PR Risk Review

LLM pull request risk review on GitHub — a Claude Managed Agent assesses each PR, upserts a PR comment, optionally requests reviewers, and notifies Discord.

Flow

1. Assess PR Risk (claude.runAgent) — Claude Managed Agent clones the repo, reviews the diff, returns structured JSON
2. Format PR review (runnerJS, tiny) — parse JSON, build comment body and Discord message
3. Create/Update PR comment — GitHub integration (posts as the integration bot)
4. Record check — saves results to the console table
5. Discord review posted — notification on completed review

Prerequisites

• Claude integration with API key on SuperPlane
• A Managed Agent and Environment in Anthropic (for checkout + review)
• GitHub integration on triggers, reviewers, and comment nodes
• GITHUB_TOKEN secret (app-codeowners) injected into the agent session for private repos
• Optional: Discord integration on Discord review posted

Setup

1. Create a Claude Managed Agent + environment that can clone GitHub repos and review PR diffs.
2. Connect Claude and bind it on Assess PR Risk.
3. Connect GitHub and bind the integration on trigger, reviewer, and comment nodes.
4. Add GITHUB_TOKEN to the app-codeowners secret (used by the agent via session secret injection).
5. Optional: Connect Discord on Discord review posted and pick a channel.

Checks run for PRs targeting main or master only (manual checks always run). Draft PRs are skipped until marked ready for review (or updated with a new push).

Discord

[Fix auth middleware](<https://github.com/acme/api/pull/42>) - alice - Risk 15/100 (low)

Pass / fail

A check passes when Claude returns "approved": true in its JSON review.

Risk score (0–100) and level (very low / low / medium / high / critical) are shown in the PR comment and console.

Source layout

File	Role
scripts/format_review.js	Parse Claude output, build comment fields
canvas.yaml	Embedded workflow

License

MIT