Skip to main content

Agent 无关的项目业务事实账本:基于 Markdown + Git 的 AI 会话业务记忆工具包(零依赖、零遥测)

Project description

FengChaoSkills 蜂巢

CI License: MIT Python 3.9+

Agent 无关的项目业务事实账本:让 AI 会话在真实开发交付后自动沉淀"业务真相 + 落地证据",在新会话开始时按最小必要集合路由回这些记忆,并严格区分"当前事实"与"历史参考"。

FengChao is an agent-agnostic, Markdown + Git based business fact ledger for AI coding sessions: it captures business truth with evidence after real delivery, routes the minimal memory set into fresh sessions, and never confuses "current fact" with "historical reference".

零第三方依赖 · 零遥测 · 零网络请求 · 纯 Markdown + Git · 记忆数据永远属于你

它解决什么问题

AI 编程会话是无状态的。代码在 git 里,但业务语义只存在于对话中:为什么审核要分两级?"主管"指什么岗位?用户上个月否定过哪个方案?会话关闭,知识就死了。更危险的是记忆污染:把"讨论过"当"已实现"、把"随口一提"当"业务规则"、把"已废弃"当"当前行为"。

FengChao 用两个别人基本没做的设计解决这两个问题:

  1. 记忆的认识论分层:五级可信度 + 冲突优先级(business-context 当前真相 > task-records/changelog 落地证据 > implemented plan > conversation-records > proposed plan)。
  2. 否定记忆:用户明确拒绝过的方案会被记录,防止 AI 把被否掉的方案再提一遍。

与 Spec 工具(OpenSpec / Spec Kit)不冲突:它们管"接下来做什么"(事前),FengChao 管"业务认知为什么是现在这样"(事后),可共存。

60 秒 Quick Start

cd your-project

# 方式一(推荐,发布后可用):免安装一次性运行
uvx fengchao-skills init     # 或 pipx run fengchao-skills init

# 方式二(当前可用):git clone 后直接跑,零依赖
git clone https://github.com/HappyLeoYang/FengChaoSkills.git ~/FengChaoSkills
python3 ~/FengChaoSkills/skills/fengchao-business-memory/scripts/fengchao.py init

init 交互选择 agent(或 --agents claude,cursor,会自动探测已有的 .claude/.cursor/ 等目录),然后立刻验证:

uvx fengchao-skills status    # 全绿即可(源码方式则用 python3 .../fengchao.py status)

init 写入了什么(全部可见、可 diff、可干净移除)

写入 内容
.fengchao/ 唯一工具安装点:config.yaml + skill 完整副本(可整体删除)
fengchao/ 唯一记忆根:FENGWANG.md 路由入口 + 六层记忆目录(属于你,建议入 git)
.claude/ .cursor/ 薄入口 + 三个斜杠命令(每个文件 ≤ 10 行,只指向 .fengchao/skill/
CLAUDE.md / AGENTS.md 一个 marker 块(≤ 15 行,disable 精确摘除,原内容一字不动)
.claude/settings.json SessionStart / Stop 两个 hook(可 --no-hooks 关闭)

日常使用:你什么都不用做

正常开发即可。路由和维护由 skill 规则 + hooks 驱动:

  • 新会话自动从 fengchao/FENGWANG.md 路由回业务记忆(SessionStart hook);
  • 真实开发交付后自动沉淀 task-record + changelog(Stop hook 门禁:remind / strict / off 三档);
  • 纯修复/重构走 lite 档只记一条 changelog,不制造记忆噪音。

想主动用,三个动词(Claude Code 中为 /fengchao:route 等):

命令 作用
/fengchao:route <关键词> 找回相关业务记忆
/fengchao:remember 把刚才的业务解释记入记忆
/fengchao:status 看看记忆系统状态

停用与卸载(三级分离,随时反悔)

uvx fengchao-skills disable     # 暂停:摘除全部注入,记忆和工具本体全保留;enable 逐字节还原
uvx fengchao-skills uninstall   # 移除工具:删 .fengchao/,记忆文档保留(它属于你)
uvx fengchao-skills uninstall --purge-memory   # 连记忆一起删,需要二次确认

disablegit diff 干净可读;每一个写入你项目的字节都有对应的干净摘除路径。即使卸载,记忆仍是一套完好的、人类可读的 Markdown 文档库。

六层记忆模型

目录 语义 写入时机
business-context/ 当前业务真相(每条规则一个现行条目,带证据链) 仅稳定业务事实落地后,经语义合并写入
task-records/ 已交付任务的业务意图、方案、证据(不可变) 仅真实开发交付后(full 档)
changelog/ 已落地变更历史(不可变) 任何真实交付(lite/full 档)
plan-records/ 最终计划(proposed → implemented 状态流转) Plan 产出最终计划后
conversation-records/ 用户业务解释、术语、偏好、否定项 用户给出长期有价值的解释后
FENGWANG.md + memory-map.md 路由入口 所有写入命令自动追加

隐私默认:只保存萃取摘要,永不保存完整对话。

支持的 Agent Surface

Claude Code(薄入口 + 斜杠命令 + hooks 硬门禁)· Cursor(rule + 命令)· OpenCode(命令 + opencode.json)· Codex / 通用 Agents(AGENTS.md marker 块)。不用 Claude Code 也能用:五个 surface + 可选 git pre-commit 钩子兜底(fengchao-skills install-git-hook)。

文档

信任声明

  • 运行时零第三方依赖fengchao.py 只用 Python 标准库,目标项目无需安装任何东西。
  • 零遥测、零网络请求:没有任何数据离开你的机器。
  • 记忆数据神圣:任何命令(uninstall/upgrade/migrate/compact)都不会在无显式确认下删除或改写记忆内容。

License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

fengchao_skills-0.2.0.tar.gz (60.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

fengchao_skills-0.2.0-py3-none-any.whl (59.3 kB view details)

Uploaded Python 3

File details

Details for the file fengchao_skills-0.2.0.tar.gz.

File metadata

  • Download URL: fengchao_skills-0.2.0.tar.gz
  • Upload date:
  • Size: 60.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for fengchao_skills-0.2.0.tar.gz
Algorithm Hash digest
SHA256 f14ec6f11d4e776ef2903810976279ed41f5b88a86723184669dc4f120907376
MD5 18afde79e34dd0122ebbb85707eafffb
BLAKE2b-256 1e6a3bc5fccdcc2bcde2776b0b98fc22339a8b9f060fbd06e380b155864249af

See more details on using hashes here.

Provenance

The following attestation bundles were made for fengchao_skills-0.2.0.tar.gz:

Publisher: release.yml on HappyLeoYang/FengChaoSkills

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file fengchao_skills-0.2.0-py3-none-any.whl.

File metadata

  • Download URL: fengchao_skills-0.2.0-py3-none-any.whl
  • Upload date:
  • Size: 59.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for fengchao_skills-0.2.0-py3-none-any.whl
Algorithm Hash digest
SHA256 4e68463adbbbe03bb15fd68f4a029e2d58783829e7fe4b6372dfb7b711ae4dae
MD5 791531de6982a7733ddcd78be76ee9d6
BLAKE2b-256 4357545ea5b58287b0165055d9bbf23d10ac9b0fa17fdab6bf6160611b99bba1

See more details on using hashes here.

Provenance

The following attestation bundles were made for fengchao_skills-0.2.0-py3-none-any.whl:

Publisher: release.yml on HappyLeoYang/FengChaoSkills

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page