SuperAPI AI Directory

test

Testing & quality

View repository

Updated 10-15 21:38

About this skill

test - Verification Stage

Invariants

  • Verify against the task document acceptance criteria.
  • Use the best available project verification path.
  • Do not require one specific browser automation tool in core workflow.
  • Always write one final report to docs/testing/.
  • When tests pass and the user approves, update TASKS.md to Approved and re-read the tracker to confirm the row moved. A stale tracker blocks downstream and headless sessions.

Runtime adapters may expose this stage as a slash command, menu action, or natural-language skill invocation. The portable stage name is test.

Workflow

  1. Read AGENTS.md.
  2. Resolve the task ID and read docs/task/{ID}-{task-name}.md.
  3. Confirm the task reached the quality gate.
  4. Build a verification plan from Acceptance Criteria and Verification Approach.
  5. Prefer existing project commands and CI-compatible checks.
  6. Run safe verification commands.
  7. Use browser automation or manual checks only when required by the task.
  8. Write docs/testing/{ID}-{task-name}.md using docs/templates/test-report.md.
  9. Move the task to Approved only after tests pass and the user approves, or leave it in Testing if fixes are needed.
  10. If the move was made, re-read TASKS.md and verify the task row is in Approved and absent from Testing. If the row is not in the expected section, the move was missed — perform the edit now before handoff.

Verification Priority

Use the narrowest reliable check first:

  1. Unit tests for pure logic.
  2. Integration tests for API, database, service, or cross-module behavior.
  3. Browser or end-to-end checks for user-visible workflows.
  4. Static checks such as typecheck, lint, or build where they are part of project quality.
  5. Manual inspection only when automation is unavailable or not cost-effective.

If no viable verification path exists, write a BLOCKED report explaining what is missing.

Report Rules

  • Write the final report only under docs/testing/.
  • Do not scatter temporary reports in the repo root.
  • Record commands exactly as run.
  • Record skipped checks with reasons.
  • Include enough evidence for the implementation stage to fix failures without session history.

Handoff

PASS:

Verification passed for task {ID}
Report: docs/testing/{ID}-{task-name}.md
Next stage after user approval: document {ID}

FAIL or BLOCKED:

Verification failed or blocked for task {ID}
Report: docs/testing/{ID}-{task-name}.md
Return to implementation with the report context.