AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

firecrawl-workflows

Run focused Firecrawl workflows for research, documentation lookup, SEO audits, QA, lead discovery and enrichment, business contacts, current jobs, fantasy decisions, brand assets, knowledge bases and website design systems. Use when the user wants a sourced answer or deliverable from a workflow rather than raw extraction or product-code integration.

Original source
firecrawl
License
ISC
Pinned revision
2bf714d9f45e85085036a533ceb943c8577aef2b

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

Ask your assistant

Use Agentic Memory skills to read firecrawl--firecrawl-workflows at revision 2bf714d9f45e85085036a533ceb943c8577aef2b, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 2

references/workflow-authoring.md

Workflow Authoring

Use this reference when adding or reviewing workflow skills.

Workflow Checklist

  • Name the real user outcome in the first paragraph.
  • Define only the blocking onboarding questions.
  • Tell the agent what artifacts to gather with Firecrawl.
  • Specify the final deliverable shape.
  • Include an evidence or citation expectation when claims come from websites.
  • Identify work that can run in parallel.
  • Keep instructions generic enough for any coding agent harness.

Harness-Agnostic Language

Use:

  • "ask the user a clarifying question"
  • "use sub-agents if available"
  • "run independent page research in parallel"
  • "use Firecrawl through the CLI or equivalent Firecrawl tools"

Avoid:

  • editor-specific function names
  • assumptions about a single modal UI
  • hardcoded sub-agent APIs
  • output paths that only one harness can access

Lightweight Onboarding

Do not run a long interview by default. First infer from the user's message, files, URLs, and surrounding context. If the agent can safely start, start.

Ask at most 1-3 concise clarifying questions only when a required input is missing or ambiguity would materially change the work. Prefer defaults for non-blocking choices and state them briefly.

Automation Inputs

For recurring jobs, each workflow should be expressible as:

workflow: skill-name
cadence: weekly
inputs:
  subject: example
  urls: []
  competitors: []
  output: report.md

The skill does not need to implement a scheduler. It should make the inputs and outputs stable enough for another system to schedule it.

Reference Repos

  • anthropics/skills: use simple skill directories and minimal metadata.
  • anthropics/financial-services: borrow outcome orientation and optional agent packaging ideas.
  • anthropics/claude-for-legal: borrow vertical workflow organization and concrete deliverable framing.

Do not copy heavier managed-agent scaffolding unless this repo starts shipping standalone hosted agents.