grill

Reviews local branch diffs adversarially, auto-fixes high-confidence findings, and loops until clean.

Updated Jan 15, 2026
One-click install
npx skills add https://github.com/bearded-giant/claude-code-config --skill grill-bearded-giant
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: grill
Source: https://github.com/bearded-giant/claude-code-config/tree/main/skills/grill
Command: npx skills add https://github.com/bearded-giant/claude-code-config --skill grill-bearded-giant

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Shipping code with hidden bugs, security holes, or missing tests is risky, and manual self-review of your own diff often misses issues. This Skill runs an adversarial pre-push review loop on your local branch, scoring every finding by severity and confidence, auto-fixing the safe ones, and flagging the rest before you open a merge request. ## Core Features & Use Cases - Adversarial review loop: Reviews the git diff against a base branch as a skeptical staff engineer, scores findings on severity (1-5) and confidence (0-1), and re-reviews after fixes for up to 5 turns. - Dual-axis auto-fix matrix: Automatically fixes high-confidence severity 2-4 findings while flagging severity-5 hard-stops and low-confidence blockers for human judgment, with a dry-run mode for report-only passes. - Kai CI preflight: Predicts whether Recharge's GitLab CI reviewer would block the MR, including size gates, labeling rules, and a predicted approve/block verdict. - Use Case: Before pushing a feature branch, run /grill to catch a missing migration rollback (sev-5, flagged), auto-fix a boundary comparison bug (sev-4, conf 0.92), and get a final SHIP IT / NEEDS WORK / BLOCK rating written to final.md. ## Quick Start Ask the assistant to grill this branch with the command /grill to run an adversarial review of the current diff before pushing.

Frequently Asked Questions about grill

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an adversarial code review on my local branch before pushing?

Invoke /grill on your local branch to review the diff against the base branch. It scores each finding by severity and confidence, auto-fixes high-confidence items, and loops up to 3 turns, writing results to final.md.

What is the difference between grill and reviewing a posted merge request?

Grill runs locally on committed branch state before an MR exists, editing files directly without staging or committing. For posted GitLab MRs, use kai-review:review-code or kai:review-adversarial instead, since grill never touches remote state.

Can I run the review without applying any automatic fixes?

Yes, pass --dry-run to score and report all findings without editing files. In dry-run mode all auto-fix dispositions become would-fix entries in the run report, and the loop stops after the first run.

How does the severity and confidence scoring decide what gets auto-fixed?

Each finding gets a severity of 1-5 and confidence of 0-1. Severity 3-4 findings auto-fix at confidence 0.85 or higher, severity 2 requires 0.95, severity 5 is always flagged for humans, and severity 1 is report-only.

When will grill refuse to run a review?

It refuses when the branch has uncommitted changes, tests changed but were not run, new behavior lacks tests, or a migration lacks a documented rollback. It also aborts when the Kai size gate exceeds 75 files or 2000 changed lines.

Does grill commit or push my changes after fixing them?

No, grill never stages, commits, pushes, or stashes. It only edits files already in the diff plus new test files, leaving all version control decisions to you after reviewing the fixes.