| name | test-execution-protocol |
| description | Test execution protocol. Use by test-runner to choose checks, run commands, and report failures. |
Test Execution Protocol
Order
- Read
.agents/PLAN.md and .agents/PROGRESS.md.
- Identify changed areas.
- Run targeted tests first.
- Add lint, typecheck, build, or broader tests when risk warrants it.
- Capture exact commands and results.
- Classify failures and suggest owner.
- Report results using
.agents/PROGRESS.md section 5. Checks from templates/PROGRESS.md, or return the same fields to the Orchestrator for recording.
Failure classes
- Backend implementation.
- Frontend implementation.
- Contract mismatch.
- Test fixture.
- Environment.
- Existing unrelated.
- Ambiguous.
- Not applicable.
Do not edit code.
Required test-run fields
For each command/check, report:
- Status:
passed, failed, partially-passed, skipped, or blocked.
- Agent:
test-runner.
- Phase or final verification label.
- Exact command in a code block when writing markdown.
- Result summary.
- Minimal relevant output, abbreviated if long.
- Failure classification checklist/value.
- Next action.
- Routed owner or
none.
Failure handoff expectations
When a failure is actionable, include enough information for the Orchestrator to route it without rerunning the command:
- Exact command.
- Pass/fail status.
- Minimal relevant output.
- Failure class.
- Suspected owner and why.
- Suggested next command after fix.
Only when explicitly assigned a durable test review or an actionable repository issue must become cross-agent evidence, read .agents/REVIEW.md as the read-only schema and write a complete assigned absent .agents/REVIEW_test-<n>.md artifact. Ordinary test output is not a review artifact; environment-only and ambiguous failures remain test evidence/routing metadata. Never write findings to .agents/REVIEW.md, self-allocate, or overwrite an artifact.
A test re-review receives the next assigned test successor artifact with predecessor/disposition references; never overwrite its predecessor.
The artifact uses a canonical finding, ordered with other findings by priority:
### F1 — [P1] Short imperative title
File: `/absolute/path/to/file.ext:123`
Comment: One concise paragraph explaining the failing scenario, relevant evidence, and needed fix when useful.
Use [P0] for a universal release, operations, or major-usage blocker; [P1] for urgent next-cycle work; [P2] for a normal eventual fix; and [P3] for a low-priority improvement. Keep titles under 80 characters, use the shortest useful file location, and state No findings. if no actionable repository issue was identified. Do not use Blocker, Important, Optional, numeric priorities, or JSON priorities. Return only artifact path, verdict, and blocking status when an artifact was assigned.
If asked to write .agents/HANDOFF.md, use templates/HANDOFF.md and the failure-routing protocol.
