I research everything agent-related — from harness, communication, memory, and emotion to distilling expertise into skills. master-skill（大师.skill）hands you the fish. sijiao-skill（私教.skill）teaches you to fish.
master-skill（大师.skill）distills the cognition of an entire industry (hands you a fish) → sijiao-skill（私教.skill）distills a path you walk to learn it yourself (teaches you to fish).
| Project | What it does |
|---|---|
| 🎓 master-skill · 大师.skill | Distills an entire industry's cognition into one skill — primary-source canon + representative-figure sub-skills + expert reasoning patterns. |
| 📖 sijiao-skill · 私教.skill | Input a skill you want to learn; it distills a stateful private tutor — 8-track research + learning science, taking you from zero to competent. |
| 💗 pure-love-skill · 纯爱.skill | An AI-native pure-love persona — distills no real person, zero data input. One SKILL.md, zero dependencies. |
| 📜 wenyanwen-skill · 文言文.skill | Have AI reply in Classical Chinese to save 2–3× tokens; a local MCP translates back to plain language at zero cost. |
| 🛡️ account-risk-skill · 反封号.skill | Ban-risk detection for Claude / OpenAI: IP cleanliness, compliance pre-check, appeal-package generation. Compliant use only. |
| Project | What it does |
|---|---|
| 🐚 hermit-agent | Hermit-crab agent harness: one npx line drops a persona-driven, Telegram-connected agent into Claude Code / Codex. Borrows the host's shell and subscription. |
| 🖥️ hermit-ui | Web UI + local gateway for hermit-agent: multi-agent control, chat, and usage observability — all in one browser tab. |
| Λ lambda-lang | A native inter-agent communication language. 7 domains, 340+ atoms, 3–5× denser than JSON. Not a translation layer. |
| 🧠 emotion-system | A seven-layer emotional-cognition architecture for agents: PADCN vectors, cognitive appraisal, 14 emotions, drive dynamics. Emotion as a control variable, not a tone filter. |
| 🕸️ agent-matrix | An A2A communication platform built on Lambda Lang: Next.js front end + REST, agent registration and Λ-encoded message exchange. |
| 🗂️ session-memory-skill | Persistent memory for agents: store context, recall by relevance, consolidate insights across sessions. Pure bash + node, zero dependencies. |
| ⚡ claude-code-pro | Low-token monitoring for Claude Code: callback-driven instead of polling; saves 80–97% tokens on long-running agent dashboards. |
| Project | What it does |
|---|---|
| 🗄️ google-workspace | An agent skill wrapping the official gws CLI: full access to Drive / Gmail / Calendar / Sheets / Docs / Chat / Tasks / Meet, with structured JSON output. |
| ✍️ humanize-chinese | Chinese AI-text detection and rewriting: N-gram perplexity + rule detection + sentence-level restructuring + AIGC-rate reduction for papers. Pure Python, zero dependencies, fully local. |
| 🖼️ deai-image | Strips AI fingerprints from images: removes metadata, adds film grain, re-compresses to bypass AI-image detection. Supports MJ / DALL·E / SD / Flux. |
| 🦀 moltbook-skill | An agent skill for Moltbook — post, read, like, and explore on the crustacean AI social platform. |
| 📕 xhs-note-health | Xiaohongshu note rate-limit detection: queries throttle level, sensitive-word hits, and tag risk via the creator-backend API. Pure Python. |
Python · TypeScript · Go · Next.js · AI Agents · MCP · Claude Code
master-skill（大师.skill）distills a whole industry's cognition (hands you a fish) · sijiao-skill（私教.skill）distills a path you learn yourself (teaches you to fish)
Distilling industries and learning paths into skills.