@djoeyg/pi-review0.1.0
request_review / steer_review: background code review by an independent, steerable reviewer model
Untested on omp
This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.
Reports 0
No reports yet.
Readme
pi-review
Gives the agent a background code reviewer it can steer.
request_reviewstarts a reviewer and returns a review id (R1,R2, …) immediately. The reviewer is a headless pi on its own model, with fresh context and read-only tools (read,grep,find,ls,bash). When it finishes, its verdict and findings arrive in the session as a message, which starts a turn if the agent is idle.steer_reviewmessages a review. While it runs, the message redirects it. After it finishes, the message starts another round in the same reviewer context, so the agent can dispute findings or ask for a re-review after fixing them.
Reviewers live until the session ends. The footer shows which are running.
/review off removes both tools from the agent and stops running reviewers; /review on restores them, and /review toggles. The choice persists in ~/.pi/agent/review.json as "enabled".
The reviewer prompt (src/reviewer.md) asks for a strict maintainability review: correctness first, then structural simplification, ownership boundaries, and illegal state.
Models
The agent picks the reviewer model per request. Offer it a menu in ~/.pi/agent/review.json: keys are provider/id, optionally with a :thinking suffix, and values say when to pick each. The first entry is the default.
{
"models": {
"openai/gpt-6.1-sol:high": "Default. Deep structural and correctness review.",
"claude-bridge/claude-opus-5-5:high": "Second opinion from another model family; large or cross-cutting changes.",
"openai/gpt-6-luna:medium": "Small, low-risk diffs where speed matters."
}
}
Without a menu, the agent may name any model that has credentials, and defaults to its own. An optional "promptFile" (absolute path) replaces the bundled reviewer prompt.
Install
pi install npm:@djoeyg/pi-review