Cross-engine dispatch layer for AI coding agents. Use when you need to: launch a worker on Codex/Claude/Gemini, recover a timed-out dispatch, steer a running worker mid-flight, or coordinate multi-model work. Trigger on: agent-mux, dispatch, spawn worker, codex worker, profile dispatch, async dispatch, steer agent, recover timeout, multi-engine. --- # agent-mux One CLI, three engines (Codex, Claude, Gemini), one JSON contract. Worker identity lives in prompt files at `~/.agent-mux/prompts/` -- markdown with YAML frontmatter that sets engine, model, effort, timeout, and system prompt. No config files, no role tables. The prompt IS the worker. ## Quick Dispatch Three patterns cover 95% of dispatches. **Profile dispatch** (the standard path -- one flag resolves everything): ```bash agent-mux -P=lifter -C=/repo "Fix the retry logic in src/client/retry.go" 2>/dev/null ``` **Async dispatch** (fire, collect later): ```bash ID=$(agent-mux -P=scout --async -C=/repo "Find deprecated API usages" 2>/dev/null | jq -r .dispatch_id) agent-mux wait --poll 30s "$ID" 2>/dev/null agent-mux result --json "$ID" 2>/dev/null ``` **Structured dispatch via stdin** (canonical machine invocation): ```bash printf '%s' '{"profile":"lifter","prompt":"Implement the fix","cwd":"/repo"}' \ | agent-mux --stdin --async 2>/dev/null ``` Parse stdout JSON. Every result has `status`, `response`, `activity.files_changed`, and `metadata.engine`. Always check `status` first. ## Profile Roster
Reviews PRs or code changes using multiple AI models, synthesizes findings, and interactively applies fixes. Uses GPT 5.2, Gemini 3 Pro, and Opus 4.5 for diverse perspectives.
- 📁 .claude-plugin/
- 📁 .github/
- 📁 agent/
- 📄 .gitignore
- 📄 .nvmrc
- 📄 CHANGELOG.md
Build autonomous AI agents with Claude Agent SDK. TypeScript v0.2.101 | Python v0.1.58.
Create new Claude Code skills for the Karma ecosystem. Use when user says "create a skill", "build a new skill", "scaffold a skill", or "design a skill". Guides through intent capture, SKILL.md writing, validation, and packaging.
Personal memory from real data exports (ChatGPT, Claude, Instagram, Google, etc.). Search and manage memories extracted from the user's data before responding to anything personal — past conversations, people, places, projects, habits, or events.
Review new skills for claude-skills-library standards. Use when adding new skills to verify they meet quality and structure requirements.
- 📄 configuration.md
- 📄 deployment.md
- 📄 extending.md
This guide is for AI agents (Claude, etc.) helping a new user set up Reacher interactively. Walk through each step in order, asking for user input where needed.
[Deprecated — use remember /remember instead; removal planned] Search and summarize past Claude Code work across all projects. Same workflow as remember; kept temporarily for compatibility.
- 📁 instructions/
- 📁 references/
- 📄 SKILL.md
电影级全流程提示词生成。用户给出故事主题，一键生成完整 markdown 文件，包含：角色设定提示词、场景提示词、完整剧本、分镜表、每个分镜的视频生成提示词。支持 Seedance 2.0 / 即梦 / Midjourney / SD。触发词：分镜、提示词、prompt、镜头、脚本、剧本、视频生成、电影风格、Seedance、生成剧本。
Create or update CLAUDE.md and AGENTS.md files following official best practices. Use when asked to create, update, audit, or improve project configuration files for AI agents, or when users mention "CLAUDE.md", "AGENTS.md", "agent config", or "agent instructions".
Open-source AEO monitoring CLI. Track how AI answer engines cite your domain across ChatGPT, Gemini, Claude, and Perplexity.
- 📁 .github/
- 📁 _shared/
- 📁 analyze-results/
- 📄 .gitignore
- 📄 ARCHITECTURE.md
- 📄 CHANGELOG.md
Research automation skills for Claude Code. Full pipeline from idea to submittable paper. Skills: /lit-review, /novelty-check, /experiment, /analyze-results, /write-paper, /research (orchestrator), /setup.
