feat(skills): add gap-analysis verification-gate skill + evals

[kreneskyp]

What

A new gap-analysis skill — a post-implementation verification gate over one targeted plan — plus an eval suite.

Skill (skills/gap-analysis/): SKILL.md + 6 step references. It checks:

1. Plan completion — every Task is status: done
2. Test-matrix verification — each matrix TC-xxx is backed by a real tracking tag in the tests
3. Underspecified code — code/behavior with no owning StR/US/FR (reverse gap) + stub detection
4. (optional) Semantic review — intent↔test↔code agreement (user-gated)

Output: one quire-validated SpecReview (analysis: gap-analysis) at reviews/YY-MM-DD-<slug>.md with a Verdict (PASS/CONDITIONAL/FAIL) and a findings table.

Evals (evals/): EV-030..EV-033 cover happy + sad branches via a shared seedGapBundle fixture, plus a new fileContains assertion (evals/lib/assert.mjs) that checks the actual Verdict/findings:

Eval	Path
EV-030	sad (combined) → FAIL
EV-031	happy → non-blocking verdict (not FAIL)
EV-032	sad (medium-only) → untraced code flagged, CONDITIONAL
EV-033	optional semantic review catches a hollow test

Validation

• make test green (102/102); make lint clean for the changed files.
• All four evals run 4/4 PASS locally (sonnet, 0 validation rejections).

Dependency

Requires agent-ix/spec-artifacts-process v0.4.0 (the gap-analysis analysis enum — agent-ix/spec-artifacts-process#5). The evals are RED-in-CI until v0.4.0 is tagged and the default-modules.yaml pin is bumped to it (same release-coupling as EV-026); proven green locally by temporarily sourcing the module from the local working tree.

🤖 Generated with Claude Code