| name | manual-verification-docs |
| description | Organize post-implementation manual verification steps, execution methods, and expected results into a checklist-oriented dated Markdown report. Use when the user asks for verification steps, acceptance checks, expected outcomes, or manual test notes for UI, API, job, permission, or notification changes after implementation. |
Manual Verification Docs
Create a human-friendly manual verification guide for AI-agent changes at a caller-specified Markdown path.
Quick start
- Collect the implementation scope, changed files, intended users, and impact area.
- Decide the output location with the caller. When used inside an operation, prefer the mission-scoped output file path from the step output contract. When used standalone, pass either an explicit output path or an output directory.
- Run the scaffold generator.
python3 .opencode/skills/manual-verification-docs/scripts/create_verification_doc.py --slug <topic> --title "<verification target>" --output-path <path>
- Fill in the generated Markdown file with checklist items, execution steps, expected results, observation points, evidence notes, and open items.
- Share the created file path with the user and confirm that no important scenarios are missing.
Workflow
1. Define the verification scope
- Identify what must be verified to determine whether the change succeeds.
- List affected surfaces such as screens, APIs, jobs, permissions, and notifications.
- Separate automated coverage that already exists from areas that still need manual checks.
2. Generate the scaffold
- Use lowercase kebab-case for
--slug.
- If the caller provides
--output-path, write exactly that file and do not create a second copy anywhere else.
- If the caller provides
--output-dir, create a dated file in YYYYMMDD-xxx.md format inside that directory.
3. Make the guide concrete
- Document startup steps, required data, permissions, and configuration values in the prerequisites section.
- For each scenario, create checklist items that a user can mark off one by one.
- Under every checklist item, include the execution steps, expected result, observation point, and an evidence or notes field.
- Make every expected result observable and specific.
- For anything that cannot be verified through normal app interaction, document an exact execution method instead of a vague instruction.
- For API verification, provide a runnable command such as
curl, including required headers, auth placeholders, and example payloads when applicable.
- For jobs, scripts, queues, or notifications, describe how to trigger the behavior and where to observe completion or side effects.
- If the exact execution method is unknown, write
Maintainer input required: with the missing detail instead of inventing a partial procedure.
- Record unverified items or known constraints at the end.
4. Review with the user
- Confirm that the intended verification environment matches reality.
- Check for missing happy paths, error paths, and permission scenarios.
- If the caller gave an exact output path, update that file rather than creating another copy.
- If the caller gave only an output directory, add another dated file only when a separate guide is truly needed.
Rules
- Write this skill file, examples, and script-facing text in English.
- Write the generated verification document in the caller-requested language, and default to English only when no language was specified.
- Describe reproducible verification steps, not implementation internals.
- Use checklist items for actionable verification work.
- Each checklist item must tell the verifier how to execute the check, especially for non-UI paths.
- Include a matching expected result for every checklist item.
- Include an observation point for every checklist item so the verifier knows where to inspect success or failure.
- Leave space for evidence, notes, or blockers under each checklist item.
- Use only the caller-specified output path or output directory.
- Do not create duplicate copies in another workspace, repository, or reports directory unless the caller explicitly asks for more than one file.
Advanced features
See REFERENCE.md for concrete patterns for API, job, notification, and permission verification steps.
Output
- Output location: caller-specified via
--output-path or --output-dir
- Scaffold generator:
.opencode/skills/manual-verification-docs/scripts/create_verification_doc.py