菜单
Use when a command takes more than ~1 minute, needs to persist (servers, watchers), or when running it in the background lets you continue making progress on other work in parallel. Examples: dev servers, long builds, background subagents. Do not use for short one-shot commands (typecheck, lint, quick tests) — just run those directly.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Fetch a web page and return its content as clean markdown. Use whenever you need to read any web content — documentation, articles, APIs, READMEs, blog posts. Always returns content to stdout; falls back gracefully through multiple extraction methods. Use --discover to explore a site's structure before fetching specific pages.
Use when a task is parallelizable, would consume many tokens (large file reads, codebase exploration, summarization), or involves grunt work that doesn't need your full reasoning ability. Spawn a sub-agent instead of doing it yourself.
| name | background |
| description | Use when a command takes more than ~1 minute, needs to persist (servers, watchers), or when running it in the background lets you continue making progress on other work in parallel. Examples: dev servers, long builds, background subagents. Do not use for short one-shot commands (typecheck, lint, quick tests) — just run those directly. |
Always use tmux. Do not use & or nohup; they won't persist across tool calls.
Start: tmux new-session -d -s mybuild 'npm test; tmux wait-for -S mybuild-done'
Wait: tmux wait-for mybuild-done
Capture: tmux capture-pane -t mybuild -p -S -
Kill: tmux kill-session -t mybuild
Tips:
tmux wait-for -S signals, tmux wait-for (no flag) blocks — use them as a pairremain-on-exit on if you need to capture output after the command exits-S - on capture-pane gets full scrollback, not just the visible pane