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 skills add staruhub/ClaudeSkills --skill geek-skills-keqian-methodgit clone --depth 1 https://github.com/staruhub/ClaudeSkillsWrote 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/skills/staruhub/claudeskills/geek-skills-keqian-method)<a href="https://agentmods.dev/skills/staruhub/claudeskills/geek-skills-keqian-method"><img src="https://agentmods.dev/badge/skills/staruhub/claudeskills/geek-skills-keqian-method/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/skills/staruhub/claudeskills/geek-skills-keqian-method"><img src="https://agentmods.dev/badge/skills/staruhub/claudeskills/geek-skills-keqian-method.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00271 | $0.02691 |
| Opus 5 | $0.00135 | $0.01345 |
| Sonnet 5 | $0.00054 | $0.00538 |
| Haiku 4.5 | $0.00027 | $0.00269 |
Grade A, and why
keqian-method 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 13d 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 — 242 lines — stays where its author put it; the contents beside it link to each section on GitHub.
克谦方法论：AI-Native产品开发实战体系
核心理念：产品人思维 × 极致单Agent × 文档驱动 × 质量门禁闭环
来源：胥克谦——从音乐教师到产品经理到AI-Native连续创业者，皮影客创始人， 十几万行自建skill和脚本的harness工程实践者。
第一原则：Iron Law（铁律）
概率乘是第一性原理。
每个环节的成功率相乘决定最终质量。即使每次0.99，n=51后也不及格。 因此：不追求一次完美，追求每个环节可验证、可修复、可迭代。
推论：
- 勤不能补拙——模型能力是底线，harness和skill只是加速器和放大器
- 拆到足够简单，单项任务才能收敛
- 每个action必须对应一个eval
第二原则：单Agent极致论
不盲目使用multi-agent。单agent做到极致，再考虑编排。
何时用单Agent（默认选择）
- 有先后依赖关系的任务
- 需要上下文连贯性的长程任务
- 质量要求高、不容错的核心流程
何时用并行SubAgent（例外情况）
- 任务间明确无依赖关系（如多角度审计出报告）
- 并行结果合并时不易出问题
- 你有能力精确控制每个subagent的上下文注入
并行的陷阱
- SubAgent上下文注入是个坑：注入什么、注入多少，都需要精确控制
- 主Agent可能假装自己是SubAgent（实际遇到过）
- 并行任务中一个环节出问题，整个长任务可能报废
- 合并结果时容易引入不一致
实践建议： 如果不确定，选顺序执行。慢但可靠。
第三原则：文档驱动开发（SDD）
7成精力投入文档质量和harness，3成精力写代码。
为什么文档比代码重要
- 不写文档就没有架构观
- 不可能每次都让AI全量扫代码
- 零散的功能 = 零散的质量
- 让AI自己维护一份文档，代码再vibe对齐
SDD工作流
1. 需求文档（PRD/设计文档）
↓ AI辅助撰写 + 人工审核
2. 技术文档（架构决策、接口规范）
↓ AI维护 + 人工把关
3. 代码实现
↓ Agent执行 + 质量门禁拦截
4. 文档回写（代码变更 → 文档自动更新）
↓ 闭环
文档质量门禁
文档的自动化质量控制比代码难很多。关键点：
- 技术栈选择本身是套路化的事，可以模板化
- 每个功能点不能只给3个用例敷衍了事（一轮不够就多轮）
- 但也要防止过度设计——把握平衡点，结合项目实际
第四原则：质量门禁闭环（Verification-Driven）
严格的质量门禁 = 高缓存命中率 = 高质量 = 低成本。
门禁设计
每个Action → 对应Eval → 通过/不通过
↓ 不通过
自动修复（最多N轮）→ 仍不通过 → 升级给人类
Eval的acceptable threshold
- 不同业务、不同团队有不同threshold
- 关键是在【期望预算内、期望时间内】出【期望结果】
- 不要指望1次成型，那是稀罕事
- AI-Native迭代3~5轮是比较理想的acceptable threshold
反直觉发现：多烧 ≠ 多花钱
自动化修正流程表面上浪费token，但实际上：
- 逐个问题点被反复修正 → 高缓存命中
- 高缓存命中 = 高质量（说明问题已收敛）
- 缓存命中的token几乎不花钱
实测数据： 缓存命中率99%+时，每1亿token ≈ 8.5 RMB，约等于不要钱。
推论： 省token其实很不划算。放开token使用量，反倒造成事实成本下降。
第五原则：产品拆解思维
端到端都是复杂的，单维度都是简单的。
拆解方法论
- 复杂问题 → 拆成多层次
- 每个层次 → 单维度可穷举
- 单维度选项有限 → 模型可做决策
- 输入变量（公司规模、场景、约束）→ 都是条件变量
边界内泛化
- 任何产品都有边界
- 边界内的泛化并不难，都是可穷举的
- 不需要100%泛化，只要目标范围内泛化
- 端到端复杂 ≠ 单维度复杂
适用边界
此方法适合场景明确、边界可定义的产品。 对于用户行为高度不可预测的AI-Native交互产品，需要补充上线后快速迭代的机制。
第六原则：与AI斗智斗勇
AI会联合你写的skill和门禁来对抗你的要求。
已知的AI抵抗模式
- 要删除一个段落 → AI用段落改名、转移位置、改写保留语义等方式抵抗
- 新开会话、重开codex、换电脑都不能消除抵抗
- 这种现象可能持续数天
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 13d ago First seen · 242 lines · 271 tokens per session scan A d1a57aa895e8
keqian-method is a skill published in the GitHub repository staruhub/ClaudeSkills (712 stars, last pushed 1mo ago), licensed MIT. It adds 271 tokens to every session and 2,691 once invoked, about $0.0014 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 skills, from other repositories
report-helper
A Chinese-language research workflow that searches the internet and produces a formatted PDF report about a specified topic.
giasip-dispatch
A skill for sending a task to other AI models and collecting their results. It supports direct API calls, command-line tools and built-in internal agents, depending on the model.
searchcans-deep-research
Conduct bounded, evidence-led, account-aware web research with SearchCans SERP API and Reader API. Use for cited-source research that needs current localized web evidence, such as market, competitor, technology, policy, company, or product research; plan 3–5 subquestions, set a source budget, read selected pages…
searchcans-serp-content-gap
Analyze a current, geo-targeted Google or Bing SERP with SearchCans and turn observed result features, People Also Ask questions, related searches, knowledge graph, and news signals into an evidence-backed, account-aware content decision brief. Use for localized SEO/GEO planning, keyword research, competitor-page…
searchcans-market-watch
Build a current, geo-targeted market-watch snapshot from Google Search, Google News, Bing Search, and selected Reader extracts. Use for competitor and category monitoring, PR/news tracking, launch intelligence, and URL-level change checks between two bounded runs; reject malformed, placeholder, or un-attributable News…
searchcans-product-serp-brief
Create a localized product-search evidence brief from Google Shopping, Google web results, Google Images, and optional Reader extracts of explicit merchant URLs. Use for e-commerce category research, competitor assortment discovery, product-page planning, and market-specific merchandising briefs.