candango-uat-runner

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Candango UAT Runner

Run UATs from `docs/features/<feature-slug>/uat.md` one scenario at a time.

All user-facing communication and written UAT reports must use `$caveman`.

## Inputs

Read:

- `docs/features/<feature-slug>/uat.md`
- `plan.md`, `issues.md`, and dashboard state when present
- repo setup docs under `docs/agents/`
- stack files such as `package.json`, lockfiles, Docker files, app config, routes, tests, and README
- relevant code paths for each UAT scenario

If feature folder is ambiguous, ask which `docs/features/<feature-slug>/` to use.

## Workflow

### 1. Prepare Run

Infer the technical stack and test surface:

- app start command
- focused test command
- seed/data setup
- browser/API/CLI path to exercise
- required env vars or credentials
- whether scenario is automated, manual, or both

Do not ask the user to inspect implementation details. Give user-facing steps only.

### 2. Guide Each Scenario

For each unchecked UAT:

1. Show UAT id/title and expected outcome.
2. Run automatable checks yourself when available.
3. Give precise manual steps for the user, based on the repo stack.
4. Ask for confirmation: passed, failed, or blocked.
5. Wait for user answer before moving to next UAT.

### 3. Mark Checked

When user confirms pass:

- update `uat.md`
- prefer existing checkbox/status format if present
- otherwise maintain a `## UAT Run Log` section:
  - `- [x] UAT-001: passed <ISO timestamp> - <short evidence>`

When blocked or failed:

- record under `## UAT Run Log`:
  - `- [ ] UAT-001: failed <ISO timestamp> - <problem>`
- do not mark checked

### 4. Failure Loop

If user says a UAT failed or describes a problem:

1. Capture exact repro, expected behavior, actual behavior, and evidence.
2. Search code/tests/docs for likely source.
3. Build a failure package with complete context:
   - UAT id/title and priority
   - user repro steps and user notes
   - expected vs actual behavior
   - evidence: logs, screenshots, command output, URLs, data state
   - related issue/plan references
   - suspected files/tests and validation commands
4. Send the failure package to `$candango-executor`.
5. `$candango-executor` restarts the fix loop with executor `$tdd` and validator review.
6. Re-run the failed UAT after executor/validator pass.
7. Return to the UAT checklist only after validation passes.
Never bypass `$candango-executor` for failed UAT fixes.

## Stop Conditions

Pause and ask when:

- UAT steps require unavailable credentials or third-party access
- environment setup cannot be inferred
- expected behavior conflicts with plan/issues
- user reports behavior that cannot be reproduced after investigation

## Output

Keep a terse caveman report:

- UATs passed
- UATs failed/blocked
- files updated
- commands run
- fix loop status when failures occurred

From lucasbayma/skills, MIT.

More from lucasbayma/skills