factory-ticket

Implement one ticket: $ARGUMENTS. You are already in its worktree, and the ticket is already claimed for you.

The dispatcher (orchestrator/tick.mjs) has done the setup: the ticket is In Progress, assigned, labelled ai:in-progress + agent:claude-code, and this worktree was created by the repo's own worktree-up.sh with its own branch, ports and database. Do not create another worktree, do not claim another ticket, and do not work anything except this one. This is the dispatched path, so it is always exactly one ticket — the protocol's human-requested ticket bundles (docs/protocol.md §6) never arrive here.

Do

1. Read the ticket and restate your approach as a comment on it.

2. Implement, touching only files matching its Owned Paths. Work discovered outside that set becomes a new Triage issue — never a widening of this one.

3. Verify with the ticket's exact Verification Command and the repo's configured verify command, on the final tree. Run only those two worktree gates: do not run bun test or the repo's full suite as a PR-opening gate. The full suite is CI's job, and concurrent dispatched worktrees make it a load-induced flake source. Never proceed past failing output; never weaken a test to get green.

4. UX critique after verification and before opening the PR when this introduces or materially changes a user-completable flow, interaction, state transition, error/recovery path, responsive layout, authentication, payment, onboarding, or destructive action. Spawn factory-ux-critic, fix in-scope FIX-FIRST findings, maximum 2 rounds, and file the rest to Triage. Skip it for isolated styling, copy-only edits, static content, icons/assets, and internal/admin-only surfaces unless the ticket identifies UX risk. State UX critique: required or UX critique: skipped — <reason> in the PR.

   The spawn prompt must carry the environment, spelled out — the subagent does not inherit your working directory, and sibling worktrees for other tickets exist right now:

   • worktree: <absolute path> — this worktree's root, written out in full. Never "the current directory".
   • How to launch and reach the app: dev server command and this worktree's port (not the repo default — worktree-up.sh assigned it a non-colliding one), or simulator target / electronAppPath, plus bin/dev-login.sh [role] where the repo has it.

   A returned VERDICT: BLOCKED - environment mismatch or unresponsive shell is a defect in your spawn prompt, not a UX finding: fix the missing or wrong path/launch details and re-spawn once. That re-spawn doesn't count against the 2-round limit — no review happened. If it blocks a second time, record UX critique: blocked — <what the environment did> and continue; do not keep re-spawning.

5. Never sleep to wait for CI. For the pushed head SHA, select only the CI workflow with gh run list --workflow ci.yml --commit <sha> --json databaseId --limit 1, then wait with gh run watch <run-id> --exit-status --interval 60. The run can lag the push, so retry the workflow-selected lookup for up to about two minutes when it is empty; do not substitute another workflow's run. Before a merge, use gh api repos/<owner>/<repo>/commits/<sha>/check-runs to assert every check run completed green. Sleep-polling is blocked by the harness, and a blocked tool call kills the run. The same rule applies to any wait — poll a condition with a real command, never a fixed sleep.

6. Push and open a PR against the configured base for this repo in config/repos.yaml, never GitHub's default branch: gh pr create --base <configured-base> --title "..." --body "Fixes <ISSUE-ID>" (append \n\nrun:$FACTORY_RUN_ID to the body when $FACTORY_RUN_ID is set, and omit it in interactive sessions when $FACTORY_RUN_ID is unset). Post the mandatory structured ## Handoff comment on the ticket before transitioning state:

   ## Handoff
   - PR: <url>
   - Verification: `<the ticket's exact command>` — pass, <one-line result summary>
   - UX critique: required — SHIP | required — FIX-FIRST resolved in <n> round(s) | required — NOT-ASSESSED, <what could not be driven> | blocked — <what the environment did> | skipped — <reason>
   - Files: <n> changed, all within Owned Paths   (or: exceptions listed with why)
   - Risks: <what the reviewer should look at first, or "none known">