Browse agent skills

Run Claude Code, Codex and other ACP agents inside Margin, with your own subscriptions and settings

Pack any git repository into clean, security-sanitized LLM prompt context with token counting.

Vigyl engine: Telegram bot, LLM supervisor, sandboxed self-healing command runner.

Prep a ticket: the agent reads it, works out the fix, and writes the plan, risk, effort and priority back onto the ticket without building anything. npx installs it into .claude/skills.

Provider-agnostic multi-AI orchestration platform: coordinates CLI agents, AI providers, memory, roles, reports and planning into one execution system.

TF Design — the local design workspace CLI: headless design runs (decks, posters, sites) through the tfdesign daemon, delegable from tfcode.

TypeScript SDK for drive9 — an agent-native filesystem with semantic search

Pi coding-agent extension that recovers truncated turns, failed streams, empty responses, and terminal tool-call tails through public session boundaries.

On-device OCR for Capacitor with word-level boxes: Apple Vision on iOS, ML Kit on Android

Production-grade OCR/ANPR license plate text cleaner, character confusion disambiguation (0/O, 1/I, 8/B), 2-line combiner, and fuzzy matcher.

Validator and linter for Arazzo documents: semantic validation, semantic linting, and opt-in JSON Schema validation.

Amtal inference SDK for JavaScript and TypeScript.

YouTube transcript tools for the Vercel AI SDK and Mastra: get YouTube video transcripts with timestamps, search YouTube, and list channel videos, via the GetYouTubeTranscript API.

YouTube transcript API for Node.js and TypeScript: get YouTube video transcripts, captions and subtitles with timestamps, search YouTube, and list channel and playlist videos. No proxies or headless browser.

Agent skill for building and reviewing UI with Echo DS: component selection, composition, tokens, and cited design reviews

Feynman-style research workflows for DeepSeek Harness behind one /feynman command (deepresearch, review, audit, review-loop, …)

/cron for DeepSeek Harness: run a prompt on a cron or plain-English schedule, parsed in code, with the model as the last resort.

Multi-agent role framework for Pi. Groups of specialist agents (coding, video, etc.) bound to per-role models, dispatched via pi-subagents. Profile-driven, gate-verified, language-agnostic.

The PopCraft command line: edit .popcraft files headlessly, call the REST API, publish to the marketplace and manage the desktop app's local plugins, themes, fonts, skills and agents.

Japanese localization pack for the Pi coding agent: Japanese system prompt rules, Japanese UI labels, and Japanese design-doc templates.

Typed, cache-aware client for the BackGold headless Content API (/api/content/v1)

TrustRail Guardian for OpenClaw: a stop button on your phone for your agent. Stops at the agent's next tool call; a tool already running will finish.

Unified LLM API with automatic model discovery and provider configuration

Kaman's chat thread for React — messages, streaming, tool calls and citations.

Framework-free chat state: stream frames, turns and the pair protocol.

twitterapi.io-backed X/Twitter search extension for pi coding agent

A small headless, provider-neutral tool-calling agent loop

Installable Orb bundle. Inserts the Computer Use preset and the floating-ball host.

Support for Agent Plugins: vendor-neutral manifest types and validators, installation, activation, digests and trusted contribution loading. Every Jini agent plugin ships here as its own subpath (@jini-ai/agent-plugins/<id>); resolving one never loads ano

Real-browser tool for DeepSeek Harness agents: drive Chromium/Firefox/WebKit through Playwright (navigate, snapshot, observe, extract, act, wait, frames, eval, console, screenshots, tabs, storage, emulation, clipboard, network mocking, tracing).