AgentSkillsCN

read-software-docs

阅读并导航软件包文档网站。当用户提供了文档网址，并希望了解某个API、SDK、框架或库时使用此功能。可通过“阅读[网址]上的文档”“在[软件包]文档中查找[主题]”“根据文档，[API]究竟有什么作用”或任何需要获取并理解在线软件文档的任务来触发。支持任何基于HTML的文档网站，包括Sphinx、Doxygen、ReadTheDocs、MkDocs，以及自定义文档站点。常见目标包括NVIDIA Isaac Sim、Omniverse Kit、OpenUSD、ROS 2、PyTorch，以及其他类似的软件包。

SKILL.md
--- frontmatter
name: read-software-docs
description: Read and navigate software package documentation websites. Use when the user provides a documentation URL and wants to understand an API, SDK, framework, or library. Triggers on requests like "read the docs at [url]", "look up [topic] in the [package] documentation", "what does [API] do according to the docs", or any task requiring fetching and comprehending online software documentation. Supports any HTML-based documentation site including Sphinx, Doxygen, ReadTheDocs, MkDocs, and custom doc sites. Common targets include NVIDIA Isaac Sim, Omniverse Kit, OpenUSD, ROS 2, PyTorch, and similar software packages.

Read Software Docs

Fetch, parse, and navigate software documentation websites to answer user questions or build understanding of APIs, SDKs, and frameworks.

Script

scripts/fetch_docs.py — run with the project venv:

code
.venv/bin/python3 <skill-dir>/scripts/fetch_docs.py <mode> <url> [options]

Modes:

ModePurposeKey Options
pageFetch a single page as clean markdown--output FILE
linksList all internal doc links on a page--filter REGEX
sitemapCrawl from a root URL, build page tree--depth N (default 1), --filter REGEX, --output FILE

Dependencies (beautifulsoup4, html2text, requests) must be installed in the .venv.

Workflow

1. Orient: discover the doc structure

Start with links mode on the user's URL (or the doc site index) to see available pages:

bash
.venv/bin/python3 scripts/fetch_docs.py links <url>

Use --filter to narrow results when the link list is large:

bash
.venv/bin/python3 scripts/fetch_docs.py links <url> --filter "api|reference|python"

For broader discovery, use sitemap with shallow depth:

bash
.venv/bin/python3 scripts/fetch_docs.py sitemap <url> --depth 1

2. Read: fetch specific pages

Fetch pages identified in step 1:

bash
.venv/bin/python3 scripts/fetch_docs.py page <page-url>

For large pages, save to a file and read selectively:

bash
.venv/bin/python3 scripts/fetch_docs.py page <page-url> --output .cache/knowledge/doc_page.md

3. Drill down: follow links as needed

If the initial page references sub-pages or related APIs, fetch those too. Repeat steps 1-2 to navigate deeper into the documentation tree.

4. Synthesize

After reading the relevant pages, produce the deliverable the user requested — this could be:

  • •A summary of an API or module
  • •A how-to guide assembled from multiple doc pages
  • •Answers to specific questions about the software
  • •Code examples based on the documented API

Save research output to .knowledge/ for easy user access (same pattern as read-arxiv-paper).

Tips

  • •WebFetch fallback: If fetch_docs.py struggles with a page (e.g., JavaScript-rendered content), fall back to the WebFetch tool which may handle it differently.
  • •API docs: For Python/C++ API reference pages, use --filter with module names to quickly find the right page (e.g., --filter "omni.isaac.core|isaacsim").
  • •Version pinning: Many doc sites include version numbers in URLs. Confirm the user wants the version in the URL before deeply reading docs for a different version.
  • •Caching: Save fetched pages to .cache/knowledge/ to avoid re-fetching during a session. Use descriptive filenames like .cache/knowledge/isaacsim_core_api.md.