spec-verify
Installation
SKILL.md
Contains Hooks
This skill uses Claude hooks which can execute code automatically in response to events. Review carefully before installing.
/spec-verify - Verification Phase
Phase 3 of the /spec workflow (features). Runs comprehensive verification: automated checks, code review, program execution, and E2E tests. For bugfix plans, use spec-bugfix-verify instead.
Input: Plan file with Status: COMPLETE
Output: Plan status → VERIFIED (success) or loop back to implementation (failure)
⛔ KEY CONSTRAINTS
- Run code review when enabled — Step 3.1 launches
changes-reviewviaTask(subagent_type="pilot:changes-review")whenPILOT_CHANGES_REVIEW_ENABLEDis not"false"(read in Step 0). To disable, use Console Settings → Reviewers → Changes Review toggle. - Only changes-review — NEVER spec-review — Do NOT launch
spec-reviewduring verification. Do NOT read or referencefindings-spec-review-*.jsonfiles — they are stale artifacts from the planning phase that were already addressed during implementation. If you encounter a spec-review findings file, ignore it completely. - NO stopping — Everything automatic. Never ask "Should I fix these?"
- Fix ALL findings — must_fix AND should_fix. No permission needed.
- Code changes finish BEFORE runtime testing — Phase A then Phase B.
- Plan file is source of truth — re-read it after auto-compaction, don't rely on conversation memory.
- Re-verification after fixes is MANDATORY — fixes can introduce new bugs.
- Quality over speed — never rush due to context pressure.