Review pull requests for bugs and breakage

This agent reads a pull request diff and posts one review comment. It reports bugs, breaking side effects, developer workflow failures, and unintended feature exposure in added or modified code. Style nits stay out.

This template is one model turn: read the diff, then post one comment.

Scaffold

agent-sdk init ./thermo-review --template thermo-review
cd thermo-review

Deploy

Sign in once, then deploy from a pushed checkout:

agent-sdk login
agent-sdk deploy --cursor-events-repo acme/api

Repeat --cursor-events-repo for each repo. Every repo must share one owner. The deployment watches those repos through your team's Cursor GitHub connection and reviews each opened PR. No webhook or local server.

Change repos later without redeploying:

agent-sdk deployment update thermo-review --event-repo acme/api --event-repo acme/web

Try it locally

Skip the deploy while you iterate. Start the relay for one repo:

agent-sdk dev --cursor-events --repo acme/api

Open a PR, or replay a known one without pushing:

agent-sdk github replay https://github.com/acme/api/pull/12 --dir .

replay needs pull access. Draft PRs are ignored. The channel writes pr/diff.patch, then starts a model turn. The model calls post_findings once. Chat replies stay in the playground. post_findings writes the GitHub comment.

Connect GitHub in Cursor first (Settings or cursor.com/dashboard) so the relay and API calls use your account. See the GitHub guide.

Evals

Use the two smoke cases, one auth-bypass fixture and one docs-only change, as a local exercise. They need no GitHub access.

agent-sdk eval

Eval sessions skip the GitHub write.

What to edit

• bot/instructions.md: what counts as a finding and the comment shape
• evals/review.eval.ts: smoke cases