Zilan Agent - Framework Budista para LLMs
RyanYao527/zilan-agent
Agent budista com lógica formal, recuperação Ágama e validação de código
Ver no GitHubFramework de agent/skill dual-track para LLMs que integra recuperação de textos Ágama, lógica budista (因明), validação de código e harness de API OpenAI. Projetado para praticantes buddhistas que desejam instrumentar pesquisa profunda, análise cross-domain e relatórios estruturados sobre textos clássicos.
Instalação
npx convez-ai RyanYao527/zilan-agent1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Recuperação e análise de textos Ágama com lógica buddhista formal
- Validação de código Python/Claude com harness OpenAI integrado
- Pesquisa profunda estruturada sobre conceitos buddhistas clássicos
- Automação de workflows de prática e reflexão meditativa
Veredito

Projeto especializado e bem-estruturado que combina rigor técnico (CI/CD, testes de regressão, validação multi-plataforma) com domínio específico buddhista. Recomendado para construtores com interesse em LLMs orientados por lógica formal e pesquisa textual estruturada; curva de aprendizado elevada.
README
A tradução para o português deste README ainda está na fila.
孜澜 · Zilan
Independent Buddhist practitioner · 独立修行者
AI 大语言模型佛学 Agent / Skill 双轨框架 · Buddhist Agent / Skill framework for AI LLMs
基于优婆塞姚磊佛学体系 · Based on Upāsaka Yao Lei's Buddhist study system
🌐 Choose your language · 选择语言
| 语言 | Language | 入口 Entry | |
|---|---|---|---|
| 🇨🇳 | 中文 | Chinese | README.zh.md |
| 🇺🇸 | English | English | README.en.md |
💡 Tip: GitHub 会根据你的浏览器语言自动显示对应版本。如果你想切换,点击上表中的入口即可。 GitHub auto-detects your browser language. Click an entry above to switch.
⚡ 30 秒快速上手 · Quick start in 30 seconds
# 1. Clone
git clone https://github.com/RyanYao527/zilan-agent.git
# 2. Copy into your Claude Code skills directory
cp -r zilan-agent ~/.claude/skills/
# 3. Optional: install the Claude Code Agent definition
mkdir -p ~/.claude/agents
cp zilan-agent/agents/zilan-claude-code.md ~/.claude/agents/zilan.md
# 4. In any Claude session, mention 孜澜 / Zilan / 因明 / 摄类学 to activate
📦 What's inside · 仓库内容
| 文件 File | 用途 Purpose |
|---|---|
SKILL.md | 完整 skill 定义(中文) · Full skill definition (Chinese) |
SKILL-en.md | 完整 skill 定义(英文) · Full skill definition (English) |
README.zh.md | 完整文档 · Full documentation (Chinese) |
README.en.md | 完整文档 · Full documentation (English) |
CONTRIBUTING.md / CONTRIBUTING-en.md | 贡献指南 · How to contribute |
CHANGELOG.md | 版本变更记录 · Release notes |
LICENSE | MIT 开源许可 · MIT license |
agents/zilan-claude-code.md | Claude Code Agent 定义 · Claude Code Agent definition |
agents/zilan-codex.md | Codex sub-agent prompt · Codex sub-agent prompt |
agents/openai.yaml | 跨平台 Agent 配置 · Cross-platform Agent metadata |
CODEX_REGRESSION_TESTS.md | Codex 回归测试矩阵 · Codex regression matrix |
docs/platform-validation.md | 平台验证状态 · Platform validation status |
docs/runtime-validation-log.md | 运行验证记录 · Runtime validation log |
docs/runtime-evidence/ | 脱敏运行证据摘录 · Redacted runtime evidence excerpts |
docs/maintenance-roadmap.md | 维护路线图 · Maintenance roadmap |
docs/installation.md | 安装与运行指南 · Installation and runtime guide |
docs/validation-evidence.md | 运行证据归档规范 · Runtime evidence policy |
docs/provider-routes.md | Provider 路线归类 · Provider route triage |
docs/openai-api-harness.md | OpenAI API harness 说明 · OpenAI API harness guide |
scripts/validate_zilan_repo.py | 仓库结构与语料 smoke 校验 · Repository invariant checks |
scripts/search_agama.py | 阿含 Markdown 检索工具 · Agama Markdown search helper |
scripts/openai_api_harness.py | OpenAI Responses API dry-run/live harness |
scripts/mock_install_smoke.py | Claude Code mock install smoke test |
.github/workflows/ci.yml | 自动化校验 · Automated CI checks |
context/摄类学工具箱.md | 摄类学推理工具链 · Collected Topics reasoning toolkit |
context/因明推理引擎.md | 因明逻辑引擎 · Buddhist logic engine |
沟通过程.md | 沟通进化轨迹 · Communication evolution log |
上传步骤.md | 上传指南 · Upload guide |
🧭 Skill / Agent dual track · 双轨模式
- Skill mode: lightweight dialogue, daily practice reflection, and simple concept explanation.
- Agent mode: explicit deep research for Agama retrieval, Buddhist logic chains, cross-domain analysis, and long reports.
- Codex: use explicit prompts such as
spawn a zilan agent/让孜澜独立深入研究一下; regression cases live inCODEX_REGRESSION_TESTS.md.
✅ Compatibility status · 兼容性状态
- Platform status:
agents/openai.yamlis the machine-readable metadata source;docs/platform-validation.mdrecords status definitions, validation evidence, and update rules. - Current validation: Codex and Claude Code are runtime-tested as of 2026-06-12; OpenAI API is
harness-readywith dry-run tests and awaits a liveOPENAI_API_KEYrun. - Runtime boundary: Codex, Claude Code, OpenAI API, DeepSeek, GLM, and Qwen routes must not be described as tested unless the platform validation document says so.
- Scholarly collation: local Agama Markdown is a searchable working corpus; publication-level work should verify against CBETA XML and parallel texts.
🧪 Engineering checks · 工程校验
python scripts/validate_zilan_repo.py --check-generated
python -m pytest
python scripts/openai_api_harness.py --case ZC-02 --json
python scripts/mock_install_smoke.py
python scripts/search_agama.py --terms "無我|非我|緣起" --limit 10
python scripts/search_agama.py --terms "非我" --passages --group-by juan --limit 10
python scripts/search_agama.py --terms "緣起" --json --limit 5
GitHub Actions runs the same invariant checks, tests, and Agama search smoke test on push and pull request.
search_agama.py text and JSON output include stable citation fields with sutra name, CBETA ID,卷, and local Markdown line references, for example 《雜阿含經》(T02n0099) 卷 1, context/agama/T0099-za-agama.md:33.
Maintenance priorities and release guardrails are tracked in docs/maintenance-roadmap.md.
Manual runtime validation evidence is tracked in docs/runtime-validation-log.md.
Installation paths are documented in docs/installation.md; evidence and transcript rules are documented in docs/validation-evidence.md; redacted evidence excerpts live in docs/runtime-evidence/; provider route triage is documented in docs/provider-routes.md; release notes are tracked in CHANGELOG.md.
🔑 唤醒关键字 · Activation keywords
主关键字 Primary: 孜澜 · Zilan
身份 Identity: 姚磊 · 优婆塞 · Upāsaka · Yao Lei
经典 Scriptural: 阿含经 · Agama
逻辑 Logical: 因明 · 摄类学 · 应成论式 · 因三相 · 四句逻辑
场景 Contextual: 数字人佛学 · 数字人修学 · Buddhist digital persona
寻遍十方谁是我,波心圆月本无澜 · 遍了十方谁是吾,波心圆月本无澜