Review Pack · Task {{N}}

--- name: ayf:review-pack description: | Build a self-contained HTML review pack for task N: task spec, git diff, self-test findings, guided review questions, and a Pass/Needs-Work verdict. Opens in any browser -- no server required. allowed-tools: - Read - Write - Edit - Bash - Glob - Grep --- # /review-pack N -- Human Review Pack Produce a single self-contained HTML file at `.ay/retros/review-pack-{N}.html` that gives a human everything needed to review task N in one page: the task spec, the full diff, the agent's self-test findings, guided questions, and a Pass / Needs-Work verdict they can copy back. No dev server, no external assets -- it opens straight from disk in any browser. ## Phase 1: GATHER INPUTS 1. **Task spec** -- read `.ay/tasks/{N}.md` (try `.ay/tasks/{N}-*.md` if the bare name is missing). If none exists, use the task's BOARD.md row as the spec and note that no task file was found. 2. **Scope dirs** -- determine the directories this task was allowed to touch, from the task spec's "work in" / files section. If none are declared, leave scope empty (the diff will cover everything). 3. **Diff** -- capture the changes for this task: ```bash git diff main...HEAD -- # scope-limited when scope is known git diff main...HEAD # fallback: whole branch ``` Use `main...HEAD` (three dots) so the diff is the branch's own work, not unrelated changes on main. If `main` is not present, fall back to the repo's default branch or `HEAD~N`. 4. **Self-test findings** -- read `.ay/retros/agent-findings.json` if it exists (produced by `/self-test`). If absent, the pack renders a "run /self-test first" note in that section. ## Phase 2: BUILD THE HTML Write `.ay/retros/review-pack-{N}.html` (create `.ay/retros/` if needed) from the template below. Replace every `{{...}}` placeholder: - `{{N}}` -- task number. - `{{TASK_TITLE}}` -- task title (from the spec heading or BOARD row). - `{{GENERATED_AT}}` -- ISO-8601 UTC timestamp. - `{{TASK_SPEC_HTML}}` -- the task spec text, **HTML-escaped** (`&`→`&`, `<`→`<`, `>`→`>`), newlines preserved (it sits inside a `

`).
- `{{GIT_DIFF_HTML}}` -- the raw diff, **HTML-escaped** the same way. If empty,
  use the literal text `No changes on this branch.`
- `{{FINDINGS_ROWS}}` -- one `` per finding from agent-findings.json:
  `T-01passdescevidence`
  (escape `description` and `evidence`; `status` drives the badge class:
  `pass`|`fail`|`skip`). If findings are absent, replace the whole `` with
  a single row: `No agent-findings.json — run /self-test first.`.

Never inject un-escaped spec or diff text -- a stray `<` in the diff would break
the page.

Template (self-contained; CSS vars mirror `Human/assets/base.css`):

```html





Review Pack · Task {{N}}



  
    
      Review Pack · Task {{N}}
      {{TASK_TITLE}} · generated {{GENERATED_AT}}
    
    
  

  1 · Task Spec
{{TASK_SPEC_HTML}}

  2 · Changes (git diff main...HEAD)
{{GIT_DIFF_HTML}}

  3 · Self-Test Findings
    {{FINDINGS_ROWS}}Test Status Description Evidence
    
  

  4 · Guided Review
    Does this solve the stated problem?
    
    Are there edge cases not covered?
    
    Is the code scope-clean (only touches what it should)?
    
  

  5 · Verdict
    
      
      
    
    
    
  




```

## Phase 3: OPEN

Open the pack in the default browser (best-effort, non-fatal):

```bash
open ".ay/retros/review-pack-{N}.html" 2>/dev/null \
  || xdg-open ".ay/retros/review-pack-{N}.html" 2>/dev/null \
  || echo "Open .ay/retros/review-pack-{N}.html in your browser"
```

## Phase 4: REPORT

Print: `Review pack ready: .ay/retros/review-pack-{N}.html`

The human reviews in the browser, picks **Pass** or **Needs Work**, answers the
three questions, and clicks **Copy review as JSON**. They paste that JSON back
into the session. On **Pass**, proceed. On **Needs Work**, act on the answers
(fix, re-run `/self-test {N}`, re-run `/review-pack {N}`).