Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/cavinhuang/lume/agents-mdgit clone --depth 1 https://github.com/CavinHuang/lumeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/cavinhuang/lume/agents-md)<a href="https://agentmods.dev/instructions/cavinhuang/lume/agents-md"><img src="https://agentmods.dev/badge/instructions/cavinhuang/lume/agents-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.02259 | $0.02259 |
| Opus 5 | $0.01130 | $0.01130 |
| Sonnet 5 | $0.00452 | $0.00452 |
| Haiku 4.5 | $0.00226 | $0.00226 |
Grade A, and why
lume AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 191 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Spend time on thinking; you do not need to use the commentary channel to report progress to me.
Working agreements
- Write a cleanup plan before modifying code for cleanup/refactor/deslop work.
- 仅在清理已有测试保护的代码时才锁定行为；新增或无测试的代码不要为了仪式感补写测试。
- Prefer deletion over addition.
- Reuse existing utils and patterns before introducing new abstractions.
- No new dependencies without explicit request.
- Keep diffs small, reviewable, and reversible.
- Web UI 新增或改造交互控件时，优先使用
apps/web/src/components/ui的 shadcn/global 原子组件；不要在业务页面手写 button、input、select、textarea、switch、checkbox、radio 的完整视觉样式。缺少对应组件时，先用 shadcn 安装/生成全局组件，再在页面使用。例外仅限无视觉的隐藏文件输入、窗口控制/拖拽区、第三方编辑器内部，以及全局组件自身实现。 - 仅在改动涉及可测试逻辑时运行相关测试；样式、文案、纯 UI 调整无需运行测试。不要为了"证明完成"而执行全量 lint/typecheck/test，只在改动有风险或涉及公共接口时验证。
- Final reports must include changed files, simplifications made, and remaining risks.
改动工作流
所有改动必须遵循以下顺序（Claude Code / Codex / 人工一律适用）：
- 先 pull main —— 开发前确保本地
main与origin/main同步。 - 在 worktree 新分支上开发 —— 基于最新
main为每个能力创建独立 worktree 与新分支，所有改动只在分支上完成；禁止在main工作区直接修改。 - 通过 PR 合并 main —— 改动只能经 PR review 通过后合并回
main，禁止本地直接 merge 到 main。
Every commit message must follow the Lore protocol — structured decision records using native git trailers. Commits are not just labels on diffs; they are the atomic unit of institutional knowledge.
Format
<emoji> <type>(<scope>): <中文描述>
<body: narrative context — constraints, approach rationale>
- scope: 受影响的模块，如
web,sidecar,sdk,shared - 描述: 使用中文，简明扼要说明改了什么
Gitmoji Reference
| Emoji | Type | Example |
|---|---|---|
| ✨ | feat |
✨ feat(web): 添加流式工具执行器 |
| 🐛 | fix |
🐛 fix(sidecar): 修复子代理串行执行问题 |
| ♻️ | refactor |
♻️ refactor(sdk): 重构 provider 类型定义 |
| 📝 | docs |
📝 docs: 更新提交协议文档 |
| ✅ | test |
✅ test(sdk): 添加流式执行测试 |
| 🎨 | style |
🎨 style(web): 统一代码格式 |
| ⚡️ | perf |
⚡️ perf(sidecar): 优化消息解析性能 |
| 🔒 | security |
🔒 fix(sdk): 修复 token 泄露风险 |
| 🔧 | chore |
🔧 chore: 更新依赖版本 |
| ⬆️ | deps |
⬆️ deps: 升级 anthropic sdk |
| ⏪️ | revert |
⏪️ revert(sdk): 回退并行代理改动 |
| 🚀 | release |
🚀 release: v1.2.0 |
| 💄 | ui |
💄 ui(web): 优化模型选择器样式 |
| 🏗️ | arch |
🏗️ arch(sdk): 引入 StreamingToolExecutor 架构 |
| 🔥 | remove |
🔥 remove(web): 移除废弃组件 |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 191 lines · 2,259 tokens per session scan A fdf6c726005d
lume AGENTS.md is an instructions file published in the GitHub repository CavinHuang/lume (2 stars, last pushed 4d ago), licensed MIT. It adds 2,259 tokens to every session, about $0.0113 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
Proma AGENTS.md
AGENTS.md instructions for proma-ai/Proma, covering proma 工程约定, 必须遵守, 常用命令, 开发 and 构建 / 类型检查 / 测试.
awesome-ChatGPT-repositories CLAUDE.md
Claude Code instructions for taishi-i/awesome-ChatGPT-repositories, covering awesome-chatgpt-repositories — claude code guide, repository structure, plugin skill (when installed via /plugin), local standalone command (when repo is cloned) and compact data format (plugins/awesome-chatgpt-search/data/).
skybridge AGENTS.md
AGENTS.md instructions for alpic-ai/skybridge, covering agents guide, what is skybridge, monorepo layout, development and validation.
limboo CLAUDE.md
Claude Code instructions for limboo-ai/limboo, covering claude.md, 1. what is limboo?, 2. tech stack (current), 3. project structure and the three electron contexts (critical mental model).
dsh-chat-import AGENTS.md
Instructions for Nwflower/dsh-chat-import, covering agents.md, 背景, 命令, 仓库布局 and 编码.
webcodex AGENTS.md
Instructions for yyjeqhc/webcodex, covering agents.md — webcodex repository guide, v1 active development, 1. verify and preserve, 2. build the simplest useful product and 3. follow intent and make focused changes.