gitmyhub

master-skill

Shell ★ 59 updated 15h ago

大师.skill — 输入行业,自动调研 6 轨[行业大佬 / 工具地图 / 工作流 / 知识正典 / 信息源 / 术语标准] → 提炼为可运行的行业 Master OS skill;装到任意 Claude Code / OpenClaw / Codex / Hermes agent 即让其进入「这一行的资深人」模式。MIT,Python + Shell。

A shell-based skill for AI coding agents that automatically researches any industry and produces a reusable knowledge package capturing expert-level mental models, workflows, and terminology.

Shellsetup: easycomplexity 2/5

This repository is a skill for AI coding agents (Claude Code, Codex, and similar tools) that automatically researches an industry and produces a reusable knowledge package for that industry. You tell the agent which specific field you are in, and it spends 30 to 60 minutes running six parallel research tracks: finding the prominent figures in that field, mapping the tool landscape, documenting common workflows, identifying canonical reference materials, finding reliable information sources, and capturing the terminology and jargon specific to that industry.

The output is a directory containing a skill file that, when loaded into any compatible AI agent, puts the agent into what the README describes as "the most senior person in this field" mode. The idea is that the agent does not just know facts about the industry but has internalized the mental models, decision rules, and working patterns that experienced practitioners actually use. The README gives examples showing the kind of responses this produces: instead of listing generic options, the agent identifies the real constraints and tradeoffs a practitioner would care about.

Along with the main skill file, the process also generates a set of bash command-line scripts, one for running an agentic research protocol on a new question, others for specific decision trees and workflow walkthroughs. These scripts have no external dependencies and work on plain bash.

The skill also optionally invokes a separate tool to distill three prominent individuals from the industry into their own sub-skills, so the agent can later respond from a specific person's perspective when asked.

Source quality is checked with 16 automated rules. Primary sources (official documentation, the author's own writing, academic papers) are required. Content from certain aggregator sites is automatically rejected. Each source is dated, and sources older than 18 months are weighted less. The skill explicitly documents what it does not cover and which parts will go out of date fastest.

Where it fits