| name | checklist |
| description | Run a category-specific quality pass (security / performance / accessibility / i18n / privacy / observability) against the current plan and code. |
| metadata | {"aiadev":{"version":"0.2.0","inputs":[{"type":"text","description":"Category name (security, performance, accessibility, internationalization, privacy, observability, or a preset-defined category)."},{"type":"dir","path":"specs/<branch>/"}],"outputs":[{"type":"file","path":"specs/<branch>/checklists/<category>.md"}],"requires":["constitution","templates/checklist-template.md"]}} |
Checklist
Apply a focused quality pass in one category. Default categories ship with the template; presets can add more.
Announce at start: "Using the checklist skill in mode."
Preconditions
plan.md exists. Running a checklist before plan is premature.
templates/checklist-template.md is available.
- The category is either a default or registered by the active preset.
Loop
- Copy the template to
specs/<branch>/checklists/<category>.md.
- Set the Scope section. Which parts of the plan does this pass cover; which are out of scope and why.
- For every item in the category's default list:
- Read the relevant code / plan / tests.
- Decide
PASS, FAIL, or N/A.
- Paste or link the evidence that supports the status.
- If the preset defines extra items, run them too. Do not silently skip.
- Failed items go in their own table with a remediation note and owner task id (create a new
T<NNN> in tasks.md if none exists).
- Sign off when every non-
N/A item is PASS or has an owner task.
Rules
- Do not mark an item
PASS without evidence. "Looks fine" is not evidence.
- Do not fix findings inline. Open a task instead; the fix goes through
implement like any other work.
- Do not combine categories. One file per category keeps reviewers honest.
Hand-off
- All items resolved → append the checklist result to the PR body in
requesting-code-review.
- Failed items with owner tasks → the owner tasks go into the next
implement run.