This is Firecrawl's nuclear option for web scraping: a browser automation tool for when standard scraping hits a wall. You scrape a page first, then use natural language to click buttons, fill forms, or navigate multi-step flows to get the data you actually need. It's the last step in Firecrawl's escalation pattern (search, scrape, map, crawl, then finally instruct), meant for content locked behind JavaScript interactions or login flows. The docs are clear about when not to use it: never for web searches, only when simpler methods fail. At 3.3K installs it's the least used tool in the Firecrawl suite, which makes sense given it's intentionally positioned as a last resort.
npx -y skills add firecrawl/cli --skill firecrawl-instruct --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
sentry/dev
degausai/wonda
github/awesome-copilot
insforge/agent-skills
github/awesome-copilot
breaking-brake/cc-wf-studio