generated: '2026-08-29'
method: searched
source: >-
  https://github.com/unclecode/crawl4ai-cloud-sdk/tree/main/python/claude-plugin/skills
  (provider-published, saved verbatim) plus one API Evangelist skill for the flow
  the provider does not cover
provider: Crawl4AI
providerId: crawl4ai
description: >-
  Crawl4AI PUBLISHES ITS OWN AGENT SKILLS. Four Claude Code skills ship inside the
  first-party crawl4ai-cloud-sdk repository, distributed through its own
  .claude-plugin/marketplace.json alongside a local stdio MCP server. They are
  saved here verbatim, unedited.
skills:
  - file: crawl4ai-crawl4ai-provider-skill.md
    name: crawl4ai
    method: searched
    provider_published: true
    source: >-
      https://github.com/unclecode/crawl4ai-cloud-sdk/blob/main/python/claude-plugin/skills/crawl4ai/SKILL.md
    description: >-
      "Web crawling, screenshots, data extraction, URL discovery. Use when the user
      asks to crawl, scrape, extract data from, screenshot, or discover URLs on a
      website." Wraps markdown / screenshot / extract / map / site / crawl / async
      plus job polling and key validation.
    references:
      - reference/crawl4ai-endpoints.md
      - reference/crawl4ai-errors.md
  - file: crawl4ai-setup-provider-skill.md
    name: crawl4ai:setup
    method: searched
    provider_published: true
    source: >-
      https://github.com/unclecode/crawl4ai-cloud-sdk/blob/main/python/claude-plugin/skills/setup/SKILL.md
    description: First-run configuration of the plugin (key, mode).
  - file: crawl4ai-switch-provider-skill.md
    name: crawl4ai:switch
    method: searched
    provider_published: true
    source: >-
      https://github.com/unclecode/crawl4ai-cloud-sdk/blob/main/python/claude-plugin/skills/switch/SKILL.md
    description: >-
      Switch the agent between the cloud API and a local pip-installed crawler,
      rewriting ~/.crawl4ai/claude_config.json and verifying with a live crawl.
  - file: crawl4ai-update-provider-skill.md
    name: crawl4ai:update
    method: searched
    provider_published: true
    source: >-
      https://github.com/unclecode/crawl4ai-cloud-sdk/blob/main/python/claude-plugin/skills/update/SKILL.md
    description: Update the plugin and its dependencies.
  - file: crawl4ai-bulk-crawl-job.md
    name: crawl4ai-bulk-crawl-job
    method: generated
    provider_published: false
    source: 'https://gate.crawl4ai.com/docs, https://gate.crawl4ai.com/llms.txt'
    description: >-
      Submit / poll / page / retry a 10,000-URL async job on the gate Cloud API —
      the surface the provider's own skills do not cover, and the one with no
      cancel operation and no idempotency key.
reference_files:
  - file: reference/crawl4ai-endpoints.md
    method: searched
    provider_published: true
    source: >-
      https://github.com/unclecode/crawl4ai-cloud-sdk/blob/main/python/claude-plugin/skills/crawl4ai/reference/endpoints.md
    note: >-
      The most complete API reference Crawl4AI publishes anywhere — every v1
      endpoint with parameter tables, defaults and response shapes. It is the
      closest thing the company has to an OpenAPI, and it lives inside a skill.
  - file: reference/crawl4ai-errors.md
    method: searched
    provider_published: true
    source: >-
      https://github.com/unclecode/crawl4ai-cloud-sdk/blob/main/python/claude-plugin/skills/crawl4ai/reference/errors.md
    note: 'Status-by-status error reference, rate-limit headers and retry strategy.'
grounding_note: >-
  No operationIds are cited anywhere in these skills because Crawl4AI publishes no
  OpenAPI for either live surface. Every path, parameter and status quoted in the
  generated skill is verbatim from the provider's own documentation.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com