Lorenzini — Git skill for Claude Code
Claude Code skills that wait for a third-party PR reviewer and decide whether its verdict actually means pass.
How to install Lorenzini
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Nanako0129/lorenzini and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Lorenzini does
Claude Code skills that wait for a third-party PR reviewer and decide whether its verdict actually means pass.
Alternatives in Git
- Fix PR — by metabase - Fetches and fixes unresolved PR comments by automatically retrieving feedback, addressing review 46.5k ★
- Loki Verify — Run Loki's deterministic PR verification on the current change and summarize the evidence verdict 1.1k ★
- Final Pass — look at the git branch we are on to see what we have been working on 200 ★
README
lorenzini
Claude Code skills that wait for a third-party pull-request reviewer and decide whether its verdict actually means *pass*.
A shark's ampullae of Lorenzini are electroreceptors: they find prey buried in sand, where there is nothing to see. That is what these skills are for. Every reviewer so far has reported findings in places its own count ignores, and every bug in this repository has been the same mistake — treating "I cannot see a problem" as "there is no problem".
What is here
| Skill | Reviewer | Status |
|---|---|---|
| `coderabbit-review-wait` | CodeRabbit (coderabbitai[bot]) |
Active on repositories with 10 stars or more |
| `copilot-review-wait` | GitHub Copilot (copilot-pull-request-reviewer[bot]) |
Active on repositories under 10 stars |
| `codex-review-wait` | Codex (chatgpt-codex-connector) |
Dormant since 2026-09-17, subscription paused |
The CodeRabbit/Copilot split is a vendor constraint, not a preference: CodeRabbit's OSS tier requires a public repository with fewer than ten stars to have its reviews triggered by hand, so those repositories moved to Copilot. Check which applies with `gh api repos/OWNER/NAME -q .stargazers_count` rather than assuming.
Each skill polls until a verdict lands on the current head commit and prints one machine-readable `RESULT=` line. They do not review code; they adjudicate what a reviewer said about it.
Install
Symlink into `~/.claude/skills/`:
git clone git@github.com:Nanako0129/lorenzini.git ~/side-project/lorenzini
for s in codex copilot coderabbit; do
ln -s ~/side-project/lorenzini/$s-review-wait ~/.claude/skills/$s-review-wait
done
Requires `gh` (authenticated) and `jq`.
The one rule
**Never infer a pass from absence.**
A verdict requires an explicit positive completion marker. No marker means keep polling — not "nothing found, therefore clean". Five separate bugs have come from breaking thi
Related Skills
Lane Safety
Safety and hygiene lane. Own redaction and every repository hygiene guard. Nothing derived from a game log rea
Commit Review
Commit workflow for this project's repos — runs an appropriateness review over uncommitted changes (to catch p
Open Contribute
Agent Skills that take you from "I want to contribute to open source" to a submitted pull request — and measur
Finishing A Development Branch
Use when implementation is complete, all tests pass, and you need to decide how to integrate the work - guides
Boot Server
Detect how this project is served, start the dev server in the background, wait until it's ready, and open the
Anvc
Agent-Native Version Control for Claude Code, Codex and Cursor. Stores each attempt your agent makes (what it
Related Agents
Skill Quarantine
Vets a third-party Claude Code skill/plugin (git URL or local path in the sandbox) for security red flags befo
Security Triage
Independently evaluate a single security finding to decide whether it is a real, exploitable vulnerability or
Recruiter Reviewer
Reviews a draft CV as a senior recruiter would — focuses on whether the CV would survive the first 10–15 secon