| name | prj-review |
| description | PR Jangler full-PR code reviewer. Use when the orchestrator dispatches the review phase for a PR or when the user runs 'prj-review' manually to produce a structured review.md from the PR's diff and changed files. |
prj-review
Overview
This skill performs a full code review of one open PR on the prj_repo. Deterministic plumbing (fetch the PR diff, fetch changed-file contents, look up project conventions in Hindsight, validate finding schemas, persist review.md, optionally post a GitHub Comment-type review) lives in Python scripts. The review itself is LLM-driven: you read the diff and changed files, produce findings, run a self-check pass, then hand the validated findings to the scripts.
Act as a careful senior reviewer. Anchor every finding to a specific file and line. Prefer signal over volume.
Conventions
- Bare paths (e.g.
scripts/run.py) resolve from the skill root.
{project-root}/... resolves from the project working directory.
- Configuration variables live under
[modules.prj] in {project-root}/_bmad/config.toml.
On Activation
The orchestrator calls this skill with a target PR number. The full review proceeds in three phases.
Phase 1: Fetch context (script)
python3 scripts/run.py --pr <pr_number> --fetch-only
This emits a JSON document to stdout containing:
pr: {number, title, body, base_ref, head_ref, author, acceptance_criteria}
diff: full unified diff (string)
files: list of {path, status, content} for each non-binary changed file
- : Hindsight bank excerpts relevant to this repo (best-effort; empty list if Hindsight is unreachable)