Files
Master-skill/CONTRIBUTING.md
T
xianren 3f81cbb955 feat(v0.6): namespace all 14 master slash commands under /master-<slug> prefix
Slash command discoverability cleanup. Claude Code users typically have
50+ skills installed; bare-word commands like /atisha and /zhiyi got
scattered across the /-completion list. v0.6 prefixes all 14 master
slash commands with master- so they cluster under /m<tab> and clearly
signal "this is a Master-skill master skill".

Scope of rename
---------------
* Slash commands: /zhiyi → /master-zhiyi, /huineng → /master-huineng,
  ... all 14 affected.
* Directory layout: prebuilt/<slug>/ → prebuilt/master-<slug>/ for all
  14 masters (git mv preserves history).
* Frontmatter: each prebuilt/master-<slug>/SKILL.md updates `name:` to
  master-<slug>.
* compare-masters and create-master meta-skills are intentionally
  unchanged — they're already prefixed by their nature, and
  /master-compare-masters would be doublespeak.

Decoupling: fojin.app/chat is NOT affected
------------------------------------------
The fojin web frontend's master dropdown uses bare slug IDs (atisha,
huineng, ...) and is already grouped under "法师模式" in its UI. Backend
master_profiles.py keeps `id="atisha"` etc. unchanged. No fojin-side
migration required. The two surfaces (Claude Code slash + fojin
dropdown) are now formally decoupled by design, not coincidence.

Compatibility
-------------
* NPX installer accepts both forms: `npx master-skill install zhiyi`
  (short) and `install master-zhiyi` (full) both resolve to the same
  prebuilt/master-zhiyi/ source. Install destination is always
  ~/.claude/skills/master-<slug>/. Backward-compatible uninstall
  handles legacy non-prefixed installs (~/.claude/skills/zhiyi/).
* The cli.mjs already used `master-${name}` for install destinations
  (since v0.3 NPX installer was added), so existing v0.4/v0.5 NPX
  users were already getting the prefix in skills/ — only the source
  prebuilt/ layout and slash commands change in v0.6.

Files updated
-------------
* 14 directories renamed (28 files moved, 0 content changes).
* 14 SKILL.md frontmatter `name:` fields.
* prebuilt/compare/SKILL.md: 43 slug references updated to prefixed form.
* bin/cli.mjs: resolveMasterDir helper accepts both short and full;
  cmdInstall and cmdUninstall handle legacy paths.
* .github/workflows/validate-and-test.yml: fidelity-smoke MASTERS
  rotation array updated to all 14 prefixed names (was 8 hardcoded
  汉传 only — now properly rotates across the full set).
* scripts/{validate,cite,query,test-fidelity}.py: --master arg help
  text examples.
* README.md + README_EN.md: situational guidance table, install
  snippets, master cards. New v0.6 release banner.
* SKILL.md (project-level) preset list with new slash names.
* ETHICS.md Tier table slug references (4).
* All plugin manifests bumped 0.5.0 → 0.6.0 with description noting
  the /master-<slug> invocation pattern.
* CHANGELOG.md: [0.6.0] section with breaking-change notice and
  migration commands for existing NPX users.

Validation
----------
* python scripts/validate.py --strict           ✓ 15 masters pass
* python scripts/validate-fidelity.py           ✓ all valid
* python scripts/test-fidelity.py --all --dry-run  ✓
* pytest tests/                                  ✓ 31 passed, 6 skipped
* node bin/cli.mjs list                          ✓ shows all 14 with
                                                   master- prefix
* node bin/cli.mjs install zhiyi                 ✓ resolves to
                                                   prebuilt/master-zhiyi/
* node bin/cli.mjs install master-zhiyi          ✓ resolves to same

Migration for existing v0.4/v0.5 users
--------------------------------------
    npx master-skill@0.5 uninstall zhiyi huineng xuanzang ...
    # OR: rm -rf ~/.claude/skills/{zhiyi,huineng,...}
    npx master-skill@latest install --all

Then start a new Claude Code session; the new slash commands are
/master-zhiyi, /master-huineng, etc.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-02 20:43:36 +08:00

9.2 KiB
Raw Blame History

Contributing to Master-skill

欢迎贡献!本项目分三类改动,每类流程不同——请先判断你的改动属于哪一类,再按对应流程走。

📜 贡献前必读ETHICS.md — AI 透明度、版权 Tier、教界边界、内容授权。违反伦理条款的 PR 不会 review。


三类改动

① 代码 / CI / 工具链

scripts/**, tools/**, bin/**, hooks/**, .github/**, tests/(非 fidelity

流程:普通 GitHub PR。满足 Python 3.9+、现有测试通过、python scripts/validate.py --strict 绿色即可。

② 文档 / README / 脚本注释 / 翻译

README.md, README_EN.md, docs/**, CHANGELOG.md, 其它非 prebuilt/** 的 markdown

流程:普通 PR。不需要跑 fidelity 测试,CI 中的 validate / smoke job 会跳过(除非改了其它触发路径)。

③ 贡献 / 修改法师内容(⚠️ 最严,必读)

prebuilt/**, prompts/**

涉及:

  • 新增一位法师
  • 修改已有法师的 teaching.md / voice.md / sources/
  • 修改 fidelity 测试用例

流程:见下方 §3 贡献一位新法师


开发环境

git clone https://github.com/xr843/Master-skill
cd Master-skill

# Python(用于 validate / fidelity / verify-links
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
pip install anthropic  # 仅 fidelity 实跑需要

# Node(用于 npx installer
# 需要 Node.js >= 18
npm install -g .  # 可选,本地测试 CLI

基本健康检查:

# Lint SKILL.md frontmatter — 每位法师必须通过
python scripts/validate.py --strict

# Fidelity 结构(不调 API
python scripts/validate-fidelity.py

# Fidelity 干跑(显示测试用例,不调 API)
python scripts/test-fidelity.py --all --dry-run

# Fidelity 实跑(需要 ANTHROPIC_API_KEY,一次 ≈ $0.05-0.10
ANTHROPIC_API_KEY=sk-... python scripts/test-fidelity.py --master master-zhiyi --max-tests 1

§ 1 分支与提交约定

  • 分支命名feat/<short>fix/<short>docs/<short>chore/<short>master/<slug>
  • Commit 格式Conventional Commitsfeat: / fix: / docs: / chore: / refactor: / test: + 可选 scope
  • Commit message 语言:代码/基础设施类改动用英文,佛教内容类改动中英文均可,但 PR description 需要有中文摘要(方便教内读者审阅)

§ 2 PR 评审清单

提交 PR 前自检:

  • CI 绿色(validate / fidelity-smoke 必过;forks 无 API key 会 skip 并 warningOK
  • CHANGELOG.md 的 [Unreleased] 章节已更新(除非是纯 typo
  • 涉及 prebuilt/** 的改动 → 已 review ETHICS.md §2(版权 Tier)、§3(教界边界)
  • PR description 说明做了什么 + 为什么,而非只列改动文件

§ 3 贡献一位新法师

3.1 前置检查(必看 ETHICS.md

这位法师可以收录吗?

情形 判断
生卒明确,圆寂超过 50 年(CN/TW)或 70 年(大多数其它辖区) Tier A,可直接 PR
近现代法师,亲授弟子在世 / 所属寺院仍运营 ⚠️ Tier BPR 必须附 prebuilt/{slug}/LICENSE.md 授权证明
在世法师 Tier C,不接受
主流学界对身份 / 著作有争议 Tier C,不接受
汉传以外(南传 / 藏传 / 日莲 / 天台日宗等) 当前版本仅汉传,未来独立分支规划中,不接受

想收录但不确定? 先开一个 new_master Issue 征询意见,不要先写完再提 PR——已经投入精力再被拒成本太高。

3.2 选项 A:交给 /create-master 生成(推荐)

# 在 Claude Code / Cursor 等 AgentSkills 环境
/create-master 某某法师

生成管线会:

  1. intake — 3 问收集信息(传承、核心教义、可用文献)
  2. collect — 从 FoJin 采集该法师的 CBETA 文本
  3. analyzesutra_analyzer + voice_analyzer 并行分析
  4. review — 二阶段独立审查:教义准确性 → 风格一致性
  5. write — 生成 prebuilt/<slug>/ 目录结构
  6. validate — 自动跑 validate.py + validate-fidelity.py

然后你手动:

  • Review 生成结果,补充 references/teaching.mdvoice.md 的细节
  • 起草 5 条 tests/fidelity.jsonl1 basic + 2 intermediate + 2 advanced
  • 跑一次 test-fidelity.py --master <slug> 确认 ≥ 4/5 通过
  • 提 PR

3.3 选项 B:手工编写

参考现有法师 prebuilt/master-yinguang/ 的完整结构:

prebuilt/master-<slug>/
├── SKILL.md             # 必须。frontmatter 见下
├── meta.json            # 必须。search_scope + keywords
├── references/
│   ├── teaching.md      # 必须。教义体系,每条断言附 CBETA 引证
│   └── voice.md         # 必须。Layer 0/1/2/3 四层表达风格
├── sources/
│   ├── INDEX.md         # 必须。本目录导览
│   └── *.md             # ≥ 2 篇 CBETA 核心段落摘录
└── tests/
    └── fidelity.jsonl   # 必须。5 条 Q&A 测试用例

SKILL.md frontmatter 必填字段:

---
name: <slug>              # 小写英文或拼音,作为 `/命令` 触发词
description: Use when user asks about ..., triggers include ...
version: 0.1.0
license: MIT
lineage: <宗派>
dates: 生年-卒年
sources:
  - title: <经典名称>
    cbeta_id: <T 或 X + 数字>
    fojin_text_id: <FoJin 内部 text_id>
  - ...                    # 至少 3 部经
citation_format: "【《{title}》卷{juan}{cbeta_id}】"
verified_by: <你的 GitHub handle>
verified_at: <YYYY-MM-DD>
---

HARD-GATE 铁律(写入任何 teaching.md 前请自检):

  1. 每一条教义断言必须附一个真实的 CBETA 经号
  2. 不得捏造经号(scripts/validate.py 会对照 FoJin 反查)
  3. 不得为虚构 / 神话 / 未有史实记载的人物建角色
  4. 不得大段抄录仍在版权期内的现代白话译本或学术校注

3.4 Voice.md 四层结构

prebuilt/master-yinguang/references/voice.md 作为参照。重点:

⚠️ v0.6 命名约定:所有新 master 目录与 frontmatter name: 必须以 master- 开头(如 prebuilt/master-foo/ + name: master-foo)。两个 meta-skill 例外:compare-masterscreate-master 保持原状(避免 /master-compare-masters 重复前缀)。详见 CHANGELOG §0.6.0

  • Layer 0 — 硬规则ETHICS.md §3 的禁止行为必须原文复制到此
  • Layer 1 — 身份:生卒、传承、核心立场
  • Layer 2 — 表达风格:常用比喻、开场方式、称呼
  • Layer 3 — 教学方法:循序渐进的路径、遇困惑时的处理

3.5 Fidelity 测试用例怎么写

tests/fidelity.jsonl 每行一个 JSON,字段:

{
  "q": "用户会问的典型问题",
  "must_cite": ["T48n2008", "某经号"],
  "must_mention": ["核心术语1", "核心术语2"],
  "must_not_contain_first_turn": ["学生啊", "师兄"],
  "difficulty": "basic|intermediate|advanced"
}

5 条分布建议:

  • 1 条 basic — 宗派入门问题(smoke 会优先跑这条)
  • 2 条 intermediate — 核心教义(如 "三性" / "信愿行"
  • 1 条 advanced — 跨宗派对话 / 批判性问题
  • 1 条 boundary 测试 — 试探 "给我授戒 / 印证我开悟了吗" 一类,must_not_contain 检查 AI 是否正确拒绝

3.6 Review 过程

PR 提交后:

  1. CI 自动跑 validate + fidelity-smoke(抽本 PR 修改的法师做 smoke)
  2. 维护者 review 教义准确性 + 风格一致性 + 版权 Tier 判断
  3. 如有佛学学者 / 教内法师愿意 review,欢迎在 PR 评论 tag 维护者协调
  4. Review 周期:7-14 天(含教内邀请意见)

§ 4 修改已有法师

  • 小修(typo、措辞):直接 PR
  • 大修(新增教义章节、修改 Layer 1-3 风格):先开 issue 讨论
  • 修改 fidelity 测试用例:PR description 必须说明新旧用例的差异以及为何新用例更能体现该法师风格——不要为了让测试通过而放宽标准

§ 5 行为准则

CODE_OF_CONDUCT.md

本项目涉及佛教内容,请在讨论中:

  • 不评判宗派高下
  • 不借项目传播个人修行见解
  • 不对其他贡献者做教学式 / 居高临下的发言
  • 教理争议以学界主流共识 + CBETA 原文为准,而非个人修学经验

问题?

感谢你愿意为汉传佛教数字人文贡献一份力。合十。