feat: consolidate requirement verdict computation across status and MCP

[jimisola]

Summary

• Adds the OpenSpec change consolidate-requirement-verdict (proposal, design, spec delta, tasks) and implements it.
• Extracts compute_requirement_status() (per-requirement implementation/automated/manual verdict) and a shared _requirement_to_dict() serializer in statistics_service.py, queried through the repository's scoped per-requirement getters.
• StatisticsService (status/report/export/MCP get_status) and the MCP get_requirement_status/get_requirements_status tools now derive their verdict from this single predicate, instead of two independent implementations that had already drifted once (fix(mcp): correctly flag skipped/missing automated tests as unmet #411).
• Deletes the duplicate _compute_meets/_build_automated_test_summary predicate from common/queries/details.py.
• Exposes include_post_build (default False) on the MCP status tools for parity with status --with-post-tests.
• BREAKING (MCP output): get_requirement_status/get_requirements_status now emit the unified status shape (completed, implementation_type, automated_tests/manual_tests), replacing the old meets_requirements/flat test_summary shape. Some verdict values also change for requirements with post-build-only SVCs, not-applicable cases, and unrecorded test executions — these now match status exactly.
• Adds MCP_0005/SVC_MCP_0005 to the reqstool SSOT and a test asserting get_requirement_status/get_requirements_status agree with StatisticsService in both scoping modes, covering the previously divergent REQ_ext002_300.
• Adds a CLAUDE.md convention note: the verdict has one source of truth; no parallel re-derivation.

Closes #412

Test plan

• hatch run dev:pytest --cov=reqstool — 940 passed
• hatch run dev:flake8 — clean
• CLAUDE.md regression smoke (status/report on test_standard/test_basic fixtures) — byte-identical to main
• reqstool status local -p docs/reqstool — 72/72 complete, SVC_MCP_0005 covered
• openspec validate --all --strict — all pass

[jimisola]

Full PR Review — Consolidated Findings

Scope: docs-only (OpenSpec proposal/design/tasks/spec-delta + CLAUDE.md note + OpenSpec config scaffold). No code changes. Every factual claim in design.md was verified against the repo at the PR HEAD — all referenced symbols exist and MCP_0005 is the correct next ID. openspec validate --strict passes.

#	Prio	Location	Finding	Fix
1	Medium	specs/mcp/spec.md:3-7	Spec delta is a tautological placeholder ("The system SHALL implement MCP_0005" / "SHALL pass SVC_MCP_0005"). It carries none of the behavioral requirement that proposal.md describes. openspec validate --strict passes (structure-only), so the stub slips through.	Write a real requirement + WHEN/THEN scenario expressing that the per-requirement MCP status tools report a verdict/structure identical to the status command, from one shared computation.
2	Low	CLAUDE.md:164	"silently drifted and caused bugs (#411, #412)" — #412 is the open consolidation tracking issue (this work), not a bug caused by drift; only #411 was the drift bug.	Reword to "caused bug #411; consolidation tracked in #412".
3	Low	tasks.md:210	Task 5.2 hardcodes "72/72 complete"; tasks 1.1/1.2 add a new requirement, so the count is stale on arrival and brittle.	Use "all requirements complete with SVC_MCP_0005 covered" instead of a fixed count.
4	Low	design.md/proposal.md	Refer to the details.py function as get_requirements_status, but the impl is get_requirements_status_all (only the MCP tool is get_requirements_status). tasks.md uses the correct name.	Align prose with get_requirements_status_all.
5	Info	openspec/config.yaml	Committed as the default scaffold with all context/rules commented out — optional opportunity to populate context:.	Optional; fine to leave as standard scaffold.

Net: clean, well-reasoned proposal. #1 is the only near-blocking item — the spec delta should express the real requirement since that file survives into the archive.

---

Automated — /x:full-pr-review

[jimisola]

Correction to the review above

After checking the existing openspec/specs/mcp/spec.md, two of my findings were wrong — retracting them:

• #1 (spec delta "placeholder") — RETRACTED. The existing mcp spec uses the exact same The system SHALL implement MCP_000X / SHALL pass SVC_MCP_000X form by design: its Purpose note states requirement/SVC content is owned by reqstool (SSOT) and injected at read time via reqstool enrich. The MCP_0005 delta correctly follows that convention; there is nothing to fix.
• #4 (function name) — RETRACTED. design.md/proposal.md correctly name the MCP tools (get_requirements_status); tasks.md correctly names the impl function (get_requirements_status_all). Both accurate in context.

Applied fixes (committed to the branch):

• #2 CLAUDE.md — reworded "caused bugs (#411, #412)" → "caused bug #411 (consolidation tracked in #412)", since #412 is the tracking issue, not a drift-caused bug.
• #3 tasks.md 5.2 — replaced the hardcoded "72/72 complete" with "all requirements complete with SVC_MCP_0005 covered".

#5 (config.yaml scaffold) left as-is (optional). openspec validate --strict still passes.

---

Automated — /x:full-pr-review

[jimisola]

Full PR Review — Consolidated Findings

7 checks run: review, code-review, smells, advanced-smells, security, cruft, testing. Security: no findings.

#	Prio	Location	Finding	Fix
1	High	docs/modules/ROOT/pages/mcp.adoc:177,189,194-206	Docs still describe the deleted MCP shape (implementation, test_summary, meets_requirements) — actively misleading for MCP API consumers.	Update to the new shape and document include_post_build.
2	High	tests/unit/reqstool/common/queries/test_details.py:959-983	include_post_build parametrized test can't prove the flag changes anything — no SVC in the fixture sets phase, so both modes select the identical SVC set.	Add a fixture SVC with phase: post-build and assert behavior actually differs.
3	High	src/reqstool/storage/requirements_repository.py:131-136	New get_svc() — load-bearing for every verdict — has zero test coverage, including the None path.	Add a unit test for found/not-found.
4	Medium	statistics_service.py:22, details.py:8	_requirement_to_dict is underscore-prefixed yet listed in __all__ and imported across a module boundary. (4/7 checks)	Rename to requirement_to_dict.
5	Medium	statistics_service.py:101-130	N+1 query pattern: per-requirement, per-SVC get_svc() round-trips replace 4 bulk fetches. Indexed, so sub-ms at today's scale per design doc. (4/7 checks)	No fix required now; optional follow-up if profiling shows it matters.
6	Medium	statistics_service.py:170-183	_compute_requirement_automated_stats double-queries annotations (checks truthiness, then get_test_results_for_svc re-fetches them internally).	Reuse the already-fetched annotations list inline.
7	Medium	statistics_service.py:101	compute_requirement_status has zero direct unit tests; the TypeError branch in _check_implementation is untested anywhere.	Add direct unit tests covering not_applicable, missing MVR/test, and the TypeError branch.
8	Medium	test_details.py:961-983	The new agreement test only proves both callers agree — they call the same predicate now, so it could pass trivially even with a predicate bug.	Add assertions on known expected values for a specific requirement.
9	Medium	test_details.py:167	_make_db_with_req default verification change (MANUAL_TEST→AUTOMATED_TEST) silently shifts which gate several pre-existing tests exercise.	Document/assert which gate each test exercises; cover the now-orphaned MANUAL_TEST+MVR combination explicitly.
10	Low	statistics_service.py:101	req parameter is untyped while siblings are.	Annotate as req: RequirementData.
11	Low	test_details.py	Test names retain old meets_requirements/compute_meets vocabulary.	Rename to test_completed_*/test_verdict_*.
12	Low	mcp/server.py:69,109	No negative-path test for include_post_build combined with an invalid id.	Parametrize the not-found test over include_post_build.

Not flagged (verified, no action needed): get_svc() SQL correctly parameterized; compute_requirement_status faithfully replicates pre-PR StatisticsService logic; no dead code left from the deleted predicate; include_post_build plumbing proportionate; no security findings.

Fixes for all of the above are being pushed to this branch.

---

Automated — /x:full-pr-review

[jimisola]

Full PR Review — Round 2 (commit e1a0145)

All 12 round-1 findings verified as correctly and completely fixed. 854 unit + 17 integration tests pass; flake8/black clean. Round 2 surfaced 2 new, minor follow-ups the fix commit's scope didn't reach:

#	Prio	Location	Finding	Fix
1	Medium	src/reqstool/common/queries/details.py:91-92 (get_urn_details)	Same duplicate-query pattern eliminated in compute_requirement_status still exists here: fetches test_annotations then calls get_test_results_for_svc(svc.id), which re-fetches the same annotations internally.	Change to repo.get_test_results_for_annotations(svc.id.urn, test_annotations).
2	Low	src/reqstool/storage/requirements_repository.py:270 (get_test_results_for_annotations)	New repository method has no direct unit test — only exercised transitively via compute_requirement_status tests. Its sibling get_svc got a direct found/not-found pair in the same commit; this one didn't.	Add a direct unit test covering the CLASS and METHOD annotation branches.

Everything else checked out clean: all 6 round-1 test-coverage fixes verified substantive (including mutation-testing the include_post_build test to confirm it's load-bearing), the _requirement_to_dict → requirement_to_dict rename is complete repo-wide, docs are consistent, no new cruft, no security findings, no over-engineering.

Fixes for both are being pushed to this branch.

---

Automated — /x:full-pr-review