claude-md-forge

仓库创建 2026年7月26日最近提交 3 天前SkillHot 收录 2 天前

▸ 精选理由

为每次会话提供可复用的行为与背景模板，适合团队化上下文管理。

这个 Skill 做什么

把原始资料转成 CLAUDE.md/AGENTS.md 的行为合约，用作会话持久上下文。

把对话、转录、策略文档等原始素材提炼成目录级别会被持续载入的行为合同（CLAUDE.md 或 AGENTS.md），让之后打开的 Claude Code 会话自动带上策略、术语和已否决路径。适合把会议记录、决策或聊天历史变成长期上下文，而不是普通 README。关键是只留下会改变 agent 行为的内容，把深度细节放到 docs/，避免冗长的说明文档。

▸ 展开 SKILL.md 英文原文

Turn source material — a Claude or ChatGPT conversation, a YouTube or podcast transcript, a strategy doc, notes — into a topic-specific agent context file (CLAUDE.md for Claude Code, AGENTS.md for Codex CLI and Cursor, or AGENTS.md plus a thin CLAUDE.md shim for a shared repo), so every future session in that directory operates with the strategy, tactics, vocabulary, and rejected paths from that source. Distills rather than summarizes: keeps only what changes agent behavior, pushes depth to docs/, refuses to produce README-style documentation. MANDATORY TRIGGERS: CLAUDE.md, AGENTS.md, agents md, claude md, turn this transcript into a CLAUDE.md, make an AGENTS.md from this conversation, convert this chat into project context, capture this session's knowledge, topic CLAUDE.md, domain context file, persistent context from transcript, YouTube transcript to CLAUDE.md, seed a project with this knowledge, agent memory from this conversation, claude-md-forge.

Skill 开发管理行为合约会话上下文持久化预置Claude CodeCodexCursor

Stars

Forks

仓库内 Skill

7 日增星

安装 / 使用

给你的 Agent 一句话（通用）

帮我安装这个 skill：https://raw.githubusercontent.com/fullREFIT/skills/main/skills/claude-md-forge/SKILL.md

或 curl 直取 SKILL.md

curl -fsSL "https://raw.githubusercontent.com/fullREFIT/skills/main/skills/claude-md-forge/SKILL.md"

SKILL.MD 节选查看完整文件 ↗

# CLAUDE.md Forge

Convert unstructured source material into a **behavioral contract** that loads into every Claude Code session opened in a directory.

## The one idea that governs everything here

A CLAUDE.md is not documentation. It is persistent context that gets prepended to every session, and every line in it competes for the model's attention against the user's actual request. So the test for inclusion is never "is this true?" or "is this interesting?" — it is:

> **If the agent did not know this, what would it do differently?**

If there is no concrete answer, the line does not belong in CLAUDE.md. It belongs in `docs/`, or nowhere.

This is why summarizing a transcript produces a ba

via SKILL·HOT · 数据来自 GitHub 公开信息 · 原文版权归作者所有