Skip to content

review-e

Scalable review worker for the engineering rig.

review-e is the structural counterpart to dev-e. It does not implement features. It reviews pull requests independently against the issue, risk policy, and acceptance criteria, then emits review outcomes back to Conductor-E.

Role

  • poll for review assignments
  • fetch PR state and review context
  • run an independent review pass
  • approve routine PRs
  • dispute risky or incorrect PRs
  • escalate human-gated changes

This repo is designed to scale horizontally. Multiple review-e workers can poll for assignments using distinct agent IDs, just like dev-e.

Current Model

The existing rig behavior described in conductor-e is:

  • Review-E reviews every PR
  • Review-E is separate from Dev-E
  • Review approval is represented as REVIEW_PASSED
  • disputed or guarded PRs should block automatic progression and trigger escalation

This repo encodes that behavior as a worker-style service rather than a single fixed bot.

One coordinator gap still exists: Conductor-E does not yet expose a dedicated review-assignment claim endpoint, so safe horizontal review claiming is not fully implemented yet. The app logic is worker-shaped and ready for that API.

Structure

src/ReviewE.Core/      Domain, ports, and review use case
src/ReviewE.Worker/    Hosted worker that polls for review assignments
tests/ReviewE.Core.Tests/
docs/
deploy/
Dockerfile

Next Steps

  • improve review heuristics and prompt policy pack
  • add integration tests against GitHub review flows

Runtime Requirements

review-e now expects these secrets at runtime:

  • github-token
  • anthropic-api-key
  • discord-bot-token

Discord posting uses the PR's stored discord-review-msg-id metadata comment, matching the older review thread workflow.