Skip to content
Back to skills

claude-to-agy

ASecurity

Offloads heavy, token-intensive reasoning and search tasks to Antigravity CLI (agy) via MCP. MANDATORY delegation for grep, git diff, git log, large files, multi-file analysis.

  • 12 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added May 26, 2026
ai-agentspythonbashgitdatabasedocumentation

Works with

  • claude code
  • terminal
  • cli
  • mcp

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned May 27, 2026

npx -y skills add rauls-kjarners/claude-to-agy --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of claude-to-agy?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for claude-to-agy
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/rauls-kjarners-claude-to-agy/badge)](https://www.skillsdirectory.com/skills/rauls-kjarners-claude-to-agy)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: claude-to-agy
description: Offloads heavy, token-intensive reasoning and search tasks to Antigravity CLI (agy) via MCP. MANDATORY delegation for grep, git diff, git log, large files, multi-file analysis.
---

# Skill: claude-to-agy

This workspace is equipped with a custom MCP bridge that connects Claude to the local `agy` CLI. It enables Claude to safely delegate massive reasoning tasks, huge file reads, and web searches without exhausting its own context window.

## Terminal Command Delegation - MANDATORY

**BEFORE** running ANY of these commands in a terminal, you MUST use `delegate_to_agy` instead:

- `grep` (any usage, including recursive) - output can be massive and needs analysis
- `git diff` (any usage) - diffs can span thousands of lines across many files
- `git log` (any usage) - history can contain thousands of commits with full messages

**NEVER** run these commands directly. No exceptions.

## When to Use `delegate_to_agy`
You should automatically use this tool whenever you encounter:
1. **Large files:** Analyzing log files, database dumps, or any file >200 lines. When in doubt, delegate.
2. **Massive context:** Trying to process more than 3 files at once.
3. **Heavy search tools:** Needing to perform `git log`, `git diff`, or `grep`.
4. **Web/external knowledge:** Web searches and documentation lookups.
5. **Adversarial review / plan critique:** Always delegate.

## How to Delegate
- **Always pass `cwd`** with your current working directory (absolute path) so agy knows where the project is.
- Formulate a clear, detailed `prompt` explaining exactly what needs to be found, analyzed, or searched.
- Optionally pass relevant file paths in the `files` array.
- Await the JSON response and use the summary/data provided by agy to fulfill the user's request.

## Configuration
The bridge supports environment variables for tuning:
- `AGY_CONNECT_TIMEOUT` - subprocess start timeout in seconds (default: `60`)
- `AGY_TOTAL_TIMEOUT` - total execution timeout in seconds (default: `600`)

## Setup Instructions (for users)
If the tool is not already active in your Claude Code environment, run the following command to register it:

```bash
claude mcp add -s user claude-to-agy python3 ~/.claude-to-agy/src/bridge.py
```

## Subagent Enforcement (Hooks)

Subagents do **not** read skill/CLAUDE.md rules and will run `grep`, `git diff`, etc. directly. Add the following PreToolUse hook to `~/.claude/settings.json` to block banned commands for **all** agents:

```json
{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "python3 ~/.claude-to-agy/src/hooks/block-direct-commands.py",
            "onError": "block"
          }
        ]
      }
    ]
  }
}
```

Files in this skill

  • CLAUDE.md2.5 KB
  • SKILL.md2.7 KB
  • pytest.ini30 B
  • src/bridge.py7.6 KB
  • tests/test_bridge.py2.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.