KeyffMS · KeyffMS · Aug 28, 2026 · Aug 28, 2026 · Aug 28, 2026
diff --git a/.agents/skills/plan-anvil/scripts/execution_contract.py b/.agents/skills/plan-anvil/scripts/execution_contract.py
@@ -39,10 +39,7 @@ def execution_contract_findings(plan_text: str) -> list[dict[str, Any]]:
                 }
             )
 
-    if not (
-        _has(plan_text, r"\bflat\b.{0,40}\bdirect[- ]child\b")
-        or _has(plan_text, r"\bagents\.max_depth\s*=\s*1\b")
-    ):
+    if not _has(plan_text, r"\bflat\b.{0,40}\bdirect[- ]child\b"):
         findings.append(
             {
                 "kind": "execution-contract-topology-missing",

diff --git a/.agents/skills/plan-anvil/tests/test_execution_contract.py b/.agents/skills/plan-anvil/tests/test_execution_contract.py
@@ -85,6 +85,11 @@ def test_flat_topology_and_single_modifier_are_enforced(self) -> None:
         self.assertIn("execution-contract-topology-missing", kinds)
         self.assertIn("execution-contract-single-modifier-missing", kinds)
 
+    def test_legacy_max_depth_does_not_satisfy_flat_topology(self) -> None:
+        text = compliant_plan().replace("flat direct-child topology", "agents.max_depth = 1 topology")
+        kinds = {item["kind"] for item in execution_contract_findings(text)}
+        self.assertIn("execution-contract-topology-missing", kinds)
+
     def test_evidence_cycle_is_enforced(self) -> None:
         findings = execution_contract_findings(compliant_plan().replace("FULL GREEN", "TEST AGAIN"))
         self.assertIn(

diff --git a/.codex/config.toml b/.codex/config.toml
@@ -1,3 +1,3 @@
 [agents]
-max_threads = 4
-max_depth = 1
+enabled = true
+max_concurrent_threads_per_session = 4
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -4,6 +4,13 @@ All notable changes to PlanAnvil are documented here.
 
 ## [Unreleased]
 
+### Changed
+
+- refresh the Codex capability baseline to 2.2 against current 2026-08-28 official documentation;
+- replace legacy agent concurrency/depth configuration with `agents.enabled` and `agents.max_concurrent_threads_per_session`;
+- enforce flat direct-child execution topology in the generated contract instead of relying on undocumented `agents.max_depth` behavior;
+- record the 2026-08-28 C01–C16 qualification attempt and its live Codex runtime blocker.
+
 ### Fixed
 
 - canonicalize event repository paths before active-run routing so source-worktree matching is stable across macOS symlink aliases and Windows path aliases;

diff --git a/capabilities/C01/README.md b/capabilities/C01/README.md
@@ -1,8 +1,11 @@
 # C01 — Capability evidence
 
-- Expected behavior: Repository skills are discovered from .agents/skills.
+- Expected behavior: Repository skills are discovered from `.agents/skills`.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS` against current official Codex skills documentation.
+- Live blocker: no authenticated Codex runtime or `codex` executable is available in the qualification environment.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`.
diff --git a/capabilities/C02/README.md b/capabilities/C02/README.md
@@ -1,8 +1,12 @@
 # C02 — Capability evidence
 
-- Expected behavior: allow_implicit_invocation false disables implicit invocation.
+- Expected behavior: `allow_implicit_invocation: false` disables implicit invocation while explicit `$skill` invocation remains available.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS`; repository metadata is configured for explicit-only activation.
+- Deterministic support: `test_skill_metadata_requires_explicit_activation` passed in GitHub Actions run #24.
+- Live blocker: no authenticated Codex runtime is available for implicit-vs-explicit prompt trials.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C03/README.md b/capabilities/C03/README.md
@@ -1,8 +1,12 @@
 # C03 — Capability evidence
 
-- Expected behavior: agents.max_depth 1 permits direct children and prevents deeper descendants.
-- Source: `DOCUMENTED`
+- Expected behavior: Generated execution contracts require an explicit flat direct-child topology without relying on a Codex nesting-depth setting.
+- Source: `CONTRACT_DEFINED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Baseline correction: current Codex documentation does not document `agents.max_depth`; PlanAnvil now enforces flat topology in deterministic contract validation.
+- Deterministic support: execution-contract topology tests passed in run #24; qualification adds a regression rejecting legacy-depth-only wording.
+- Live blocker: no authenticated Codex runtime is available to capture the required subagent event tree.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C04/README.md b/capabilities/C04/README.md
@@ -1,8 +1,11 @@
 # C04 — Capability evidence
 
-- Expected behavior: Nested agents are available when configured depth permits them.
+- Expected behavior: Codex subagent workflows use current agent enablement/concurrency settings; PlanAnvil does not require nested descendants.
 - Source: `DOCUMENTED`
-- Release-gating: `conditional`
-- Current result: `NOT_RUN`
+- Release-gating: `no` for baseline 2.2
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS`; current docs expose `agents.enabled` and `agents.max_concurrent_threads_per_session`, not a nesting-depth knob.
+- Live blocker: no authenticated Codex runtime is available for subagent spawning evidence.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+This capability is informational for PlanAnvil 2.2 because generated execution deliberately requires a flat direct-child topology.
diff --git a/capabilities/C05/README.md b/capabilities/C05/README.md
@@ -1,8 +1,11 @@
 # C05 — Capability evidence
 
-- Expected behavior: Planning-agent handoffs use explicit files and hashes.
+- Expected behavior: Required reviewer handoffs use explicit immutable files and hashes.
 - Source: `CONTRACT_DEFINED`
-- Release-gating: `conditional`
-- Current result: `NOT_RUN`
+- Release-gating: `yes`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Deterministic support: review-bundle integrity, immutable-write and tamper-detection tests passed in run #24.
+- Live blocker: the mandatory fresh reviewer handoff cannot be exercised without an authenticated Codex runtime.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C06/README.md b/capabilities/C06/README.md
@@ -1,8 +1,12 @@
 # C06 — Capability evidence
 
-- Expected behavior: PreToolUse covers supported calls but not every equivalent path.
+- Expected behavior: `PreToolUse` covers supported local function-tool paths but not every equivalent path.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS`; current hooks documentation lists supported shell, patch, MCP and local-function paths plus hosted/specialized exceptions.
+- Deterministic support: hook guard tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available to capture supported and bypass-path hook events.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C07/README.md b/capabilities/C07/README.md
@@ -3,6 +3,9 @@
 - Expected behavior: The Git guard rejects the configured unsafe-command corpus.
 - Source: `CONTRACT_DEFINED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Deterministic support: destructive-Git denial and hook-diagnostic classification tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available to capture the required live hook decisions and postconditions.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C08/README.md b/capabilities/C08/README.md
@@ -1,8 +1,12 @@
 # C08 — Capability evidence
 
-- Expected behavior: PreCompact can stop compaction.
+- Expected behavior: `PreCompact` can stop compaction.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS`; current hooks documentation states `continue: false` stops before compaction.
+- Deterministic support: checkpoint-required compaction tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available for manual and automatic compaction trials.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C09/README.md b/capabilities/C09/README.md
@@ -1,8 +1,11 @@
 # C09 — Capability evidence
 
-- Expected behavior: Compaction is allowed after checkpoint creation without a permanent loop.
+- Expected behavior: Compaction is allowed after checkpoint creation without a permanent stop loop.
 - Source: `CONTRACT_DEFINED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Deterministic support: valid checkpoint acceptance and recovery tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available to demonstrate stop → checkpoint → compact → recover without a loop.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C10/README.md b/capabilities/C10/README.md
@@ -1,8 +1,12 @@
 # C10 — Capability evidence
 
-- Expected behavior: PostCompact and SessionStart can provide recovery context.
+- Expected behavior: `PostCompact` and `SessionStart` can provide recovery context.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS`; current hooks documentation supports model-visible context for these events and compact-source continuation.
+- Deterministic support: recovery hook tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available for post-compaction continuation evidence.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C11/README.md b/capabilities/C11/README.md
@@ -3,6 +3,10 @@
 - Expected behavior: Project instructions follow documented directory scope and precedence.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS`; current docs define root-to-CWD discovery, `AGENTS.override.md` precedence, fallbacks and nearest-guidance override behavior.
+- Deterministic support: instruction-map and conflict tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available to record actual instruction sources.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C12/README.md b/capabilities/C12/README.md
@@ -1,8 +1,12 @@
 # C12 — Capability evidence
 
-- Expected behavior: project_doc_max_bytes may truncate automatic instruction loading.
+- Expected behavior: `project_doc_max_bytes` can truncate automatic instruction loading.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS`; current configuration/AGENTS documentation defines the combined instruction byte limit.
+- Deterministic support: explicit read/hash freshness tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available for a truncation fixture.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C13/README.md b/capabilities/C13/README.md
@@ -1,8 +1,12 @@
 # C13 — Capability evidence
 
-- Expected behavior: SubagentStart can add context but is not a startup blocker.
+- Expected behavior: `SubagentStart` can add context but `continue: false` does not stop subagent startup.
 - Source: `DOCUMENTED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Documentation check: `PASS` against current Codex hooks documentation.
+- Deterministic support: hook code compiles and the full suite passed in run #24.
+- Live blocker: no authenticated Codex runtime is available to spawn a subagent and capture the event.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C14/README.md b/capabilities/C14/README.md
@@ -1,8 +1,11 @@
 # C14 — Capability evidence
 
-- Expected behavior: Planning isolation preserves source branch, SHA, index, and files.
+- Expected behavior: Planning isolation preserves the source branch, SHA, index and files.
 - Source: `CONTRACT_DEFINED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Deterministic support: planning-worktree isolation, source-preservation and destination-safety tests passed in run #24.
+- Live blocker: the repository policy requires a complete committed capability package tied to an actual Codex run.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C15/README.md b/capabilities/C15/README.md
@@ -3,6 +3,9 @@
 - Expected behavior: Blind review is immutable and detects seeded contract defects.
 - Source: `CONTRACT_DEFINED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Deterministic support: review tamper detection, immutable publication and independent author-role tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available for a fresh blind reviewer against seeded defects.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.
diff --git a/capabilities/C16/README.md b/capabilities/C16/README.md
@@ -1,8 +1,11 @@
 # C16 — Capability evidence
 
-- Expected behavior: Git probe reports refs, branches, worktrees, index, commits, and cleanup.
+- Expected behavior: The Git probe reports refs, branches, worktrees, index, commits and cleanup accurately.
 - Source: `CONTRACT_DEFINED`
 - Release-gating: `yes`
-- Current result: `NOT_RUN`
+- Current result: `BLOCKED`
+- Qualification attempt: `2026-08-28`
+- Deterministic support: reversible probe, source preservation, cleanup and hook-failure classification tests passed in run #24.
+- Live blocker: no authenticated Codex runtime is available to execute the required matrix under supported Codex permission modes.
 
-Do not change the result to `REPRODUCED` until this directory contains the complete sanitized package defined in `capabilities/README.md`. Record the exact Codex version, model slug, operating system, permission mode, project trust, fixture commit, setup, cleanup, expected result, actual sanitized result, evaluation, and hashes.
+Do not change the result to `REPRODUCED` until the complete sanitized live package exists.