Instruments

Built to be used, not just admired.

Every project here is a working instrument. Clone it, run it, break it — then tell us what happened.

Open source · v0.x

Argus

Grow a research tree from a messy field.

Argus takes a broad topic and keeps digging through it, round after round. Each pass adds sources, notes, relationships, open questions, and report updates to the same growing research tree — so when you come back, you never start from zero. Everything is packaged as one Claude Code skill. One window, one command, and Argus leads you through setup, dashboard launch, the research loop, and report updates. **Many eyes. Never asleep.**

deep researchclaude code skillknowledge graph

GitHub

Open source · guide

Skills System Guide

Stop trigger conflicts — organize agent skills as one coherent system.

A field guide to the layer Anthropic's single-skill docs don't cover: how multiple agent skills behave as a system. Learn where trigger conflicts come from, when a new skill deserves to exist, and how to keep a growing skill library coherent instead of chaotic. Includes a full HTML authoring guide and bilingual (EN/中文) documentation.

claude codeskillssystems engineering

GitHub

Essay · PDF

The Past, Future & Destiny of Agents

Why agents will not disappear — only change form.

A working essay on the deep structure of agent technology. It argues the scattered techniques — prompting, RAG, ReAct, context engineering, MCP, skills, harnesses, loops — are not a pile of tricks but one repeated motion: **the model hits the edge of what it has already compressed, external structure grows around it, and that structure later gets partially absorbed back into the model.** One-line thesis: **the agent is the model outside the model.** Published as PDF in English and 中文.

essayagentspdf

GitHub Docs →