Token导航 LogoToken导航

Agent Skill · 研究检索

documentation(文档)

用于辅助文档、README、Markdown、说明文和内容稿件的整理与改写。它适合让 Agent 提炼结构、补齐章节、统一术语、检查链接或把零散材料整理成可读文档。使用时应保留项目已有事实、命令和路径,不要把未确认的信息写成确定结论;涉及对外文案时,还需要控制语气,避免过度营销或夸大能力。

documentation 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为接触敏感数据;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 62,244。

总安装 339周安装 14GitHub Stars 公开资料未说明下载 111核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:documentation(文档)
来源仓库:https://github.com/usmanmehmood55/documentation
仓库路径:skills/documentation
安装命令:
npx skills add https://github.com/usmanmehmood55/documentation --skill documentation
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/usmanmehmood55/documentation --skill documentation

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

用于辅助文档、README 和 Markdown 内容的整理与改写。适合提炼结构、补齐章节或统一术语,生成可读性强的文档。使用时应保留项目已有事实,避免将未确认信息写成确定结论。涉及对外文案时需控制语气,防止过度营销或夸大能力。documentation 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/documentation

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/usmanmehmood55/documentation --skill documenta…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要接触敏感数据,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

documentation 怎么安装?

在终端执行 npx skills add https://github.com/usmanmehmood55/documentation --skill documentation。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

documentation 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

documentation 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

documentation 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

documentation 用的人多吗?

公开数据是累计安装 339 次,近一周安装 14 次,下载 111 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

documentation 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
documentation

文档 · 本页

339—接触敏感数据未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Documentation

When asked to review, create, maintain, or sync documentation, follow this workflow and these rules.

1. Goal

Keep documentation aligned with the current codebase and easy to maintain.

Priorities, in order:

  • correctness
  • current-state only
  • consistency with the codebase
  • non-duplication
  • minimal necessary churn
  • readability

Treat code, config, and actual file paths as the source of truth. Do not invent behavior that is not present in the repository.

2. Scope

2.1. Primary documentation surface

Consider these as the main documentation surface unless the user narrows scope:

  • README.md
  • CONTRIBUTING.md
  • CHANGELOG.md
  • files under docs/ or similar documentation directories
  • nested README.md files in packages, services, or modules

2.2. Exclusions

Do not edit:

  • files explicitly marked as frozen, archived, or generated
  • historical or release-history sections of CHANGELOG.md unless the user explicitly asks
  • roadmap, proposal, migration-story, or retrospective documents unless the user explicitly asks
  • non-Markdown files unless the user explicitly wants the same rules applied

2.3. Scope discipline

Prefer the smallest correct edit set.

  • If the user names a file, start there.
  • If related docs must change to avoid contradictions, update them too.
  • Do not expand into a repo-wide rewrite unless the user asked for a full pass.
  • If related files appear stale but are outside the requested scope, mention them in the final report instead of editing them by default.

2.4. Documentation config

If docs/docs-config.json exists, use it as the source of truth for documentation topology.

This includes:

  • which docs are part of the documentation surface
  • each doc's status, such as active, frozen, archived, or generated
  • each doc's role and declared topics
  • canonical topic ownership from topic_map

Use this config for topology only. Do not treat it as the source of truth for product behavior, API contracts, config defaults, or command truth.

If the config is missing or incomplete, infer the documentation surface from the repository. If it is missing and the task is in scope, create it before continuing and mention that in the final report.

3. Modes

First determine which mode applies.

3.1. Review only

Use when the user asked for a documentation review, drift check, or audit without edits.

Output:

  • files reviewed
  • inconsistencies found
  • stale or duplicated content found
  • recommended edits
  • anything that could not be verified from the codebase

3.2. Targeted update

Use when the user asked for a specific file or topic update.

Edit only the requested area plus any directly dependent docs that would become incorrect or contradictory if left unchanged.

3.3. Topic sync

Use when the user asked to sync a topic across docs, such as:

  • environment variables
  • API behavior
  • CLI commands
  • file paths
  • configuration defaults
  • naming changes

Update every doc that materially describes that topic.

3.4. Full documentation pass

Use when the user asked for a general documentation upkeep pass or repo-wide sync.

Check the main documentation surface and make focused edits where needed. Avoid rewriting accurate sections just for style.

4. Workflow

4.1. Determine exact scope

Classify the request as one of:

  • review only
  • specific file update
  • topic-specific sync
  • full documentation pass
  • new document in existing style

If the user names a file or topic, prioritize that area first.

4.2. Load documentation topology

Before editing, load documentation topology from docs/docs-config.json when present.

Use it to determine:

  • which docs are in scope
  • which docs must not be edited
  • which doc is canonical for the requested topic
  • whether related docs should be updated or only reported as out of scope

If available, prefer helper-script output over manual inference for inventory, topic ownership, frozen-doc detection, heading checks, line-wrap checks, and Markdown link checks.

If docs/docs-config.json is missing and the task is in scope, create it with subskills/documentation-topology/scripts/index_docs.py --root <target-repo> --write-config before continuing.

4.3. Inspect the source of truth

Before editing, inspect the codebase and config relevant to the request.

Check as relevant:

  • public APIs, endpoints, request and response behavior
  • CLI commands, scripts, and examples
  • environment variables
  • config keys, defaults, and supported values
  • feature availability and limitations
  • file names, paths, and module names
  • build, test, and run instructions
  • ownership or topology diagrams if present

Do not preserve claims that the codebase does not support.

4.4. Identify drift

Look for:

  • outdated behavior descriptions
  • old file paths or command names
  • mismatched env var names
  • duplicate explanations across docs
  • conflicting terminology for the same concept
  • speculative or future-looking wording
  • historical comparisons outside explicitly historical docs
  • stale diagrams or architecture summaries

4.5. Edit minimally

Preserve accurate content. Change only what is needed for:

  • correctness
  • clarity
  • consistency
  • removal of stale or duplicated content

Avoid stylistic churn that does not improve correctness or maintainability.

4.6. Keep related docs aligned

If one change affects another document, update the related document when needed to prevent contradictions.

Examples:

  • renaming an env var in README.md may require updates in docs/CONFIG.md
  • changing API behavior may require updates in both usage docs and architecture docs
  • renaming a command may require updates in setup, examples, and troubleshooting

4.7. Use helper scripts

When available, use the lightweight helper scripts in this skill to reduce manual scanning and improve consistency.

Preferred scripts:

  • subskills/documentation-topology/scripts/index_docs.py
  • subskills/documentation-topology/scripts/build_topic_map.py
  • subskills/documentation-topology/scripts/create_topic.py
  • subskills/documentation-topology/scripts/check_frozen_docs.py
  • subskills/documentation-formatting/scripts/check_heading_style.py
  • subskills/documentation-formatting/scripts/check_line_wrap.py
  • subskills/documentation-formatting/scripts/check_doc_links.py

Script rules:

  • prefer JSON output when available
  • treat exit code 0 as no issues found
  • treat exit code 1 as issues found
  • treat exit code 2 as config or usage error
  • treat exit code 3 as runtime failure
  • always pass --root <target-repo> to topology scripts; do not rely on the current working directory
  • prefer script output over manual inference for topology, frozen status, heading style, wrap issues, and link validity
  • use python <this-skill>/subskills/documentation-topology/scripts/index_docs.py --json --root <target-repo> --write-config to bootstrap docs/docs-config.json when it does not exist
  • use python <this-skill>/subskills/documentation-topology/scripts/build_topic_map.py --json --root <target-repo> --suggest-topics when reviewing or expanding per-document topic lists so the agent can see candidate topics inferred from filenames and headings
  • use python <this-skill>/subskills/documentation-topology/scripts/create_topic.py --json --root <target-repo> <topic> to register new topics explicitly in the config after the user or the agent has decided they belong in the documentation topology
  • summarize relevant script findings in the final report

4.8. Validate formatting

After editing, validate all changed Markdown files with the bundled formatter.

Use:

python <this-skill>/subskills/documentation-formatting/scripts/format_markdown.py --check [flags] <file.md> [...]

Rules:

  • Resolve the formatter path relative to this skill directory, not the target repository.
  • Only use the bundled formatter at <this-skill>/subskills/documentation-formatting/scripts/format_markdown.py.
  • Do not substitute a repository-local formatter, a third-party formatter, or a different custom script for this validation step unless the user explicitly asks for that tool instead.
  • Use the narrowest flags that enforce this skill's formatting rules.
  • If no narrower scope applies, use --all.
  • If multiple files changed, validate all of them in the same run.

Flags:

  • --headings for heading numbering only
  • --tables for Markdown table formatting only
  • --wrap for prose wrapping only
  • --spacing for spacing cleanup only
  • --all for a full formatting pass

Behavior:

  • In review-only mode, report formatter drift but do not write changes.
  • In edit modes, run --check first.
  • Show the user all formatter warnings and drift before any write action.
  • When --headings or --all are used with --check, review any warnings about missing or mis-numbered heading numbering before asking the user.
  • Ask the user before running any write-mode formatter pass.
  • If drift is found in files already within scope and the user approves, run the formatter in write mode on those files before finishing the task.
  • If formatter drift would cause broad unrelated churn, report it and limit write-mode formatting to the files already in scope.

Write mode:

python <this-skill>/subskills/documentation-formatting/scripts/format_markdown.py [flags] <file.md> [...]

4.9. Final report

At the end, report:

  • which files were reviewed
  • which files were changed
  • what was corrected
  • any important inconsistencies found
  • any related files left unchanged due to scope limits
  • anything not changed because the codebase did not support it

Use concrete summaries such as:

  • Removed outdated command examples from README.md
  • Updated env var naming to CLOUDFLARE_API_TOKEN across docs
  • Corrected API behavior in docs/API.md to match current implementation
  • Noted stale references in docs/DEPLOYMENT.md but left them unchanged because they were outside the requested scope

5. Documentation rules

5.1. General writing rules

Use:

  • present tense
  • concise, factual wording
  • repo terminology and actual names from the codebase
  • links to canonical docs instead of repeating detailed explanations
  • only commands, paths, values, fields, and examples verified from the codebase

Prefer:

  • short paragraphs
  • lists for scanability
  • explicit file names, commands, and values when verified

Avoid:

  • speculative statements
  • marketing language
  • historical narration
  • TODO-style prose in user-facing docs unless explicitly requested
  • duplicate explanations across multiple files

5.2. Current state only

Except for CHANGELOG.md and clearly archived or historical documents:

  • describe only what exists now
  • remove outdated content
  • remove completed TODO notes
  • remove future-looking statements unless the user explicitly asked for a roadmap or proposal

5.3. No past or version references

Outside CHANGELOG.md and explicitly historical documents:

  • do not use version tags in prose
  • do not compare current behavior to previous behavior
  • do not describe changes in before/after form
  • state current behavior only

Allowed exception:

  • CHANGELOG.md may retain release-history structure and version headers

5.4. No duplicate content

Maintain a single source of truth.

Default content boundaries:

  • README.md = overview, setup, basic usage, doc navigation
  • architecture docs = current structure, responsibilities, system behavior
  • config docs = configuration reference and operational details
  • API docs = interface contract and usage
  • CHANGELOG.md = release history

Rules:

  • do not repeat the same detailed explanation in multiple docs
  • keep brief mentions short and link to the canonical document
  • do not add self-links such as a README linking to itself
  • remove duplicated resolved-decision or acceptance-criteria content unless the user explicitly wants it preserved

5.5. Consistency

Keep names, values, and wording aligned across docs.

  • use the same env var names everywhere
  • use the same file names and paths everywhere
  • use the same terminology for the same concept everywhere
  • do not hardcode conflicting values in multiple places
  • when exact values come from code or config, match them exactly
  • where values are configurable, describe them as configurable instead of copying possibly divergent values into multiple docs

5.6. Documentation config maintenance

When docs/docs-config.json exists and the task is in scope:

  • keep doc paths aligned with actual files
  • add newly created docs to the config
  • remove deleted docs from the config
  • update doc status when a doc becomes frozen, archived, or generated
  • keep each doc's topics aligned with what the document materially covers, not just a single filename-derived label
  • keep topic_map aligned with canonical topic ownership
  • do not require every topic in topics to appear in topic_map; shared coverage topics may appear in multiple docs without a single owner
  • do not assign the same canonical topic to multiple docs unless the user explicitly wants that ambiguity
  • treat topics as explicit config entries, not as a hardcoded built-in list
  • use topics for all meaningful topics the doc materially covers
  • use canonical_topics only for the smaller subset the doc owns as the primary source of truth
  • do not create topics from every heading mechanically; use filenames, headings, and document content as evidence for topic candidates, then keep only the meaningful stable topics
  • when headings represent stable areas such as API groups, subsystem domains, workflows, or operational sections, include them as topic candidates instead of collapsing the doc to one generic topic

5.7. Sensitive information

Do not document secrets or sensitive operational values.

  • do not include API keys, passwords, tokens, private certificates, connection strings with credentials, or other secret material in docs
  • do not copy real secret values from .env, config files, CI settings, or deployment systems into Markdown
  • when documentation needs an example, use placeholders such as <API_KEY> or your-token-here
  • if the repository already contains sensitive values in docs, remove or redact them when the task is in scope and mention the issue in the final report

5.8. Formatting

Enforce these Markdown syntax rules in all edited or newly created Markdown files unless the user explicitly requests otherwise:

  • Use dashes (-) for bullet points.
  • Wrap normal prose to about 80-85 characters per line.
  • Do not force-wrap code blocks, inline code, URLs, or table rows in ways that reduce readability.
  • Prefer lists over long comma-separated prose.
  • Do not use 1) or 2) heading styles.

5.9. Headings

Enforce numbered headings in all newly created Markdown files and in edited Markdown files where headings are added, removed, moved, or already numbered, unless the file is explicitly frozen or the user explicitly requests a different style.

Rules:

  • The first heading (#) is the document title and is not numbered.
  • Use numbered headings after the title.
  • Use consistent hierarchical numbering such as:

- ## 1. Heading - ### 1.1. Subheading - #### 1.1.1. Sub-subheading

  • When inserting, removing, or moving sections, renumber affected headings so the numbering remains correct.
  • Do not use 1) or 2) heading styles.

Examples:

  • # My Awesome Software
  • ## 1. Installation
  • ### 1.1. Windows installation
  • #### 1.1.1. Quick installation fix

5.10. Tables

Use tables only when they improve readability.

Use tables for:

  • comparisons
  • field references
  • option matrices
  • compact command or config summaries

Avoid tables for long narrative text.

If a table becomes too wide or awkward in raw Markdown, convert it to a list.

Keep table cells short and factual.

5.11. Diagrams

Preserve useful diagrams when they still reflect the codebase.

If a diagram is stale:

  • update it to match the current structure, or
  • simplify it if the existing detail is noisy or misleading

Do not remove a useful diagram just to replace it with prose by default.

5.12. Behavior when uncertain

  • prefer shorter, factual wording
  • prefer removing questionable claims over preserving them
  • if the codebase does not confirm a claim, do not state it as fact
  • if the user explicitly asks for a style that conflicts with these rules, follow the user for that task while keeping the rest of the docs consistent

6. Invocation examples

Use this skill for requests like:

  • Do a documentation upkeep pass
  • Sync docs with the codebase
  • Review docs for outdated or duplicated content
  • Update README to match the current project behavior
  • Enforce documentation standards on docs/ARCHITECTURE.md
  • Create a new doc that matches the existing documentation style

Do not use this skill for:

  • roadmap or future-plans documents
  • speculative design writing
  • migration narratives or retrospectives
  • release-note narration unless explicitly requested
  • historical comparisons unless explicitly requested

7. Editing standard

When editing documentation:

  • preserve useful existing content when accurate
  • rewrite only as much as needed
  • avoid unnecessary repo-wide style churn
  • prefer correctness over comprehensiveness
  • prefer canonical linking over repeated explanation
  • keep the result easy to scan and maintain
  • update useful topology or ownership diagrams when they still add value
  • mention out-of-scope drift in the final report instead of silently ignoring it

平台分布

Codex

36.8%
按下载量换算41

Claude

31.56%
按下载量换算35

Cursor

16.65%
按下载量换算18

Gemini CLI

9.4%
按下载量换算10

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills