Skip to content

AGENTS.md

Latest commit

 

History

History
79 lines (61 loc) · 2.96 KB

AGENTS.md

File metadata and controls

79 lines (61 loc) · 2.96 KB
Raw

Firecrawl Workflows Repo

Distribution repo for Firecrawl-powered workflow skills.

Safe To Edit

Current authored skills:

  • skills/firecrawl-workflows/
  • skills/firecrawl-website-design-clone/
  • skills/firecrawl-deep-research/
  • skills/firecrawl-seo-audit/
  • skills/firecrawl-lead-research/
  • skills/firecrawl-qa/
  • skills/firecrawl-competitive-intel/
  • skills/firecrawl-company-directories/
  • skills/firecrawl-dashboard-reporting/
  • skills/firecrawl-knowledge-base/
  • skills/firecrawl-knowledge-ingest/
  • skills/firecrawl-lead-gen/
  • skills/firecrawl-market-research/
  • skills/firecrawl-research-papers/
  • skills/firecrawl-demo-walkthrough/
  • skills/firecrawl-shop/

Plugin metadata and top-level docs are also safe to edit:

  • .cursor-plugin/
  • .claude-plugin/
  • .codex-plugin/
  • README.md
  • AGENTS.md
  • references/

Intent

Use this repo when an agent should produce a real deliverable with Firecrawl, such as:

  • website design-system extraction
  • recurring market scans
  • SEO audits
  • QA reports
  • research briefs from web evidence
  • literature reviews over published papers, including biomedical, clinical, and other life-science topics, through the paper index rather than web search
  • knowledge bases
  • lead lists
  • research-backed content or planning artifacts
  • other structured workflows that demonstrate Firecrawl in action

Do not use this repo as the main source for:

  • one-off web search or scraping during the current session
  • product-code Firecrawl integrations
  • CLI command reference

Those belong to the core CLI skills and build skills.

Authoring Rules

  • Keep each SKILL.md concise and trigger-oriented.
  • Keep onboarding lightweight: infer from context first, then ask at most 1-3 blocking questions only if needed.
  • Stay harness-agnostic; do not hardcode Claude Code, Cursor, Codex, Hermes, or any specific tool/function API.
  • Prefer wording like "ask the user a clarifying question" or "use sub-agents if available."
  • Target a named deliverable with a clear output format.
  • Make workflows automation-friendly by defining required inputs, steps, artifacts, and verification.
  • Flag parallelizable work generically when competitors, URLs, pages, or sources can be processed independently.
  • Use Firecrawl through the CLI or equivalent Firecrawl tool surface available in the host environment.
  • Keep the paper index and web search distinct. Paper work goes through firecrawl_research_* (MCP) or firecrawl research (CLI); categories: ["research"] on Firecrawl search is a website filter — it points ordinary web search at research-affiliated sites, PubMed, bioRxiv, medRxiv, and arXiv among them, and returns their web pages rather than paper records from that index.

Repo Shape

Expected layout:

  • skills/<skill-name>/SKILL.md for each skill
  • references/ for shared authoring and packaging guidance
  • plugin metadata under .cursor-plugin/, .claude-plugin/, and .codex-plugin/

Keep supporting references one level deep.