web-search
Search the internet with Google or DuckDuckGo, inspect results, and extract selected pages as readable Markdown. Use when answers require current web sources, JavaScript-rendered pages, or browser access that can handle bot protection. Prefer dedicated tools such as GitHub CLI for GitHub data and curl for direct files or simple URLs.
来源信息
- 仓库
- ogulcancelik/agent-skills
- 最近来源活动
- 2026年9月2日 23:55
- 检测到的 SKILL.md 语言
- 英语
- 星标
- 80
- 分支
- 1
安装方式
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
检查来源文件
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
文件资源管理器
26 个文件正在显示 SKILL.md
SKILL.md
来源说明 · 只读预览- name
- web-search
- description
- Search the internet with Google or DuckDuckGo, inspect results, and extract selected pages as readable Markdown. Use when answers require current web sources, JavaScript-rendered pages, or browser access that can handle bot protection. Prefer dedicated tools such as GitHub CLI for GitHub data and curl for direct files or simple URLs.
- license
- MIT
- compatibility
- Requires Node.js 20.19+, Bun for dependency installation, network access, and a local Chromium-family browser.
# Web Search
Use the bundled CLI to search through a local browser, visit results, and extract pages as clean Markdown for agent context. The persistent browser daemon reduces bot-detection failures and avoids repeated browser startup costs.
The entry point is `{baseDir}/web-search.js`, where `{baseDir}` is the absolute directory containing this `SKILL.md`. Always invoke that absolute path. Never run `./web-search.js` from the caller's project directory.
## Setup
If dependencies are missing, run `bun install` in the skill directory.
The CLI auto-detects Chromium-family browsers. Override detection with
`WEB_SEARCH_BROWSER_BIN` or `--browser-bin <path>`.
## Search and fetch
```bash
{baseDir}/web-search.js "query"
{baseDir}/web-search.js "query" -n 10
{baseDir}/web-search.js --from <result-set-id> --fetch 1,3,5
{baseDir}/web-search.js --url https://example.com
{baseDir}/web-search.js --url https://example.com --full
```
Each search prints a result-set ID. Always pass that ID with `--from` when fetching numbered results; result sets expire after ten minutes. Use `--full` only when the complete page text is needed.
## Browser daemon
Direct calls use a local daemon by default for a warm browser session.
```bash
{baseDir}/web-search.js --daemon status
{baseDir}/web-search.js --daemon start
{baseDir}/web-search.js --daemon stop
{baseDir}/web-search.js --daemon restart
{baseDir}/web-search.js --no-daemon --url https://example.com
```
## Troubleshooting
Treat one blocked page as a normal site-specific failure. Retry once, use another result, or try the direct URL before changing browser state.
For repeated timeouts, daemon errors, or blocks across multiple attempts:
1. Check `{baseDir}/web-search.js --daemon status`.
2. Run `{baseDir}/web-search.js --daemon restart` and retry.
3. If blocks persist, inspect `health.profileDir` in the status output. Stop the daemon, delete only that hidden skill profile directory, then start the daemon again. This resets stale cookies and browser storage without touching the user's normal browser profile.
Never clear the profile after a single block. Profile reset is a last resort for persistent failures.
## Reporting
Use the URLs actually searched or fetched as sources. Cite those URLs when
answering from web material.
在 GitHub 查看