Ship Pr

Ship PR

Risk → merge authority

Self-assess; user policy overrides.

Kent's standing policy (2026-08-08): auto-ship (squash-merge + verify deploy) once AI reviewer feedback is addressed and CI is green, unless high risk. High risk still parks ready-for-review unless merge authority was granted explicitly.

• Low — green CI; nits ignorable; squash-merge when policy allows.
• Medium — wait for AI reviewer(s); address valid feedback (ignore insignificant nits / already-fixed / wrong); then merge when policy allows.
• High — leave ready-for-review unless the user granted merge authority.

AI reviewers

Prefer Cursor Bugbot (Cursor Bugbot check / cursor[bot] review comments). Never comment bugbot run or cursor review yourself, including via gh or GitHub request as kody-bot. Trigger with kody:@kentcdodds/bugbot using { prUrl } or { owner, repo, prNumber }. Do not pass a GitHub account.

import triggerBugbot from 'kody:@kentcdodds/bugbot'

await triggerBugbot({ prUrl })

CodeRabbit: if it is rate-limited, errored, or otherwise unavailable, do not wait on it for low/medium risk — proceed with Bugbot + CI. Only wait on CodeRabbit when the change is high risk (or the user explicitly asks).

Loop

1. Mark ready — kody:@kentcdodds/github/pr/set-review-status { prUrl, status: 'ready' } (or owner/repo/prNumber).
2. Wait for CI — gh pr checks (or compose loop-on-ci / fix-ci).
3. Fix failures; for medium+, wait on AI reviewer(s) (Bugbot first; see above for CodeRabbit) and address valid feedback. Rebase only when actually unmergeable. For medium+, also run npm run control-kody -- preview (or npm run preview:manual-test) as the seeded user with data for this change (--request / session cookie; see control-kody and preview-manual-test). After merge, npm run control-kody -- health --origin https://kody.codes --sha <merge>.
4. Green + (medium+: valid feedback cleared) → break.
5. Push → repeat.

Gates ≠ CI

Blocked on soak / parity CHECK / calendar gate → end the run and schedule a wake (execute + workflows.create({ runAt, idempotencyKey }) calling createRun). Don't sleep-poll or code-thrash an intentional time window. Leftovers that wait on that gate need a GitHub issue (Cleanup: title); see cleanup-after-migrations.

Batch related expand steps into fewer PRs when risk posture allows.

Merge / deploy

When policy + risk allow: squash-merge via kody:@kentcdodds/github/pr/merge { prUrl, mergeMethod: 'squash' }, watch deploy. Useful: pr/get-checks, request, graphql on the same package.

Done → Discord

Always summarize (merged or not) via kody:@kentcdodds/discord/send-shipped-pr. That export formats kind ship-pr, fetches Cursor token cost from the usage API, and includes the model you pass. Never invent a dollar figure or a model id. When the work deployed user-visible pages, put clickable links to those pages in extras (see below) so Kent can open the live result from Discord.

title (required): a human headline of the change itself so the Discord post is glanceable. Example: OpenAPI spec fetches now count against daily quota. Do not use ship owner/repo#N as the title — repo, PR, and agent already appear as links.

difficulty (required): 'Easy' | 'Medium' | 'Hard'. Always pass it. Distinct from Risk (merge authority). Easy = small/localized; Medium = several files or real behavior change; Hard = architecture, migrations, subtle correctness, or wide blast radius. The Discord export renders this on its own line.

agentId (required):

• In a Cursor Cloud Agent VM, read it from the metadata socket: curl -fsS --unix-socket "${CURSOR_AGENT_SOCKET:-/run/cursor/api.sock}" http://cursor-agent/v1/meta-data/agent/id
• Otherwise pass the bc- id from the agent URL you were launched as (https://cursor.com/agents/{id}).

model (deterministic — never infer):

• In a Cursor Cloud Agent VM, read the model that served this turn from the metadata socket (not a guess from writing style): curl -fsS --unix-socket "${CURSOR_AGENT_SOCKET:-/run/cursor/api.sock}" http://cursor-agent/v1/meta-data/turn/model If you selected Auto, this is the concrete model that served, not Auto. See https://cursor.com/docs/cloud-agent/metadata
• Outside a managed VM, pass the model.id used at create/launch if you still have that record.
• If the key is missing (404 / empty), omit model so the export posts Model pending — do not invent one.

Deployed pages (when the work actually shipped UI/routes):

• If merge+deploy produced live pages people can click, put those URLs in extras as Discord markdown links (one per line item). Prefer the specific routes/pages that changed, not only the site root.
• Skip this when nothing user-visible deployed (library-only, docs-only with no hosted page, parked/blocked, or deploy not reached).
• Do not invent URLs. Only include pages you verified or that the deploy output / PR preview / production URL clearly maps to.

import sendShippedPr from 'kody:@kentcdodds/discord/send-shipped-pr'

export default async function main() {
	return sendShippedPr({
		agentId: 'bc-…', // metadata socket or launch URL
		model: 'grok-4.6', // metadata turn/model (deterministic)
		title: 'OpenAPI spec fetches now count against daily quota',
		difficulty: 'Medium',
		status: 'Shipped', // or Parked / Blocked
		summary: 'One-screen what shipped and why it is done.',
		prUrl: 'https://github.com/owner/repo/pull/123',
		repo: 'owner/repo',
		extras: [
			'CI green',
			'[Account](https://kody.codes/account)', // only when that page actually deployed
			'[Connect](https://kody.codes/connect)',
		],
	})
}