Verify

Multi-stage quality gate. Run after finishing implementation, before superpowers:finishing-a-development-branch.

Fail-fast: Stop at the first failing stage. Do not continue.

---

Stage 1 — Get Changed Files

git diff HEAD --name-only

Store this list. Every conditional stage below checks against it.

---

Stage 2 — Lint

Run only on changed files (never without args — it will timeout):

bun lint <changed-files-space-separated>

STOP — do not continue to Stage 3.

---

Stage 3 — Typecheck

bun typecheck

STOP — do not continue to Stage 4.

---

Stage 4 — Coding Standards Audit (conditional)

Trigger: Any changed file matches **/*.ts or **/*.tsx

Dispatch a general-purpose subagent via the Task tool with this prompt:

Read ./coding-standards/SKILL.md (the full skill, including the Code Smell Checklist).

Run through every item in the Code Smell Checklist against the following changed files:

Changed files: <list of .ts/.tsx files>

For each violation found, report:
- File path and approximate line number
- Which checklist item it violates
- The offending code snippet

Report violations only. If everything is clean, report "No violations found."

STOP — do not continue to Stage 5.

---

Stage 5 — UI Verification (conditional)

Trigger: Any changed file matches src/app/**/*.tsx, src/components/**, or src/styles/**

5a — Determine Acceptance Criteria

1. Check the current conversation — user may have described expected behavior, a Linear ticket, or a spec
2. If a Linear issue is active or referenced → fetch it with mcp__linear__get_issue
3. If still unclear → use AskUserQuestion with the most logical options for this context, always including an "Other (describe)" option

5b — Start Dev Server (if not already running)

if curl -s http://localhost:3000 > /dev/null 2>&1; then
  echo "Dev server already running — skipping start"
  DEV_STARTED_BY_VERIFY=false
else
  bun dev &
  DEV_PID=$!
  DEV_STARTED_BY_VERIFY=true
  until curl -s http://localhost:3000 > /dev/null 2>&1; do sleep 2; done
  echo "Dev server ready"
fi

5c — Dispatch Agent-Browser Task

Spawn a general-purpose subagent via the Task tool. In the prompt:

• Tell it to read ./agent-browser/SKILL.md for the full command reference and session management patterns
• Include the acceptance criteria from 5a
• Include the relevant route(s) to navigate to
• Ask it to use --session-name verify (auto-saves/restores Clerk session across runs)
• Ask it to use screenshot --annotate at key states (gives visual layout + numbered refs)
• Ask it to report pass/fail per criterion and close the session when done

Example prompt (adapt route and criteria to this feature):

Read ./agent-browser/SKILL.md for browser automation command reference, then verify the following at http://localhost:3000.

Use --session-name verify for ALL commands. This auto-saves Clerk session state after
first login so subsequent runs skip authentication.

Route: /[relevant path]

Acceptance criteria:
- <criterion 1>
- <criterion 2>

Workflow:
1. agent-browser --session-name verify open http://localhost:3000/[route]
2. agent-browser --session-name verify wait --load networkidle
3. Check current URL — if redirected to Clerk login:
   a. agent-browser --session-name verify snapshot -i
   b. Fill email with $AGENT_BROWSER_EMAIL, password with $AGENT_BROWSER_PASSWORD
   c. agent-browser --session-name verify wait --url "**/localhost:3000**"
   (On subsequent runs, Clerk session is restored automatically — skip this step)
4. For each criterion: navigate/interact as needed, take annotated screenshot, verify
5. agent-browser --session-name verify screenshot --annotate  (at each key state)
6. Report pass/fail per criterion with screenshot evidence
7. agent-browser --session-name verify close

Credentials: $AGENT_BROWSER_EMAIL / $AGENT_BROWSER_PASSWORD from env.example → .env.local. To reset an expired session: agent-browser state clear verify.

5d — Stop Dev Server (only if we started it)

if [ "$DEV_STARTED_BY_VERIFY" = "true" ]; then
  kill $DEV_PID 2>/dev/null
fi

---

Stage Summary

Changed files contain	Stage triggered
**/*.ts or **/*.tsx	Stage 4: Coding standards audit
src/app/**/*.tsx, src/components/**, src/styles/**	Stage 5: UI verification
Neither	Stages 1–3 only

---

Final Report

When all stages pass, output:

✅ Lint              — N files
✅ Typecheck
✅ Coding standards — N files, no violations  (if triggered)
✅ UI verified       — N/N criteria passed    (if triggered)

→ Ready for: superpowers:finishing-a-development-branch