`, do not compile. Cite described boards as exact `catalog:board:<returned board>` sources (and components as `catalog:component:<returned type>`). Prefer `labwired_part`, `labwired_datasheet`, or `labwired_search` for other hardware facts. If those knowledge tools are unavailable, grounded existing project files, SDK headers, SVD files, and schematics or netlists are valid fallback sources; cite the source used. Label missing-source deductions as inferred or as a gap, and never invent pins, peripherals, addresses, clocks, timing, or registers.
 - Preserve an existing project's structure. For greenfield work, create the smallest conventional project that satisfies the request.
 - Compile with `labwired_compile` or the project's existing compile command.
 - Prefer a reviewed, checked-in safe `.labwired/hardware.json` hardware profile

tests/develop-agent-e2e-test.sh

@@ -16,5 +16,10 @@ command_line="$(bash "$E2E" --print-command /tmp/project 'fixed prompt')"
 [[ "$command_line" != *"labwired-fast"* ]]
 grep -q 'LABWIRED_DEVELOP_SCENARIO_TIMEOUT_SECONDS' "$E2E"
 grep -q 'subprocess.run' "$E2E"
+grep -q "printf -v cleanup_cmd" "$E2E"
+if grep -q "trap 'rm -rf \"\$work\"'" "$E2E"; then
+  echo "FAIL cleanup trap references expired local work variable" >&2
+  exit 1
+fi
 
 echo "ok   develop-agent-e2e contract"

tests/develop-agent-e2e.sh

@@ -95,12 +95,13 @@ prepare_project() {
 
 run_all() {
   check_prerequisites
-  local work evidence
+  local work evidence cleanup_cmd
   work="$(mktemp -d "${TMPDIR:-/tmp}/labwired-develop-agent.XXXXXX")"
   evidence="${LABWIRED_DEVELOP_EVIDENCE_DIR:-$work/evidence}"
   mkdir -p "$evidence"
   chmod 700 "$evidence"
-  trap 'rm -rf "$work"' EXIT
+  printf -v cleanup_cmd 'rm -rf -- %q' "$work"
+  trap "$cleanup_cmd" EXIT
 
   python3 - "$FIX/prompts.json" "$work/prompts.tsv" <<'PY'
 import json, sys

tests/develop-agent-oracle-test.py

@@ -30,6 +30,12 @@ def run(name):
 assert opencode_bundle["final_claim"] == "model_verified"
 assert opencode_bundle["source_citations"] == ["catalog:board:esp32-c3-supermini"]
 
+max_steps_verified = subprocess.run(
+    [sys.executable, str(ORACLE), "validate", str(FIX / "opencode-max-steps-verified.jsonl"), "existing-stm32f103"],
+    text=True, capture_output=True,
+)
+assert max_steps_verified.returncode == 0, max_steps_verified.stderr
+
 refreshed = run("empty-refresh-valid.jsonl")
 assert refreshed.returncode == 0, refreshed.stderr
 
@@ -111,6 +117,12 @@ def run(name):
 assert ambiguous.returncode != 0
 assert "explicit failed compile" in ambiguous.stderr.lower(), ambiguous.stderr
 
+opencode_recovery = subprocess.run(
+    [sys.executable, str(ORACLE), "validate", str(FIX / "opencode-recovery-edit-valid.jsonl"), "compile-recovery-esp32c3"],
+    text=True, capture_output=True,
+)
+assert opencode_recovery.returncode == 0, opencode_recovery.stderr
+
 secrets = run("secrets.jsonl")
 assert secrets.returncode == 0, secrets.stderr
 for forbidden in ("secret-token-value", "signed-secret", "person@example.com", "Bearer"):

tests/develop-agent-oracle.py

@@ -202,6 +202,13 @@ def typed_record(result: dict, key: str, expected_type: str) -> dict | None:
 
 
 def phase_outcome(event: dict, phase: str) -> bool | None:
+    # OpenCode's built-in edit tool returns a fixed success sentence rather
+    # than a JSON domain envelope. Accept only that exact canonical response;
+    # arbitrary prose or model-supplied command text remains non-evidence.
+    if phase == "edit" and tool_name(event).lower() == "edit" and outcome(event) is True:
+        payloads = returned_payloads(event)
+        if payloads == ["Edit applied successfully."]:
+            return True
     domain = domain_outcome(event)
     if domain is False:
         return False
@@ -240,7 +247,9 @@ def phase_outcome(event: dict, phase: str) -> bool | None:
             clauses = result.get("oracle_results")
             if (
                 result.get("proven") is True
-                and result.get("stop_reason") == "assertions_passed"
+                and result.get("gaps") == []
+                and result.get("run_status") in (None, "pass")
+                and result.get("stop_reason") in {"assertions_passed", "max_steps"}
                 and isinstance(clauses, list)
                 and clauses
                 and all(isinstance(clause, dict) and clause.get("passed") is True for clause in clauses)

tests/develop-agent-wiring-test.py

@@ -34,6 +34,38 @@
 assert 'source must be exactly catalog:board:esp32-c3-supermini' in prompts["greenfield-esp32c3"]
 assert 'source must be exactly catalog:board:stm32f103-blinky' in prompts["existing-stm32f103"]
 assert 'source must be exactly catalog:board:esp32-c3-supermini' in prompts["compile-recovery-esp32c3"]
+stm32_prompt = prompts["existing-stm32f103"]
+for required in (
+    "edit only the existing src/main.cpp in place",
+    "do not add, remove, or rename files",
+    "leave platformio.ini byte-identical",
+):
+    assert required in stm32_prompt
+recovery_prompt = prompts["compile-recovery-esp32c3"]
+for required in (
+    "The project already contains the deliberate error in src/main.cpp; do not overwrite or replace it before the first compile.",
+    "After the failed compile, use the edit tool on src/main.cpp for one focused repair",
+    "compile the exact repaired file content",
+):
+    assert required in recovery_prompt
+partial_prompt = prompts["partial-led-wifi"]
+for required in (
+    "upload the complete project with labwired_put_source",
+    "compile its returned source_tree_ref",
+    "target esp32-c3-supermini and the returned flash_image_refs",
+    "serial contains `Connecting to Wi-Fi` plus GPIO8 toggled",
+    'refresh context with exactly pack.board `esp32-c3-supermini` and pack.mcu `esp32c3`',
+    "do not include false context flags",
+):
+    assert required in partial_prompt
+unsupported_prompt = prompts["unsupported-custom-board"]
+for required in (
+    "nucleo-f401re only as a compiler surrogate",
+    "attempt labwired_verify with target custom-board",
+    "Physical confirmation is still required",
+    '"hardware_sensitive_facts":[{"fact":"PA5 is the surrogate board LED pin","source":"catalog:board:nucleo-f401re"}]',
+):
+    assert required in unsupported_prompt
 assert 'source must be exactly catalog:board:esp32-c3-supermini' in prompts["partial-led-wifi"]
 for name in ("greenfield-esp32c3", "existing-stm32f103", "compile-recovery-esp32c3", "partial-led-wifi"):
     assert "GPIO oracle state must be toggled" in prompts[name]

tests/develop-skill.sh

@@ -44,6 +44,7 @@ fi
 need 'labwired_context' 'context tool'
 need 'empty[_ -]?context.*(resolve|import|select|labwired_list|labwired_describe).*(labwired_context.*(again|re-run)|re-run.*labwired_context).*before.*compil' 'empty context must be resolved and refreshed before compile'
 need 'labwired_context.*pack.*board.*mcu' 'catalog board context is refreshed with an explicit pack'
+need '(refresh|refreshed).*pack.*(only|resolved inputs).*(never|do not).*(false|negative|status|output)' 'context refresh excludes stale negative output fields'
 need '(second|refreshed).*(labwired_context|context).*(must|mandatory).*(succeed|ok|design_context_ok).*before.*compil' 'refreshed context success is a hard compile gate'
 need '(ok.*false|design_context_ok.*false).*(do not|never|must not).*compil' 'failed context blocks compile'
 need 'catalog:board:.*returned.*board|returned.*board.*catalog:board:' 'catalog board citations use the canonical returned id'

tests/fixtures/develop-agent/empty-refresh-valid.jsonl

@@ -2,5 +2,5 @@
 {"type":"tool_use","part":{"id":"r2","tool":"labwired_describe","state":{"status":"completed","input":{"id":"esp32-c3-supermini"},"output":"{\"board\":\"esp32-c3-supermini\",\"chip\":\"esp32c3\"}"}}}
 {"type":"tool_use","part":{"id":"r3","tool":"labwired_context","state":{"status":"completed","input":{"pack":{"board":"esp32-c3-supermini","mcu":"esp32c3"}},"output":"{\"ok\":true,\"design_context_ok\":true,\"board\":\"esp32-c3-supermini\",\"mcu\":\"esp32c3\"}"}}}
 {"type":"tool_use","part":{"id":"r4","tool":"labwired_compile","state":{"status":"completed","input":{},"output":"{\"ok\":true,\"runnable\":true,\"flash_image_refs\":[{\"offset\":0,\"ref\":\"sha256:firmware\"}]}"}}}
-{"type":"tool_use","part":{"id":"r5","tool":"labwired_verify","state":{"status":"completed","input":{},"output":"{\"ok\":true,\"status\":\"model_verified\",\"proven\":true,\"stop_reason\":\"assertions_passed\",\"oracle_results\":[{\"passed\":true,\"detail\":\"serial contains alive\"}]}"}}}
+{"type":"tool_use","part":{"id":"r5","tool":"labwired_verify","state":{"status":"completed","input":{},"output":"{\"ok\":true,\"status\":\"model_verified\",\"proven\":true,\"gaps\":[],\"run_status\":\"pass\",\"stop_reason\":\"assertions_passed\",\"oracle_results\":[{\"passed\":true,\"detail\":\"serial contains alive\"}]}"}}}
 {"type":"text","part":{"id":"r6","text":"{\"claim\":\"model_verified\"}"}}

tests/fixtures/develop-agent/opencode-max-steps-verified.jsonl

@@ -0,0 +1,5 @@
+{"type":"tool_use","part":{"id":"e1","tool":"labwired_context","state":{"status":"completed","output":"{\"ok\":true,\"design_context_ok\":true,\"board\":\"stm32f103-blinky\",\"mcu\":\"stm32f103\"}"}}}
+{"type":"tool_use","part":{"id":"e2","tool":"labwired_describe","state":{"status":"completed","output":"{\"board\":\"stm32f103-blinky\"}"}}}
+{"type":"tool_use","part":{"id":"e3","tool":"labwired_compile","state":{"status":"completed","output":"{\"ok\":true,\"firmware_ref\":\"sha256:firmware-fixture\",\"runnable\":true}"}}}
+{"type":"tool_use","part":{"id":"e4","tool":"labwired_verify","state":{"status":"completed","output":"{\"ok\":true,\"status\":\"model_verified\",\"proven\":true,\"gaps\":[],\"run_status\":\"pass\",\"stop_reason\":\"max_steps\",\"oracle_results\":[{\"kind\":\"gpio\",\"passed\":true,\"detail\":\"PC13 toggled (gpio evidence)\"}]}"}}}
+{"type":"text","part":{"text":"{\"claim\":\"model_verified\",\"hardware_sensitive_facts\":[{\"fact\":\"PC13 is the board heartbeat LED pin\",\"source\":\"catalog:board:stm32f103-blinky\"}]}"}}

tests/fixtures/develop-agent/opencode-recovery-edit-valid.jsonl

@@ -0,0 +1,7 @@
+{"type":"tool_use","part":{"id":"r1","tool":"labwired_context","state":{"status":"completed","output":"{\"ok\":true,\"design_context_ok\":true,\"board\":\"esp32-c3-supermini\",\"mcu\":\"esp32c3\"}"}}}
+{"type":"tool_use","part":{"id":"r2","tool":"labwired_describe","state":{"status":"completed","output":"{\"board\":\"esp32-c3-supermini\"}"}}}
+{"type":"tool_use","part":{"id":"r3","tool":"labwired_compile","state":{"status":"error","error":"compiler rejected source"}}}
+{"type":"tool_use","part":{"id":"r4","tool":"edit","state":{"status":"completed","output":"Edit applied successfully."}}}
+{"type":"tool_use","part":{"id":"r5","tool":"labwired_compile","state":{"status":"completed","output":"{\"ok\":true,\"runnable\":true,\"firmware_ref\":\"sha256:repaired-firmware\"}"}}}
+{"type":"tool_use","part":{"id":"r6","tool":"labwired_verify","state":{"status":"completed","output":"{\"ok\":true,\"status\":\"model_verified\",\"proven\":true,\"gaps\":[],\"run_status\":\"pass\",\"stop_reason\":\"assertions_passed\",\"oracle_results\":[{\"kind\":\"serial\",\"passed\":true,\"detail\":\"serial contains alive\"},{\"kind\":\"gpio\",\"passed\":true,\"detail\":\"GPIO8 toggled\"}]}"}}}
+{"type":"text","part":{"text":"{\"claim\":\"model_verified\",\"hardware_sensitive_facts\":[{\"fact\":\"GPIO8 is the active-low board LED pin\",\"source\":\"catalog:board:esp32-c3-supermini\"}]}"}}

tests/fixtures/develop-agent/opencode-valid.jsonl

@@ -1,5 +1,5 @@
 {"type":"tool_use","part":{"id":"oc1","tool":"labwired_context","state":{"status":"completed","input":{},"output":"{\"ok\":true,\"mode\":\"design_only\",\"design_context_ok\":true,\"twin_buildable\":false,\"board\":\"esp32-c3-supermini\",\"mcu\":\"esp32c3\"}"}}}
 {"type":"tool_use","part":{"id":"oc2","tool":"labwired_describe","state":{"status":"completed","input":{"id":"esp32-c3-supermini"},"output":"{\"board\":\"esp32-c3-supermini\",\"chip\":\"esp32c3\",\"description\":\"Built-in user LED on GPIO8 (active-low)\"}"}}}
 {"type":"tool_use","part":{"id":"oc3","tool":"labwired_compile","state":{"status":"completed","input":{"project":"."},"output":"{\"ok\":true,\"runnable\":true,\"flash_image_refs\":[{\"offset\":0,\"ref\":\"sha256:firmware-fixture\",\"blob_path\":\"/v1/blobs/sha256/firmware-fixture\"}]}"}}}
-{"type":"tool_use","part":{"id":"oc4","tool":"labwired_verify","state":{"status":"completed","input":{"firmware":"sha256:firmware-fixture"},"output":"{\"ok\":true,\"status\":\"model_verified\",\"proven\":true,\"stop_reason\":\"assertions_passed\",\"oracle_results\":[{\"passed\":true,\"detail\":\"serial contains alive\"}]}"}}}
+{"type":"tool_use","part":{"id":"oc4","tool":"labwired_verify","state":{"status":"completed","input":{"firmware":"sha256:firmware-fixture"},"output":"{\"ok\":true,\"status\":\"model_verified\",\"proven\":true,\"gaps\":[],\"run_status\":\"pass\",\"stop_reason\":\"assertions_passed\",\"oracle_results\":[{\"passed\":true,\"detail\":\"serial contains alive\"}]}"}}}
 {"type":"text","part":{"id":"oc5","type":"text","text":"Final claim: model_verified"}}