Stats
Actions
Available In
Tags
Plugin
pr-review-journal
Per-PR review journal: parses verdict-block disposition replies, infers verdicts from historical thread prose, and records the result as auditable JSON. Ships with paired skills for the orchestrator and reviewer sides of a review thread.
What's Inside
Skills2
pr-review-triage
/pr-review-triage
Address findings from automated PR reviewers (CodeRabbit, Codex, GitHub Copilot review, in-house bots) and human reviewers with a documented reasoning trace and a parseable verdict for each thread. Use whenever the user mentions handling PR comments, responding to a reviewer, triaging review findings, addressing CR/Codex/Copilot feedback, resolving review threads, or "fixing" things in a PR after a review pass. Triggers on phrases like "address findings on PR
pr-reviewer
/pr-reviewer
Review a pull request the way a careful engineer would — look for architectural risk, ADR conflicts, blast radius, and root causes rather than style nits that linters and CR already catch. Use whenever the user mentions reviewing a PR, asking for a second opinion, running `@claude review`, doing a pre-merge audit, or wanting an independent pass on a PR that's already been through CodeRabbit / Codex / Copilot review. Triggers on phrases like "review PR
README
pr-review-journal
A per-PR review journal that records every reviewer recommendation and the project's verdict on it, paired with two Claude Code skills that encode the reviewer-side and orchestrator-side discipline.
Designed to work with any GitHub-app reviewer — CodeRabbit, Codex (via
chatgpt-codex-connector), Claude PR review (via agentic-ops/review.yml),
GitHub Copilot review, Greptile, Qodo, or an in-house bot — by treating
reviewer behaviour as config, not code.
The tool is Python stdlib + shell, depends only on python3 and gh. Runs in
any repo where gh is authenticated; no install required.
What's in the box
| Path | Purpose |
|---|---|
tools/review-journal/review_journal.py | The journal tool (1 Python module, stdlib only) |
tools/review-journal/sync-pr.sh, extract-pr.sh | Shell wrappers around the tool |
tools/review-journal/install/ci-check.yml | Drop-in GitHub Actions workflow (warning mode) |
tools/review-journal/tests/ | 35 shell tests (captured fixtures + goldens; no network) |
skills/pr-review-triage/ | Claude Code skill — orchestrator side (how to dispose of findings) |
skills/pr-reviewer/ | Claude Code skill — reviewer side (how to issue findings cleanly) |
.claude-plugin/plugin.json | Claude Code plugin manifest |
Install as a Claude Code plugin
/plugin marketplace add loganrooks/pr-review-journal
/plugin install pr-review-journal@loganrooks/pr-review-journal
This installs both skills into your Claude Code setup. The tool itself
(tools/review-journal/) is available at ${CLAUDE_PLUGIN_ROOT}/tools/review-journal/
inside skill instructions; for use outside Claude Code (CI, ad-hoc) you'll also
want a vendored copy in your repo — see below.
Install the tool into a consumer repo
The tool is portable: copy the directory + a config file.
# 1. Vendor the tool (pin to a tag for reproducibility).
git clone --depth 1 --branch v0.1.0 https://github.com/loganrooks/pr-review-journal.git /tmp/prj
cp -r /tmp/prj/tools/review-journal otherrepo/tools/
# 2. Create a minimal .review-journal.json at the consumer repo's root.
cat > otherrepo/.review-journal.json <<'EOF'
{
"enforcement_mode": "warning",
"reviewers": ["coderabbitai", "chatgpt-codex-connector"],
"journal_dir": ".planning/review-journal"
}
EOF
# 3. (Optional) Install the CI workflow snippet.
cp /tmp/prj/tools/review-journal/install/ci-check.yml otherrepo/.github/workflows/review-journal.yml
A submodule, subtree, or one-liner installer would all also work; pick what fits your repo's vendoring conventions.
The verdict-block discipline
Every reply on a review thread starts with a fenced block:
```review-verdict
verdict: ACCEPTED_MODIFIED
commit: 14b240b
finding_category: source-resolution-correctness
reviewer: chatgpt-codex-connector
notes: PID-first match; bundle fallback kept for relaunch-between-pick-and-start.
```
Verdict vocabulary (eight values):
ACCEPTED,ACCEPTED_MODIFIED— finding accepted;commitrequiredDEFERRED— accept but defer the action;notesrequiredREJECTED_FALSE_POSITIVE,REJECTED_BAD_FIT,REJECTED_REGRESSION—notesrequiredOBSOLETE— finding no longer applies;commitrequiredDUPLICATE— same as another thread
Full per-verdict semantics and config reference: see
tools/review-journal/README.md.
Quick start
# Sync the current verdict state (parses existing blocks; does not infer).
bash tools/review-journal/sync-pr.sh 7 --repo owner/repo
# Backfill verdicts for threads that pre-date the discipline.
bash tools/review-journal/extract-pr.sh 7 --repo owner/repo
Output lands at <journal_dir>/pr-7.json + an index.json summary across
all PRs.
Tests
bash tools/review-journal/tests/run-tests.sh
35 tests covering block parsing, sync schema, inference, portability, config, profile flexibility, extensibility, provenance, and robustness. Tests use captured fixtures and golden expected outputs; no network.
History
Extracted from loganrooks/tap-n-filter,
where it started as a local fix for a workflow problem: rejecting a reviewer's
suggestion (because it conflicts with a local convention CR's training data
doesn't know about) leaves no durable record of why. The journal mechanises
the "Reasoning over acceptance" principle.
The standalone-repo extraction (this repo) also fixed several upstream bugs identified by Codex review on the first port consumer; see the changelog for v0.1.0.
License
MIT
Stats
Version0.1.0
LanguageShell
Stars0
MaintenanceExcellent
LicenseMIT
Last Commit
Added
Available In
Tags
Categories
