Route, extract, and validate a scraping job (URL or local file) via the universal-scraping-architect skill — refuses to deliver unvalidated data.
复制下面这句话,粘贴给 Claude Code、Codex、Cursor 等 AI 编程工具,它会读取安装说明并在你确认后完成安装。
请阅读 https://ai.atlankj.com/install/asset/gh-claude-skills-9d48eedfe57c ,按照其中的说明把「cs-scrape」安装到你(当前 AI 工具)中。执行前先告诉我将运行的命令和写入的位置,等我确认。
查看 AI 将读取的安装说明正在读取 GitHub 原文…
内容来自 GitHub 原始文件,由原作者维护。在 GitHub 查看
Run a gated extraction pipeline for $ARGUMENTS using skills/universal-scraping-architect/SKILL.md.
$ARGUMENTS is empty, ask for the URL or file path plus the desired output format — do not guess.robots.txt and plan rate limits; refuse disallowed targets.os.getenv('FIRECRAWL_API_KEY'); if a key appears inline anywhere, fix that first.Route — state the mode and why (per the skill's routing rules): Mode 1 Firecrawl (public/JS-heavy URL, bulk crawl) · Mode 2 local Python (local files, private data, simple static HTML) · Mode 3 hybrid (Firecrawl extract + pandas clean).
Budget — estimate API quota / token limits before multi-page jobs; add checkpointing + pagination.
Extract — start from the matching runner template (run from the plugin root; --sample previews the summary shape offline):
python3 skills/universal-scraping-architect/scripts/firecrawl_example.py --sample
python3 skills/universal-scraping-architect/scripts/local_bs4_example.py --sample
Validate (mandatory, exit-code gated):
python3 skills/universal-scraping-architect/scripts/validate_extraction.py extracted_output.json --json
status: ok) → continuewarning = empty output, error = malformed JSON) → fix and re-extract; never deliver unvalidated dataThen check required fields and duplicates against the job spec.
Deliver — CSV (tabular) / JSON (nested) / Markdown (docs, chunked), per the user's requested format, with a summary of mode chosen, row counts, empty values, and the validation verdict.