stack(pr138): integrate D7 implementation on top of evidence-first gate

.codex/skills/development-workflow/SKILL.md

@@ -29,9 +29,14 @@ Do not use this skill for document-only relocation/classification tasks; use `do
 
 ## Required evidence block in feature aggregation doc
 
-Each active feature aggregation doc MUST contain an `## Evidence` section with at least:
+Each active/in_review feature aggregation doc MUST contain an `## Evidence` section with at least:
 - commands executed (exact commands)
 - command results (pass/fail + key output summary)
+- contract delta (`schema`/`error semantics`/`retry`; each as changed or `none/n.a + reason`)
+- golden case file updates
+- regression summary (runner output)
+- observability/failure localization (`start/tool_call/end/fail` + locator fields)
+- structured review report (module boundary/state/concurrency/side-effect/coverage)
 - behavior verification (happy path + changed error branch)
 - risks and rollback notes
 - review/merge-gate evidence links (review request, key review threads, merge gate result)
@@ -75,10 +80,12 @@ Reference contract: `docs/guides/Evidence_Truth_Implementation_Strategy.md`.
 - verify status consistency: feature doc is source of truth, linked docs are non-conflicting
 - verify `docs/**` can stand alone as the current-state record without depending on OpenSpec internals
 - verify governance CI scope is complete (frontmatter by mode, link resolution, evidence block completeness, TODO->change mapping, checkpoint mapping)
-- verify feature evidence block includes commands, results, behavior verification, risks, rollback, and review links
+- verify feature evidence block includes acceptance-pack semantics and structured review report content
+- verify review links include both intent PR and implementation PR records
 
 4. review-merge-gate
 - request review with explicit evidence links from the feature aggregation doc
+- reviewer default path is evidence-first, then risk-targeted code sampling (contract boundary + control-flow/side-effects)
 - process review feedback thread-by-thread and keep evidence section updated with fix commits
 - require explicit non-blocking merge gate signal (approval or equivalent repo policy signal) before archive
 - keep mailbox records linked: temporary coordination notes vs retained audit evidence

.github/pull_request_template.md

@@ -16,10 +16,26 @@ List all touched files and why each file changed.
 
 If this PR is a large diff (>300 changed lines), explain why split PRs are not possible and provide a split follow-up plan.
 
+## Intent / Implementation Gate (required)
+- Intent PR link (docs-only): `...`
+- Intent PR merged into `main` before this implementation started: [ ] Yes
+- Implementation PR link (this PR): `...`
+
 ## Acceptance Criteria
 - [ ] Criteria 1
 - [ ] Criteria 2
 
+## Acceptance Pack (required)
+- Contract Delta (`schema` / `error semantics` / `retry` dimensions; each as changed or `none/n.a + reason`):
+  - `...`
+- Golden Cases (new/updated file names):
+  - `...`
+- Regression Summary (runner outputs):
+  - `...`
+- Observability and Failure Localization (`start/tool_call/end/fail` + `run_id/tool_call_id/capability_id/attempt/trace_id` + one of `error_code`/`error_type`/`exception_class`/`ToolResult.error`):
+  - `...`
+- Structured Review Report attached: [ ] Yes
+
 ## Test Evidence (required)
 - Local commands run:
   - `...`
@@ -35,6 +51,27 @@ If this PR changes high-risk runtime paths (auth/concurrency/execution control),
 - Main risk points:
 - Rollback plan:
 
+## Structured Review Report (required)
+### Changed Module Boundaries / Public API
+- `...`
+
+### New State
+- New cache/global/singleton state:
+- Lifecycle and cleanup:
+
+### Concurrency / Timeout / Retry
+- New concurrency points:
+- Timeout/retry locations:
+- Upper bounds:
+
+### Side Effects and Idempotency
+- Side effects:
+- Anti-duplication strategy:
+
+### Coverage and Residual Risk
+- Covered tests/evaluations:
+- Residual risks not covered:
+
 ## Dependency and Lockfile Changes
 - Lockfile changed in this PR: [ ] Yes [ ] No
 - If yes, manifest updated in same PR (`requirements.txt`/`pyproject.toml`/`package.json`): [ ] Yes [ ] No [ ] N/A

.github/workflows/ci-gate.yml

@@ -140,5 +140,5 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v4
 
-      - name: Check governance evidence truth contract
+      - name: Check governance evidence-first contract (structure + semantics)
         run: ./scripts/ci/check_governance_evidence_truth.sh

dare_framework/plan_v2/__init__.py

@@ -7,16 +7,21 @@
 from dare_framework.plan_v2.registry import SubAgentRegistry
 from dare_framework.plan_v2.types import (
     Milestone,
+    PlanStateName,
     PlannerState,
+    STEP_STATES,
     Step,
     Task,
+    is_valid_state_transition,
 )
 from dare_framework.plan_v2.prompts import PLAN_AGENT_SYSTEM_PROMPT, SUB_AGENT_TASK_PROMPT
 from dare_framework.plan_v2.tools import (
     CreatePlanTool,
     DecomposeTaskTool,
     DelegateToSubAgentTool,
+    FinishPlanTool,
     ReflectTool,
+    ReviseCurrentPlanTool,
     ValidatePlanTool,
     VerifyMilestoneTool,
 )
@@ -25,15 +30,20 @@
     "CreatePlanTool",
     "DecomposeTaskTool",
     "DelegateToSubAgentTool",
+    "FinishPlanTool",
     "Milestone",
     "PLAN_AGENT_SYSTEM_PROMPT",
+    "PlanStateName",
     "Planner",
     "SUB_AGENT_TASK_PROMPT",
+    "STEP_STATES",
     "PlannerState",
     "ReflectTool",
+    "ReviseCurrentPlanTool",
     "Step",
     "SubAgentRegistry",
     "Task",
     "ValidatePlanTool",
     "VerifyMilestoneTool",
+    "is_valid_state_transition",
 ]

dare_framework/plan_v2/planner.py

@@ -9,7 +9,9 @@
 from dare_framework.plan_v2.tools import (
     CreatePlanTool,
     DecomposeTaskTool,
+    FinishPlanTool,
     ReflectTool,
+    ReviseCurrentPlanTool,
     SubAgentTool,
     ValidatePlanTool,
     VerifyMilestoneTool,
@@ -35,6 +37,8 @@ def __init__(
             self._tools.extend([
                 CreatePlanTool(self._state),
                 ValidatePlanTool(self._state),
+                ReviseCurrentPlanTool(self._state),
+                FinishPlanTool(self._state),
                 VerifyMilestoneTool(self._state),
                 ReflectTool(self._state),
                 DecomposeTaskTool(self._state),