hughcube

AI Hook — AI skill for Claude Code

AI community

Ultra-fast AI-agent hook dispatcher & autonomous rule engine (Rust + embedded QuickJS).

How to install AI Hook

This entry records only its repository, not the path inside it, so there is no exact command to give. Open hughcube/ai-hook and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Hook does

Ultra-fast AI-agent hook dispatcher & autonomous rule engine (Rust + embedded QuickJS). Write guardrail rules once, run them across Claude Code, Google Antigravity, CodeBuddy and OpenAI Codex. Sandboxed, zero-dependency, ~2-3ms per tool call.

Alternatives in AI

Pixel Agents — Compressed Reference — VS Code extension with embedded React webview: pixel art office where AI agents (Claude Code terminals) are an 6.5k ★
Cc Mini — Ultra-light Harness scaffolding for AI agents, a mini version of claude code 825 ★
Embeddedskills — An open-source collection of embedded development and debugging skills for Claude Code, Copilot, TRAE, and oth 593 ★

README

ai-hook — Unified Multi-Agent Security Interceptor & Autonomous Rule Dispatcher

[![CI](https://github.com/hughcube/ai-hook/actions/workflows/ci.yml/badge.svg)](https://github.com/hughcube/ai-hook/actions/workflows/ci.yml) [![Release](https://img.shields.io/github/v/release/hughcube/ai-hook)](https://github.com/hughcube/ai-hook/releases) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

**It exists for two things: hiding complexity, and making rules pleasant to write.** **① Hide the complexity** — hook-protocol differences across agents, platform quirks, and subprocess/context-fetching plumbing are all wrapped into one native binary: the same rules are written once and run everywhere, with no per-agent, per-platform script matrix. On Windows it also collapses the old one-process-per-fetch stutter into a single process with millisecond-scale in-process evaluation. **② Better rule authoring** — a rule is just plain JS: a rich `ctx` context, native microsecond `sys` primitives, one uniform deny/ask/inject decision protocol with desktop/terminal interaction, plus `list`/`test`/`bench`/`tutorial` tooling — instead of hand-written shell glue and cross-platform trial and error. Zero variable pollution, a single-process closed loop, and extreme speed all fall out of this design. Hooks for **Claude Code**, **Google Antigravity**, **CodeBuddy / WorkBuddy**, **OpenAI Codex**, **Gemini CLI**, **OpenCode** — one rule set, every agent, on Windows / macOS / Linux.

**OpenCode note:** OpenCode has no out-of-process hook protocol of its own — it loads in-process JS/TS plugins (see [opencode.ai/docs/plugins](https://opencode.ai/docs/plugins/)). ai-hook reaches it through the community bridge [`magarcia/opencode-claude-hooks`](https://github.com/magarcia/opencode-claude-hooks), which forwards Claude-Code-shaped envelopes and sets `OPENCODE_COMPAT=1`. Install the bridge (or use the in-process plugin appro