Token导航 LogoToken导航

Agent Skill · 研究检索

docs-steward(文档管理员)

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。

docs-steward 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,权限为调用外部服务;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 44,154。

总安装 931周安装 40GitHub Stars 2下载 326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:docs-steward(文档管理员)
来源仓库:https://github.com/wyattowalsh/agents
仓库路径:skills/docs-steward
安装命令:
npx skills add https://github.com/wyattowalsh/agents --skill docs-steward
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/wyattowalsh/agents --skill docs-steward

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。

负责维护项目本地文档的质量、架构与技术栈时效性。支持 Starlight、Docusaurus 等主流框架的内容图导航管理。提供非破坏性初始化、重新索引与迁移目标设定功能。输入为自然语言请求或框架名称即可触发相应维护流程。重点在于保持内容与代码变更的同步而非内容生成。

docs-steward 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/docs-steward

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/wyattowalsh/agents --skill docs-steward;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

    它需要调用外部服务,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

docs-steward 怎么安装?

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

docs-steward 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。

docs-steward 需要什么权限?

权限级别是调用外部服务。该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

docs-steward 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。

docs-steward 用的人多吗?

公开数据是累计安装 931 次,近一周安装 40 次,下载 326 次,GitHub 2 星。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。

docs-steward 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

  • 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

  • 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
docs-steward

文档管理员 · 本页

9312调用外部服务未明确通过

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

Docs Steward

Maintain docs quality, architecture, and framework currency in project-local repositories.

Input: $ARGUMENTS — mode keywords, framework names, migration goals, or natural-language docs requests.


Canonical Vocabulary

TermMeaningNOT
docs frameworkPrimary platform rendering docs (Starlight, Docusaurus, Fumadocs, Sphinx, MkDocs)"site generator" (too broad)
theme layerVisual and component skin on top of a framework"framework"
content graphNavigation + page relationships + generated indexes"folder list"
initNon-destructive docs bootstrap in an existing repository"reinitialize/overwrite"
syncRegenerate framework artifacts from source docs state"deploy"
maintainRead-only health checks for structure, links, drift, and build integrity"rewrite"
enhanceImprove existing docs clarity, structure, UX, and discoverability"recreate"
matrix runExecute workflows for all detected frameworks in one repository"auto"
migrationPlanned transition from one docs framework to another with parity checks"instant convert"
version refreshUpdate reference knowledge to latest stable framework/tool versions"blind upgrade"
project-localSkill is installed and used within the current project scope"global install"

Dispatch Table

Route $ARGUMENTS to mode:

$ARGUMENTS patternModeStart at
(empty) or autoAutoMode A
framework <name> <action> (<action> ∈ `syncenhancemaintain
framework <name>Framework-targeted (default action: maintain)Mode B
init / init <name>InitMode I
init-sync / init-sync <name>Init+SyncMode I
sync / sync <name>SyncMode C
enhance / enhance <path>EnhanceMode D
maintain / maintain <name>MaintainMode E
research / research versionsResearchMode F
matrixMatrixMode G
migrate <from> -> <to>MigrateMode H
generate api <module>GenerateMode J
generate adr <decision>GenerateMode J
generate runbook <process>GenerateMode J
generate onboardGenerateMode J
generate glossaryGenerateMode J
Natural language: "generate API docs/reference"GenerateMode J
Natural language: "create an ADR/decision record"GenerateMode J
Natural language: "write a runbook/onboarding guide"GenerateMode J
Natural language: "set up <framework> docs from scratch and keep it in sync"Init+SyncMode I
Natural language: "docs are stale/broken/outdated"MaintainMode E
Natural language: "improve docs UX/content/nav"EnhanceMode D
Natural language: "upgrade latest framework versions"ResearchMode F
Requests to build app APIs, skills, or MCP serversRefuseRedirect

Classification Gate

  1. Detect frameworks using references/framework-detection.md.
  2. If one framework is detected, continue with that framework.
  3. If multiple frameworks are detected and user did not specify target, ask user to choose each run (interactive mode).
  4. If no framework signal exists, ask user for docs stack before editing (interactive mode).
  5. If the run is headless/non-interactive, apply the Headless Fallback Contract.

Headless Fallback Contract

When no clarifying exchange is possible:

  1. If framework target is ambiguous or multiple frameworks are detected without explicit target, run Mode E (Maintain) as a read-only Mode G (Matrix) across detected frameworks.
  2. Only allow mutating auto paths when framework selection is explicit or single-framework high-confidence and trigger category is safe per Mode A.
  3. Process frameworks in deterministic order: starlight/astro, docusaurus, fumadocs, sphinx, mkdocs.
  4. Emit a warning when headless fallback was applied and mutating modes (init, init-sync, sync, enhance, research, migrate) were skipped pending explicit framework selection.
  5. If no framework is detected, return a read-only maintain report with "framework target required" and no edits.

Live Version Research (Required for "latest")

When user requests latest versions (explicitly or implicitly), always refresh version facts before applying changes:

  1. Query package registries (npm/PyPI) with tool-assisted checks.
  2. Update framework reference snapshots in references/*.md.
  3. Record version source and date in the edited reference.
  4. Only then propose dependency or config updates.

Never claim "latest" without evidence from current registry data.


Mode A: Auto

Default orchestrator mode.

A.1 Trigger categories

Classify detected change signals before choosing a mode:

  • Content-only: page prose/examples/frontmatter edits without nav/config/build impact.
  • Structure/config: docs/framework/config/navigation/build-related changes (sidebar/nav trees, framework config, docs build wiring, generated docs artifacts).
  • Dependency/version: docs framework/theme/plugin dependency or lockfile/version changes.

A.2 Action path

  1. Detect docs frameworks.
  2. If multiple frameworks, ask which one to operate on this run.
  3. Route by trigger category:

- Content-only -> Mode D (Enhance), then optional Mode E (Maintain) check. - Structure/config -> Mode C (Sync) + Mode E (Maintain). - Dependency/version -> Mode F (Research) + Mode C (Sync) + Mode E (Maintain).

  1. If trigger category or intent is unclear, ask one focused clarifying question before edits.

A.3 Auto-sync safety

  1. Auto-sync is allowed only when framework target is explicit or single-framework high-confidence.
  2. In headless ambiguous multi-framework runs, preserve read-only fallback (Mode E matrix); do not auto-sync.
  3. If confidence is low, framework signals conflict, or scope is not docs-local, downgrade to Mode E (Maintain).

Mode B: Framework-targeted (framework <name> [<action>])

Supported names:

  • astro / starlight
  • docusaurus
  • fumadocs
  • sphinx
  • mkdocs

Grammar: framework <name> [<action>], where <action> is one of sync, enhance, maintain, research, or migrate. If <action> is omitted, default to maintain. Run the mapped mode for the selected framework: sync -> Mode C, enhance -> Mode D, maintain -> Mode E, research -> Mode F, migrate -> Mode H.


Mode C: Sync

Bring generated docs artifacts into a consistent state.

  • Verify framework-native advanced component support and rendering paths (Mermaid, code snippets, tables, embeds, tabs/admonitions).

C.1 Astro + Starlight

  • Verify Astro config and Starlight integration.
  • Regenerate docs artifacts with project commands (for this repo: uv run wagents docs generate).
  • Build-check docs output.

C.2 Docusaurus

  • Validate docusaurus.config.*, sidebars, and docs route structure.
  • Rebuild generated docs assets and run build sanity checks.

C.3 Fumadocs

  • Validate next.config.*, fumadocs-* package setup, and MDX content tree.
  • Regenerate indexes/navigation where applicable and run Next build checks.

C.4 Sphinx

  • Validate conf.py, extension set, and theme package alignment.
  • Build with strict warnings enabled for docs quality gates.

C.5 MkDocs

  • Validate mkdocs.yml, plugin stack, and nav structure.
  • Run strict build checks and detect plugin drift.

C.6 Sync output contract

  • Include an "Advanced component render check" summary for Mermaid, code snippets, tables, embeds, and tabs/admonitions.
  • Note framework-specific support/plugins used and flag unsupported components with safe fallbacks.
  • Include accessibility notes (diagram text alternatives, labeled code fences, titled embeds, and tab/admonition semantics).

Mode D: Enhance

Improve docs quality without changing framework identity.

Enhancement targets:

  • Information architecture (nav clarity, section depth, landing pages)
  • Writing quality (scannability, examples, API/task orientation)
  • Internal linking (orphan reduction, related-links strategy)
  • Visual documentation UX (callouts, tabs, code grouping, admonitions, diagrams, embeds, tables)
  • Consistency (style, heading depth, frontmatter, metadata)

Ask clarifying questions when enhancement direction is ambiguous ("developer docs", "marketing docs", "API docs", or "tutorial docs").

D.1 Advanced component strategy

  • Use advanced components when they materially improve comprehension, not decoration.
  • Choose framework-native primitives for the active stack and keep syntax/style consistent within each page.
  • Prefer: Mermaid for flows/architecture, code snippets for implementation steps, tables for comparisons, embeds for canonical demos/media, tabs/admonitions for variants and cautions.
  • Keep outputs accessible: add plain-language context and text fallback for diagrams, language labels on code blocks, meaningful captions/titles, and avoid color-only or tab-only critical content.
  • Use references/advanced-components.md for framework-specific syntax patterns and safe fallbacks.

D.2 Enhance output contract

  • Summarize which advanced components were added/updated, why they help, and any framework constraints or fallbacks.

Mode E: Maintain

Read-only diagnostics and remediation planning.

Checks:

  1. Broken links and anchor drift
  2. Stale/generated file mismatch
  3. Navigation dead-ends and orphans
  4. Theme/plugin dependency drift
  5. Build warnings/errors
  6. Framework mismatch in mixed repos

Output format:

  • Critical (must-fix)
  • Warning (should-fix)
  • Suggestion (nice-to-have)
  • Next commands to run

Mode F: Research

Refresh framework/theme references to latest stable versions and advanced patterns.

Workflow:

  1. Resolve package/version facts for each active framework.
  2. Update reference snapshot sections.
  3. Note migration-relevant deltas (breaking changes, deprecated APIs, config shifts).
  4. Return a concise change summary with confidence and citations/source commands.

Mode G: Matrix

Execute a controlled run across all detected frameworks in the repository.

Use for:

  • Monorepos with multiple docs stacks
  • Parallel migration programs
  • Framework parity audits

Matrix run output must keep results grouped per framework and include cross-framework conflicts.


Mode H: Migrate (migrate <from> -> <to>)

Migration is in scope for v1.

Supported paths:

  • Docusaurus -> Fumadocs
  • Sphinx -> MkDocs
  • Sphinx -> Starlight
  • MkDocs -> Docusaurus

Migration phases:

  1. Inventory and parity baseline
  2. Content and nav mapping
  3. Theme/component mapping
  4. Build/test parity checks
  5. Incremental rollout plan

Do not promise one-shot full conversion; prefer staged migration with checkpoints.


Mode I: Init (init, init <framework>, init-sync, init-sync <framework>)

Bootstrap docs site wiring in an existing repository without destructive rewrites. This mode applies to explicit init* commands and implicit "bootstrap docs + keep synced" requests in existing codebases. Load references/init-sync-existing-repos.md plus the selected framework reference before edits.

I.1 Framework selection behavior

  1. If <framework> is provided, target that framework directly.
  2. If no <framework> is provided and exactly one framework is detected, target the detected framework.
  3. If multiple frameworks are detected and no explicit target is provided, ask user to choose; in headless runs, apply the Headless Fallback Contract.
  4. If no framework is detected and no explicit target is provided, ask user to select a supported framework; in headless runs, return read-only Mode E output with "framework target required."

I.2 Non-destructive bootstrap contract

  • Only create missing docs scaffold/config files; do not overwrite existing files by default.
  • Preflight guard: if target docs root already exists and is non-empty, do not recreate or replace that root; continue in-place with missing-file scaffolding only unless user explicitly requests overwrite.
  • Preserve existing docs content, navigation, and framework config when present.
  • If requested framework conflicts with existing docs framework wiring, stop and return maintain findings + conflict summary (use Mode H for migrations).

I.3 Follow-up flow

  • init: bootstrap, then recommend sync and maintain follow-up.
  • init-sync: bootstrap, then run Mode C (Sync) and Mode E (Maintain) in the same run.
  • Always report what was created, what was skipped, and why.

Mode J: Generate (generate <type> [<target>])

Generate technical documentation from source code. Load references/generate-mode.md for detailed procedures, output formats, and script usage.

Sub-modes:

CommandOutputKey Script
generate api <module>API reference with signatures, docstrings, coverageapi-surface-extractor.py, doc-coverage-analyzer.py
generate adr <decision>MADR-format architecture decision recordadr-scaffolder.py
generate runbook <process>Operational runbook with commands from codebase—
generate onboardNew contributor onboarding guide—
generate glossaryTerm definitions extracted from code and docs—

J.1 Workflow

  1. Identify target scope (module path, process name, or whole repo).
  2. Run applicable scripts to extract structured data from source.
  3. Transform extracted data into documentation following the format in references/generate-mode.md.
  4. Present draft for review. Ask clarifying questions if context is ambiguous.
  5. Write output to appropriate location (e.g., docs/api/, docs/decisions/).

J.2 Generate output contract

  • Generated docs must trace to source code (include file paths and line numbers).
  • Never fabricate API signatures or docstrings; extract or flag as undocumented.
  • ADRs follow MADR v3 format from data/adr-template.json.
  • Use docstring format conventions from data/docstring-formats.json matching the project's language and style.

Scope Boundaries

In scope: Docs framework operations, non-destructive docs bootstrap, version refresh, build/health diagnostics, quality enhancement, docs migrations, technical documentation generation from code.

NOT for:

  • Creating/editing unrelated product features or backend APIs
  • Creating new skills, agents, or MCP servers
  • CI/CD redesign outside docs pipeline needs
  • Non-docs frontend application implementation

Reference File Index

Load references on demand; do not load all at once.

FileContentLoad When
references/framework-detection.mdFramework signal map, multi-framework routing rulesAll modes
references/advanced-components.mdFramework-specific Mermaid/codeblocks/tables/embeds patterns with safe fallbackssync/enhance
references/init-sync-existing-repos.mdNon-destructive framework-aware bootstrap + follow-up sync workflows for existing reposinit/init-sync
references/astro-starlight.mdAstro + Starlight advanced setup and checksframework/sync/maintain/enhance
references/docusaurus.mdDocusaurus advanced config and plugin checksframework/sync/maintain/enhance
references/fumadocs.mdFumadocs + Next advanced setup and checksframework/sync/maintain/enhance
references/sphinx.mdSphinx + theme matrix (Shibuya/PyData/Furo/Book)framework/sync/maintain/enhance
references/mkdocs.mdMkDocs + Material/plugin stack checksframework/sync/maintain/enhance
references/migrations.mdMigration playbooks and parity templatesmigrate/matrix
references/version-refresh.mdLatest-version refresh workflow and evidence rulesresearch + any latest-version request
references/generate-mode.mdGenerate sub-mode procedures, output formats, script usagegenerate
data/docstring-formats.jsonDocstring format standards per language (Google, NumPy, Sphinx, JSDoc, TSDoc)generate api
data/adr-template.jsonMADR v3 template structure and file naming conventionsgenerate adr

Critical Rules

  1. Ask at least one clarifying question before edits when user intent or target framework is ambiguous.
  2. In multi-framework repositories, auto mode must ask which framework to operate on each run unless explicitly overridden.
  3. Never claim "latest" versions without fresh registry-backed evidence and a reference snapshot update.
  4. Keep operations project-local by default; do not assume global install context.
  5. Run framework-appropriate build or validation checks after docs changes.
  6. Maintain mode is read-only: diagnose first, then propose concrete fixes.
  7. Migration mode must be phased and reversible; avoid destructive one-pass rewrites.
  8. Refuse out-of-scope requests and route to the correct specialized skill.
  9. Init mode is non-destructive in existing repos: create missing files only unless user explicitly requests overwrite.
  10. Generate mode must extract from source code; never fabricate signatures, docstrings, or API details.

平台分布

Codex

37.14%
按下载量换算121

Claude

29.46%
按下载量换算96

Cursor

18.23%
按下载量换算59

Gemini CLI

10.24%
按下载量换算33

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

调用外部服务

该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

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