review

Reviews branch diffs against main for SQL safety, race conditions, and LLM trust boundary issues.

Updated Jan 13, 2026
One-click install
npx skills add https://github.com/psyche911/AntigravityTest --skill review-psyche911
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: review
Source: https://github.com/psyche911/AntigravityTest/tree/main/.claude/skills/gstack/review
Command: npx skills add https://github.com/psyche911/AntigravityTest --skill review-psyche911

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code review often misses structural issues that tests don't catch, such as SQL injection risks, race conditions, and unvalidated LLM output written to databases. This Skill automates a pre-landing review of your branch's diff against main using a structured two-pass checklist. ## Core Features & Use Cases - Two-Pass Review: Runs a CRITICAL pass (SQL & data safety, race conditions, LLM output trust boundary) followed by an INFORMATIONAL pass (dead code, test gaps, magic numbers, view performance). - Structured Findings: Outputs every issue with file:line citations and one-line fixes, classifying them as blocking or non-blocking. - Interactive Fix Workflow: For each critical issue, asks whether to fix it now, acknowledge it, or dismiss it as a false positive, then applies approved fixes. - Use Case: Before merging a feature branch that adds an LLM-generated email field, run the review to catch missing format validation and an N+1 query in the new view code. ## Quick Start Ask the assistant to run the pre-landing review on the current branch's diff against main.

Frequently Asked Questions about review

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

FAQPage Schema
How do I review a git diff before merging a pull request?▼

Run the review workflow on your feature branch. It fetches the latest main, computes the full diff with git diff origin/main, and applies a structured checklist covering SQL safety, concurrency, and LLM trust boundaries, outputting findings with file:line citations.

What code issues does an automated PR review checklist catch?▼

The checklist catches SQL string interpolation, TOCTOU races, N+1 queries, unvalidated LLM output written to databases, conditional side effects, dead code, test gaps, weak entropy sources, and view performance problems like inline styles and O(n*m) lookups.

Can the review modify my code automatically?▼

It is read-only by default. Files are only modified when you explicitly choose "Fix it now" on a critical issue via the interactive prompt. It never commits, pushes, or creates pull requests.

What happens if I run the review on the main branch?▼

The review stops immediately with the message that there is nothing to review. It also stops if git diff origin/main produces no changes, so stale or empty branches produce no output.

Which findings block merging versus being informational?▼

Critical findings—SQL and data safety, race conditions and concurrency, and LLM output trust boundary violations—block shipping. Informational findings like magic numbers, dead code, prompt issues, and test gaps are reported but non-blocking.