Agy Review
Code review of your local git changes by Google Antigravity (agy), with read-only repo access and no permission prompts.
Free
Works with the AI Agents you already use
agy-review
Example session with this skill installed
Use agy-review to prepare this repository for a safe release.
- Check that Google Antigravity and the read-only review guard are correctly configured.
- Independently review the current Git changes for correctness, reliability, and security vulnerabilities.
- Verify every finding against the actual source code. Reject unsupported or hallucinated findings.
- Return a prioritized report with file locations, CWE classifications where relevant, evidence, recommended fixes, and a release verdict.
- Check whether Graphify is installed and configured.
- With my approval, configure Graphify to refresh the repository knowledge graph through Antigravity in the background after commits, merges, branch switches, and rebases.
- Use incremental updates and avoid consuming Claude Code or Codex agent tokens for Graphify's LLM operations.
- Show the final review verdict, Graphify configuration, and verification commands.
Do not modify application source code or silently install dependencies.
- Read your context and instructions
- Compiled the agy-review
agy-review — Repository Review & Graphify Setup
1. Environment Verification
Start by checking the Antigravity environment and read-only review guard.
bun plugins/agy-review/scripts/agy-review.mjs --check
bun plugins/agy-review/scripts/agy-review.mjs --verify
The verification must establish that read operations work and prohibited write operations are blocked before agentic review proceeds.
A failed verification requires investigation before continuing.
2. Independent Security Review
Run the security review against the current changes:
bun plugins/agy-review/scripts/agy-review.mjs \
--security \
--format json \
--fail-on high \
--output security-review.json
Evidence Verification
| Model claim | Verification handling |
|---|---|
| Supporting code found | Eligible for a verified finding |
| Incorrect line number | Corrected against source |
| Evidence not found | Rejected |
| Speculative finding | Reported separately |
| Insufficient coverage | Inconclusive, not approved |
Security Report Structure
Every verified finding can include
- Severity and priority
- CWE classification
- Source file and line
- Exact supporting evidence
- Explanation and risk
- Recommended remediation
- Confidence and grounding status
CI Verdict
| Exit code | Meaning |
|---|---|
0 | Successful review, no failing gate |
1 | Error or unparseable model response |
2 | Invalid usage or missing prerequisite |
3 | Working tree changed during agentic review |
4 | Inconclusive review |
10 | Severity gate failed |
The application code is not automatically fixed by the reviewer.
3. Graphify Prerequisite Check
Run
bun plugins/agy-review/scripts/agy-review.mjs --graphify doctor
Check for
- Signed-in Antigravity CLI
- Graphify installation
- Python OpenAI client availability
- Existing knowledge graph
- Git hook configuration
If any dependency is missing, report the required action and obtain approval before making environment changes.
4. Configure Graphify
After approval
bun plugins/agy-review/scripts/agy-review.mjs --graphify setup --yes
This configures the Antigravity-backed update workflow and installs background Git hooks.
Existing unrelated hook content is preserved. By default, Graphify's own hooks are replaced to avoid competing graph writers.
If no knowledge graph exists, initialize it once:
bun plugins/agy-review/scripts/agy-review.mjs --graphify update --full
5. Automatic Knowledge Graph Maintenance
After setup, the following events trigger background graph refreshes:
| Git event | Action |
|---|---|
| Commit | Background update |
| Merge | Background update |
| Branch checkout | Background update on branch switch |
| Rebase or amend | Background update |
Graphify parses source code locally and uses its cache to identify changed documentation.
The Antigravity model handles semantic extraction and community naming through the temporary local bridge.
The Git command does not wait for the graph update to finish.
6. Inspect Graphify Status
Run
bun plugins/agy-review/scripts/agy-review.mjs --graphify status
Inspect the configured hooks, latest result, model calls, token usage, and any errors.
The agent may continue working with the current graph while an update is running.
7. Remediation and Follow-Up
For verified high-priority findings
- Inspect the cited code and validate the recommended fix.
- Apply approved changes using a separate development workflow.
- Repeat the independent review.
- Compare results with the previous JSON report.
- Resolve critical issues before release.
- Recheck Graphify status after subsequent Git operations.
Final Deliverables
- Evidence-verified code and security report
- Explicit release verdict
- List of rejected or inconclusive findings
- CI-compatible JSON report
- Graphify prerequisites and setup status
- Background Git hook configuration
- Knowledge graph status and troubleshooting information
Outcome: Independent, verifiable review results and a Graphify workflow that moves graph-update LLM processing away from Claude/Codex agent sessions to Antigravity.
Actual findings, setup state, and execution metrics depend on the repository and environment.
Connects securely to your tools. The creator never sees your data.
About this skill
agy-review — Evidence-Verified AI Code Review & Automatic Graphify
Two powerful workflows. One toolkit. Powered by Google Antigravity.
Review your code with independently verified AI findings, detect security vulnerabilities, and keep your repository's knowledge graph automatically updated—without wasting your primary coding agent's tokens on graph maintenance.
agy-review is an open-source developer toolkit connecting Claude Code, OpenAI Codex, and terminal workflows to Google Antigravity (agy).
It combines independent code and security reviews, multi-model analysis, task delegation, translation, CI/CD automation, and Graphify integration.
🔍 1. Evidence-Verified Code Review
AI code review that has to show its work.
Every reported finding must include an exact quotation from the affected source code.
agy-review automatically:
- Checks quoted evidence against real repository files.
- Corrects inaccurate line references.
- Rejects findings whose evidence cannot be located.
- Separates speculative or unsupported findings.
- Calculates the final verdict from verified findings.
- Generates Markdown or structured JSON reports.
Review uncommitted changes, staged files, commits, branches, individual functions, classes, line ranges, or pasted snippets.
Supported verdicts include approve, approve_with_comments, request_changes, and inconclusive.
An inconclusive review is never silently treated as approval.
🛡️ 2. Security Review & Vulnerability Analysis
Go beyond ordinary code review with security-focused analysis.
Capabilities include:
- OWASP and CWE mapping
- Exploitability analysis
- Concrete attack scenarios
- Untrusted-input data-flow reasoning
- Secret and dangerous-pattern scanning
- Caller and callee context analysis
- Prioritized, evidence-backed remediation
Run security reviews on authentication, database queries, file handling, networking, cryptography, and other security-sensitive code.
🧠 3. Automatic Graphify Knowledge Graph Updates
Stop spending Claude Code and Codex agent tokens refreshing your knowledge graph.
agy-review integrates with Graphify to maintain a repository's knowledge graph automatically through Google Antigravity.
Graphify processes code locally using Tree-sitter and maintains its graph, cache, clustering, and reports.
agy-review redirects Graphify's LLM-dependent operations to Antigravity:
- Semantic extraction from changed documents
- Entity and relationship extraction
- Community naming
This avoids using your active Claude Code or Codex session for those graph updates.
Automatic Git Hooks
Graphify updates run in the background after:
- Git commits
- Git merges
- Branch checkouts
- Git rebases and amends
Git operations return immediately without waiting for the graph refresh.
Incremental Processing
Graphify's existing cache detects unchanged documents and avoids sending them repeatedly to the model.
Local Antigravity Bridge
A temporary, loopback-only OpenAI-compatible endpoint connects Graphify to agy.
- No separate LLM API key required
- No Ollama required
- Configurable Antigravity model
- Configurable reasoning effort and concurrency
- Temporary bearer-token authentication
- Recognized secrets redacted before model submission
Graphify still uses your Antigravity quota. The savings concern Claude/Codex agent tokens, not all model usage.
Documented Graphify Benchmark
On the author's small Windows 11 test repository:
| Metric | Observed |
|---|---|
| Git commit return | 0.3 seconds |
| Background graph update | 22 seconds |
| Antigravity calls | 1 |
| Claude/Codex agent tokens | 0 |
| Graph nodes | 10 → 14 |
| Graph edges | 9 → 12 |
Results depend on repository size, changed documents, model choice, and available quota.
🤖 4. Multi-Model Review & Agent Delegation
Run reviews using available Antigravity models.
- Choose models and reasoning effort.
- Cross-check findings across multiple models.
- Track agreement between independent reviews.
- Delegate self-contained tasks through Antigravity.
- Integrate reviewer and worker sub-agents into larger workflows.
🌍 5. Translation & Localization
Translate documentation and other content through Antigravity with configurable:
- Target language
- Tone and audience
- Glossary and terminology
- Structural preservation checks
Verify that code snippets, URLs, placeholders, and headings survive translation.
⚙️ 6. CI/CD & Automated Quality Gates
Bring evidence-based reviews into automated development workflows.
- Structured JSON reports
- Stable finding IDs
- Severity-based exit codes
- Comparison of new, persisting, and resolved findings
- Automated review-and-fix loops
- Explicit handling of inconclusive results
A verified high-severity finding can trigger a failing CI gate.
🔒 7. Security & Permissions
Static review modes run without giving the AI model filesystem tools.
Agentic reviews require a deny-by-default read-only guard, with verification before use and working-tree change detection.
Graphify's Antigravity model receives text and returns text; it does not need filesystem write or execution tools. Graphify itself performs local processing and writes graph artifacts.
These protections are not an operating-system sandbox.
Source code and documents submitted to Antigravity are processed by Google's models. Only use the toolkit with material you are authorized to share.
🚀 Requirements
Core toolkit:
- Google Antigravity CLI (
agy), authenticated - Bun or Node.js 18.18+
- Git for repository-based operations
Additional Graphify requirements:
- Graphify (
graphifyy) - Python
openaipackage in Graphify's environment
The Graphify integration is optional. Code and security reviews do not require Graphify.
💡 Who Is This For?
- Software engineers using AI coding agents
- Claude Code and Codex users
- Security-conscious developers
- Open-source maintainers
- Teams maintaining large repositories
- Developers building persistent code knowledge graphs
- Engineers automating review and CI/CD workflows
Open Source & Documentation
MIT Licensed — Free and Open Source
GitHub: https://github.com/Ealireza/agy-review
Official Website: https://ealireza.github.io/agy-review/
Graphify Documentation: https://github.com/Ealireza/agy-review/blob/main/docs/GRAPHIFY.md
Review your code. Verify the evidence. Keep your knowledge graph fresh. Let Antigravity handle the work.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 8 days ago
- Free with an Agensi account
- Software
Git
Python
Node.js