Borrowing it
Nothing to install: this file belongs to hanzhcn/laohan-skills. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/hanzhcn/laohan-skills/main/CLAUDE.mdgit clone --depth 1 https://github.com/hanzhcn/laohan-skillsWrote 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/hanzhcn/laohan-skills/claude-md)<a href="https://agentmods.dev/instructions/hanzhcn/laohan-skills/claude-md"><img src="https://agentmods.dev/badge/instructions/hanzhcn/laohan-skills/claude-md/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/instructions/hanzhcn/laohan-skills/claude-md"><img src="https://agentmods.dev/badge/instructions/hanzhcn/laohan-skills/claude-md.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.03470 | $0.03470 |
| Opus 5 | $0.01735 | $0.01735 |
| Sonnet 5 | $0.00694 | $0.00694 |
| Haiku 4.5 | $0.00347 | $0.00347 |
Grade A, and why
laohan-skills CLAUDE.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 9d 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 — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.
laohan-skills project rules
Brand name
- 对外品牌名统一为
laohanAI。所有口播稿、封面提示词、系列标签、安装器标题和后续生成内容不得再使用旧名“寒武纪AI”或“寒武纪说AI”。
Talking-head video workflow
真人口播动画默认采用 CODEX_DIRECT + Remotion:
- 新episode的导演预制默认调用
laohan-daoyanv2.1.2:它只读取视频项目当前method-v5.json和director-state-v5.md。第一请求先从内容发散2—3个在视觉隐喻、空间组织、运动逻辑或素材角色上本质不同的完整方向,方向形成后才按需读取最小范围Plugin知识,再统一收敛;第二请求只做最终复审,对真正值得增强的位置内部发散2—3个本质不同的候选。非穷尽技术搜索地图只提醒搜索维度,不是必选清单,未入选候选不落盘。两请求都停在WAITING_FOR_FOOTAGE。完成时必须同步本期_status.md,并在最后一次写入后实时运行bash scripts/check-episode-contract.sh episodes/<slug> config;不能用开始时PASS或干净Git代替episode准入。它不得剪辑、生成clean/SRT、写旧beat-sheet/EDL/motion-plan、实现Remotion或渲染。旧METHOD_LAB只在Jeffrey明确要求历史路线时读取laohan-daoyan/references/method-lab-legacy.md。 - ⑧—⑪由同一个 Codex 任务连续完成:完整观看原片并剪辑、写direct brief与按需source manifest、直接实现Remotion、渲染完整candidate并交Jeffrey观看。
laohan-donghua、AST、双renderer与proposal/extension/seal只用于Jeffrey明确发起的METHOD_LAB或历史episode。 laohan-sucai/SKILL.md:⑩仅在 source manifest 有真实 PROOF/BROLL 请求时供应 provider/license/source/SHA 完整且 visually_verified 的本期素材;无请求标 not_applicable,不创建空素材任务。laohan-bianpai/SKILL.md:全流程的唯一状态路由器;从laohanAI视频创作根目录运行,按 artifact + mechanical gate 报告唯一下一步。⑫—⑭在 laohan-yunying 定义前必须 BLOCKED。- V5旧期已有pending candidate再接入V5.1时,只能使用workflow受控迁移冻结原candidate与director-state SHA;bianpai视该期D1/D2为历史兼容完成并停在
JEFFREY_REVIEW,不得倒退覆盖已生成样片。 - episode 存在且通过 workflow
verify-episode-supersession.mjs的supersession-record.json时,bianpai 必须优先返回SUPERSEDED并拒绝 vendors/status/next/check;不得覆盖旧 executor lock 或把作废期迁移到新 runtime。 laohan-cheat/SKILL.md:只把 episode 接到上游cheat-on-content,不能保留本地教程型评分公式。教程与观点内容必须分开 calibration lane;state 迁移只允许先 dry-run。laohan-yunying/SKILL.md:⑫—⑭的抖音运营编排器。发布准备要求bianpai check --require final;真实发布仍必须 Jeffrey 确认。它保存 publish-record、数据快照和评论洞察,随后交给 upstream cheat-publish/retro;不自动发布、回复或改预测。- ①的 experiment 必须预先声明合法
metric_keys和 T+N 窗口;⑬只有匹配该窗口、播放与全部预注册指标非空的measurement_role: TARGET快照可完成,其他观测只能标CONTEXT,不得被 retro 当作假设结论。 bianpai vendors写 schema 3 的FROZEN_ON_RESUMEvendor preflight,冻结UP_TO_DATE|READY_LOCAL_AHEAD状态和 Cheat/dbskill 完整 HEAD SHA。status/next只做本地 HEAD 对照;任一 HEAD 漂移或旧 schema 2 均 BLOCKED,不能猜填或静默迁移。新 episode 必须走sync-content-vendors.sh --new-episode并创建00-编排/executor-lock.json;registry/runtime 漂移时进行中 schema 2 episode BLOCKED,不得刷新 preflight 静默换执行器。platform-display-evidence.json是⑫可选的发布后展示核对;缺失不阻断,存在时必须绑定本期截图/导出、aweme、标题、封面 SHA、final SHA 和三项 MATCHED 人工结论。错绑记录必须使⑫未完成,绝不自动修正或伪造证据。- ④ score 后只算 PARTIAL,最终盲预测 RECORDED 后才算 COMPLETE;production/final/full 不得接受 PENDING。⑧媒体/SRT、⑨时间轴/PROOF、⑪ workspace/handoff/qa-evidence 都由 workflow 机械 gate 验证。
- ③和⑤不能只凭当前稿 hash 通过:schema 2 的③还必须绑定未过期 ruleset review/expiry/scan 时间并写
platform_guarantee: false;⑤的04-深扫报告.md必须声明review_status: CLEAR与unresolved_issue_count: 0,04-事实核验.md必须声明fact_check_status: CLEAR、contradicted_count: 0、unverifiable_count: 0。CONTENT_CLAIMS 在 AUTONOMOUS_RUN 不因通用 HIGH/CRITICAL 确认规则询问用户,而是回②最多三轮修订;不得伪 CLEAR。 - schema 2 NATIVE 不再用 EDL
visual.render_plan决定审美:⑨先写 semantic contract/art direction/parity plan/native challenge,再确定性编译animation-ast.json;⑪的两个 proposal/extension 必须绑定 AST/proposal SHA、完整 state/transition 与 capability allow-list。旧 render_plan 仅保留 PARITY 兼容回归,不能作为表现上限。 - NATIVE 生产顺序必须是 proposal →
prepare-renderer-extension→ scaffold → native implementation →seal-renderer-workspace→ base render →apply-review-captions→ QA。未 seal workspace、renderer 内嵌字幕或缺 base/SRT/caption-style/candidate 哈希链都必须失败;Remotion/HyperFrames 的 base 都不得自行烧字幕。 - ⑪ QA 的字幕安全区和有声审片结论必须来自独立
qa-reviews/<renderer>.json,绑定 candidate SHA、审阅者、时间及具体说明;任何命令行 PASS 都无效。每个 selected beat 均须有 entry/middle/exit 三帧证据。 - ⑧的实际字幕必须经过
spoken-script-variance.json与当前稿的事实差异审阅。任何事实偏离都回②→⑤→④,⑨不得自行解释或继续生产;仅CLEAR、事实偏离为零的当前记录可放行。 - bianpai
vendors成功后必须为本期写入绑定 runtime lock 的00-编排/vendor-preflight.json;status/next缺少或检测到漂移时只能 BLOCKED,不能隐式联网或继续路由。 - ⑤必须输出当前稿/事实报告绑定的 claim ledger。每个 PROOF beat 与 proof asset 都要绑定
SUPPORTED claim_id和同源 evidence;只有 URL 或本地文件不能冒充该画面已证明口播事实。 - 配套V5.1回归使用项目中真实存在的
node scripts/test-v5-director-method.mjs和node scripts/test-codex-direct-contracts.mjs,并先运行node scripts/check-workflow-runtime.mjs;不得引用不存在的聚合脚本。修改被workflow-runtime-lock.json锁定的skill后必须同步更新内容SHA,否则bianpai必须BLOCKED。 - dbskill 不再由
npx --all更新:完整 vendor 在~/Documents/dbskill,并由 workflow 的sync-content-vendors.shfast-forward 后受管 symlink 到 Claude/Codex 通用 skills。视频 workflow 只固定调用 dbs-script-flow/dbs-resonate,条件调用 hook/ai-check/spread;其余能力见项目docs/dbskill-编排映射.md。 laohan-daoyanv2是V5导演预制的轻入口,不另建方法真源;实时config因registry/runtime漂移失败时必须报告阻断并停止,禁止手改lock。只有Jeffrey明确授权,才可由视频工作流的受控导演预制期迁移修复。V5仍为PILOT,真实本期candidate经Jeffrey验收前不得称为accepted或效果最优。⑩laohan-sucai按当前素材合同执行;laohan-donghua只保留METHOD_LAB历史兼容。laohan-xiazai的抖音下载按其references/douyin.md:先用opencli douyin user-videos取得临时play_url并立即下载;opencli 失效才降级到移动端 UA + iesdouyin_ROUTER_DATA,再按文档后续降级。抖音不存在 YouTube/yt-dlp 路径。对标/cheat 批量文字转录用 MLX 未量化 large-v3;⑧停顿和词级时间真值用 whisper-timestamped large-v3。ASR-B01 证明 MLX 会合并静音间隔,不能替代⑧;不要使用会出现 NaN/吞吐异常的 Torch MPS。路径见 workflowCLAUDE.md第24条。- schema 2 的①由
laohan-redian唯一主写 signals/candidates/source-health 与00-选题.*;laohan-douyinsousuo是可独立调用的 OpenCLI 只读取证 adapter,只写抖音证据,两者不合并、不做 programmatic skill-to-skill 调用。screening_summary在同一个 candidates 文件内留下全部 signals→8—12 longlist(不足8个时全部)→短名单链路,不新增节点或第二套打分器。①至少比较两个候选,唯一 SELECTED 必须绑定 PRIMARY;平台可用时再绑定 PLATFORM_SIGNAL,平台预检/查询失败则platform_alignment与content_gap均可UNAVAILABLE,不单独阻塞。当前 OpenCLI 1.8.6 的douyin search只有结果样本,hashtag hot是全站热点,hashtag search实测失败且hot --keyword不能可靠过滤,禁止反推搜索量、供给总量、增长或内容缺口。source health 仍覆盖DISCOVERY|DOUYIN_SEARCH|PRIMARY_PROOF并绑定本期结果 SHA;每个 metric key 都有正整数 T+N/DAY、方向、基线和抖音创作者中心来源。opencli hackernews top是当前命令,禁止hackernews hot。不得为①安装 TrendRadar/RSSHub/Obsei 或另一套浏览器/爬虫;外部项目只作设计参考。laohan-fengmianqiuzhiv1.12默认只走秋芝方向,从45套词典选择3个不同模板族,按推荐程度生成01—03三张9:16完整封面;不恢复排字脚本或评分流程。Jeffrey可改选,没有改选时发布准备默认以01为源候选。 laohan-chuangzuo1.7.0 的新期②必须写schema 3并通过laohan-chuangzuo/scripts/check-script-contract.mjs。固定开场为“嘿,你有没有这种感觉,”。热点短小允许自然短稿;内容单位必须是“新判断+支撑+观众价值”,同义内容不得用于凑字数或时长。人味废话与内容重复分开,仍保留至少4种结巴、语气词、自嘲、反问或停顿设备;分层时使用连续1、2、3……。旧episode继续按executor lock中的1.6.x schema 2验证,不迁移既有稿件。- 上游 Cheat 的 prediction anatomy 现统一为所有 calibration 阶段都写 7 组件、bucket、概率、中枢与反事实;cold-start 只降低 Confidence、拉平分布。工作流
AUTONOMOUS_RUN可用Review Mode: AUTONOMOUS_AGENT_PROXY,但不得伪造 Jeffrey 确认或用户覆盖。旧2026-07-12-native-method-real-run已在拍摄前 SUPERSEDED;当前生产期2026-07-16-codex-direct-production已真实完成①(92条发现信号、30条OpenCLI抖音关键词证据、3个候选,选中C01),唯一下一步是②写稿。
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.
- 9d ago First seen · 48 lines · 3,470 tokens per session scan A 2209f559fa36
laohan-skills CLAUDE.md is an instructions file published in the GitHub repository hanzhcn/laohan-skills (11 stars, last pushed 13d ago), licensed MIT. It adds 3,470 tokens to every session, about $0.0174 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 2026-08-30.
Other instructions, from other repositories
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
codex AGENTS.md
AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
spec-kit AGENTS.md
AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.