| name | dev-test-gaps |
| description | Internal /dev requirement-to-evidence audit after authenticated implementation. |
| user-invocable | false |
| disable-model-invocation | true |
| hooks | {"PreToolUse":[{"matcher":"Write|Edit|MultiEdit|NotebookEdit","hooks":[{"type":"command","command":"bun ${CLAUDE_PLUGIN_ROOT}/hooks/orchestrator-mutation-guard.ts --workflow dev"}]},{"matcher":"Bash","hooks":[{"type":"command","command":"bun ${CLAUDE_PLUGIN_ROOT}/hooks/orchestrator-mutation-guard.ts --workflow dev"}]}]} |
Dev test-gap audit
!bun ${CLAUDE_SKILL_DIR}/../../scripts/load-constraints.ts dev-test-gaps
Use the receipt-selected {planFile, planHash} and current TaskList only. Do not read a fixed
SPEC.md or PLAN.md, and do not produce VALIDATION.md, a progress ledger, or a visible review
artifact.
Audit
- Resolve and rehash the approved dev receipt immediately before auditing. Reject stale identity.
- Read every
REQ-NN, task criterion, real-test contract, and requirement-to-test mapping from the
selected generated plan.
- Inspect concrete unit, integration, and required E2E evidence. A skipped test, mock-only test,
structural grep, or an old command result is not coverage.
- Run the relevant tests fresh and read their exit status and output. User-facing requirements need
the declared real/E2E evidence, not unit-only evidence.
- Classify every requirement
COVERED, PARTIAL, or MISSING, naming test files, test names,
evidence commands, and raw outcomes.
For every PARTIAL or MISSING criterion create exactly one current-plan TaskList finding with
item_kind: test-gap, requirement ID, evidence, and any dependency. If a gap reveals an
implementation defect, create a repair work item/dependency for dev-implement; this audit does
not fix production code. Test-only changes still require a valid RED/GREEN result and independent
verification.
Return contract
Return, rather than persist, this complete result:
Test-gap audit result
- Plan: {planFile, planHash}
- Requirement → test matrix: [{requirement, criterion, tests, evidence, status}]
- Findings: [TaskList IDs]
- Fresh suite evidence: [commands, exit codes, relevant output]
- Verdict: COVERED | GAPS_FOUND
- Next ready wave: [TaskList IDs]
Only COVERED with no open gap finding admits dev-verify. Otherwise route the repair IDs through
the shared implementation adapter and re-audit every requirement.
