菜单
Extract clean Markdown from web pages via the self-hosted Crawl4AI service at $CRAWL4AI_URL. Use when WebFetch returns junk on JS-rendered or anti-bot pages, or for whole-site ingestion. Keyless — never install a crawler, puppeteer, or hosted Firecrawl.
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。
基于 SOC 职业分类
Prioritize open TODO items by effort and impact, recommend what to work on next. Add the `full` keyword to also weigh project notes and stash context.
Create an isolated git worktree in a new tmux window via `just wt new`. Use when the user asks to use, create, spin up, or work in a worktree in a jolo container.
Browse, screenshot, or visually verify a local web app using the pre-installed browser-check and playwright-cli (never install your own browser). Use when asked to screenshot, browse, open/view a page, check the UI, or confirm the site is running.
Resume a session — read the paste-back handoff note to recall what we last worked on, falling back to a git+TODO status when none exists. Add the `full` keyword to also deep-scan project notes and memory.
Write the paste-back handoff note (scratch/resume-session.md) so the next session can hit the ground running. Add the `full` keyword to also save durable knowledge to denote notes + TODO and commit.
Get oriented on host-level, cross-project knowledge by reading the stash README map. Use when starting fresh in a container, or when you need the lay of the land for hardware, dotfiles/preferences, Emacs, Linux, or cross-project conventions that live outside the current repo.
| name | j-crawl |
| description | Extract clean Markdown from web pages via the self-hosted Crawl4AI service at $CRAWL4AI_URL. Use when WebFetch returns junk on JS-rendered or anti-bot pages, or for whole-site ingestion. Keyless — never install a crawler, puppeteer, or hosted Firecrawl. |
Self-hosted Crawl4AI reachable at $CRAWL4AI_URL (already passed into every
container — do not hardcode the host). Keyless: no API token today, no
per-container secret. NEVER install a crawler, puppeteer, playwright-extra, or
reach for hosted Firecrawl — the endpoint is already there.
WebSearch (built-in). Crawl4AI is not a search engine.WebFetch first. It's cheaper./md (below).
Reach here only when WebFetch returns junk or JS-gated content.playwright-cli (local browser), not Crawl4AI./md callcurl -s -X POST "$CRAWL4AI_URL/md" \
-H 'Content-Type: application/json' \
-d '{"url": "https://example.com", "f": "fit"}'
f: "fit" is the heuristic content filter (default; no LLM call). Omit for default.c entirely. It must be a string if sent ("0"); boolean c: false → HTTP 422.-H "Authorization: Bearer $CRAWL4AI_TOKEN" only if a token is set (it isn't today).{"url": "...", "filter": "fit", "markdown": "# ...", "success": true}
markdown field, gated by success: true — NOT
nested under result.markdown.fit_markdown.# heading.Then synthesize from the returned Markdown.