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 chenyuxiaojin/xiaochen-skills --skill cyxj-deepergit clone --depth 1 https://github.com/chenyuxiaojin/xiaochen-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/skills/chenyuxiaojin/xiaochen-skills/cyxj-deeper)<a href="https://agentmods.dev/skills/chenyuxiaojin/xiaochen-skills/cyxj-deeper"><img src="https://agentmods.dev/badge/skills/chenyuxiaojin/xiaochen-skills/cyxj-deeper/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/chenyuxiaojin/xiaochen-skills/cyxj-deeper"><img src="https://agentmods.dev/badge/skills/chenyuxiaojin/xiaochen-skills/cyxj-deeper.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 | $0.00196 | $0.01501 |
| Opus 5.5 | $0.00078 | $0.00600 |
| Sonnet 5.5 | $0.00039 | $0.00300 |
| Haiku 4.5 | $0.00020 | $0.00150 |
Grade A, and why
cyxj-deeper 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 5d 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 — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.
cyxj-deeper · 做深,不做大
方法论来自 Felix Rieseberg《How to be more ambitious: A tutorial for humans with models》 (https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/ ,2026-10-13): 模型的能力早已远超大多数人的用法,真正的瓶颈是人不敢要。这个 skill 的任务是替用户把要求往深处提—— 不是加更多功能、更多页数,而是做成这个领域顶尖的人会交出的那个版本。
用户是非程序员——解释直白,术语第一次出现就一句话说清。
流程(三步,顺序执行)
第 1 步:四问(动手前必须答完)
先从对话上下文、项目文件、~/项目/内容创作/log/index.md 推断任务背景,推断得出来的不许再问。然后逐条回答:
- 顶尖的人会多做什么:这个领域顶尖的从业者/创作者,在这里会多做哪些用户没要求的事?
- 最原始的材料:他们会从哪种最原始的材料起步——后台数据、评论区、原片素材、一手来源、原始数据—— 而不是别人已经消化过的二手总结?
- 具体方法和工具:他们会用哪些具体方法和工具?要真的去用(跑数据、做对照、写脚本、抽帧), 不是只换一套专家口吻。
- 外部裁判:除了模型和用户之外,什么能证明结果是对的?平台数据、可自动跑的检查脚本、对照样本、 真实观众反馈。不许模型给自己打分。
涉及外部事实(平台规则、工具能力、行业做法)要联网查证,查不到就标「未验证」,不许拿记忆冒充事实。
第 2 步:重写任务,给用户确认
把原任务重写成深度版,格式:
## 结论(一句话:深度版要交付的结果,用动词定义)
## 四问
1. 顶尖的人会多做:……
2. 起步材料:……(在哪、能不能拿到)
3. 方法和工具:……
4. 外部裁判:……(怎么判对错,谁来判)
## 重写后的任务
(一段可以直接执行的需求)
## 需要你拍板的
(只列答案会改变做法的 1~3 个决定)
然后停下等用户确认。用户说「行/开干」才进入第 3 步;用户改了就按改的来。
第 3 步:执行
按确认后的深度版执行。过程中遇到「做不到」先别退回原版——换方法再试;真卡住了,说清卡在哪、 已经试过什么,再请用户决定。交付时附上第 4 问的外部裁判结果(数据、脚本输出、对照),而不是自评。
原则(来自原文)
- 定结果,不派单任务。 问一句「目标背后的目标是什么」:不是「修这个崩溃」,是「保证 app 不崩」; 不是「改这篇稿」,是「这条视频在某平台达到某个数据」。前提是让模型自己能拿到反馈、能复现、能验证。
- 用专家的方法,而不是专家的用词。 想用 markdown 表格糊弄的地方,坚持上数据库、统计、对照实验。
- 敢定不合理的目标,并当作合理去推。 网页 10 秒加载,就直接要 200 毫秒;模型说做不到,告诉它能做到,让它继续。
- 用外部裁判。 结果由数据、脚本、样本来判,不由模型或用户的感觉来判。
- 用动词定义作品。 季度汇报不是「一份 PPT」,是「30 分钟内让忙碌的人信服一个因果故事」; 一条视频不是「一个 3 分钟成片」,是「前 5 秒让刷到的人觉得这事和他有关」。
- 组专家队伍,但做他们会做的事,不角色扮演。 编剧去拆结构、剪辑去看节奏、音频设计去测响度、 红队观众去找划走点——每个角色交付他那一行真正会交付的东西,而不是模仿他的口气说话。
原文深度 prompt(原样保留)
I want a deeper version of this task, not a bigger one. Before rewriting it: (1) say what a top professional in this field would do here that I'm not asking for; (2) name the rawest material they would start from instead of what has already been digested for me; (3) name the specific methods and tools they would use; (4) say what, other than you or me, would show the result is right. Then rewrite my prompt to ask for exactly that.
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.
- 5d ago First seen · 90 lines · 196 tokens per session scan A 8ced048f8a8f
cyxj-deeper is a skill published in the GitHub repository chenyuxiaojin/xiaochen-skills (8 stars, last pushed 5d ago), licensed MIT. It adds 196 tokens to every session and 1,501 once invoked, about $0.0008 per session on Opus 5.5. 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 2026-10-20.
Other skills, from other repositories
triage
Work a repo's needs-triage backlog, and outside contributors' issues and pull requests, in batches the decider approves in one pass: each item gets a cheap check and one proposed move with a plain-words reason, and only approved moves are applied. Use when asked to triage or work through a needs-triage pile, or to…
retro
Read a sample of one repo's past agent sessions, Claude Code and Codex, find the friction the agents hit, and return a short ranked list of fixes in plain words, each labeled and routed to whoever owns the fix. Use when the user asks for a retro or retrospective on agent work, asks what keeps going wrong in a repo's…
ingest
For a media file or URL shared with a goal, transcription requests, or takeaways from a call, demo, playtest, or video, build a transcript and timestamped visual evidence packet with a manifest and recommended next workflow.
show-me-your-work
For autonomous multi-phase or unattended work, a reviewer needing an evidence trail, or an explicit decision-log/show-your-work request, keep an append-only decision log with why, evidence, and result. Record decisions and corrections, not hidden reasoning transcripts.
to-tickets
When asked to file tickets, break down approved work, or move a settled plan onto the board, turn the spec, closed decision map, or pressure-tested conversation into verifiable work items with real dependencies and non-ticket blockers.
setup
For a new TeamWorkflow installation, binding refresh, or drift audit after a pack release, inspect and bind the project workflow. Reuse explicit recorded decisions with their sources; ask about missing, conflicting, or changed authority and bindings.