CrewAI
SkillAI & models"Routes agents using or contributing to CrewAI, including crews,
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the CrewAI skill
What this skill tells your AI
The instructions your AI receives, as published by vectorspacelab/arex-skill in skills/repositories/repo-skills/crewai/SKILL.md and read by ahel’s review.
Use this repo skill when the task involves CrewAI, the crewai Python package, the crewai CLI, official CrewAI tools, CrewAI Flows, memory/knowledge/RAG, multimodal file inputs, observability hooks, or contributing to the CrewAI monorepo.
Start Here
- Read Repository provenance before deciding whether this skill matches a current CrewAI checkout or needs refresh.
- Read Installation and dependencies when choosing package extras, optional dependencies, Python versions, or inspection commands.
- Read Troubleshooting for cross-cutting install/import, optional dependency, credentials, CLI/project-root, telemetry, and repo-development failures.
- Run check_crewai_environment.py for a safe local import/version/CLI diagnostic; it does not run crews, call LLMs, start MCP servers, or use credentials.
Route by Task
- Use core-runtime for
Agent,Task,Crew,Process, kickoff modes, outputs, guardrails, callbacks, planning, reasoning, checkpoint basics, JSONC crew definitions, and direct-code crew design. - Use flows-and-events for
Flow,@start,@listen,@router, state, routing labels, plotting, persistence, checkpointing, human feedback, event listeners, and event ordering. - Use cli-and-projects for
crewai create, JSON-first and classic project scaffolds,crewai run,train,test,replay,chat,deploy,uv, project templates, and CLI troubleshooting. - Use tools-and-mcp for official
crewai_tools, customBaseToolorCrewStructuredToolwork, MCP adapters, tool publishing, integration tools, optional packages, and credential boundaries. - Use memory-knowledge-and-rag for
Memory,Knowledge,knowledge_sources, RAG loaders, embedding providers, vector stores,RagTool, andreset-memoriesbehavior. - Use llm-and-providers for
LLM, provider model strings, API keys, base URLs, Azure/OpenAI-compatible/Anthropic/Bedrock/Google/Snowflake settings, streaming, tool calls, response models, custom LLMs, and LiteLLM migration. - Use observability-and-hooks for tracing, telemetry, observability providers, event listeners, LLM/tool hooks, kickoff hooks, output logs, task outputs, and security fingerprints.
- Use files-and-multimodal for
crewai-files,input_files, file source resolution, provider file constraints, multimodal agents, upload cache, and file/document tool adjacency. - Use repo-development only for contributing to this CrewAI checkout: focused tests, docs versioning, Edge docs edits, frozen snapshot rules, workspace metadata, and safe native verification selection.
Common Decisions
- Prefer JSON-first projects from
crewai create crew <name>unless the user is maintaining an existing classic Python/YAML crew or needs decorator-heavy customization. - Treat
custom:<name>tools and{"python": "module.attribute"}JSONC references as trusted-code boundaries; inspect before running projects from untrusted sources. - Do not run LLM-backed crews, hosted deploy/login commands, MCP servers, network-backed tools, or credential-bound integrations just to inspect a project. Use the bundled diagnostic scripts first.
- Use provider-specific sub-skills for optional dependency and credential choices instead of installing all extras or assuming all integrations are available.
- For repo docs edits, modify only
docs/edge/<lang>/...; do not editdocs/v*/frozen snapshots or delete/renamedocs/images/assets outside release-freeze work.
Safe Diagnostics
python scripts/check_crewai_environment.py --jsonreports installed CrewAI package versions, top-level import results, and CLI command names.python sub-skills/core-runtime/scripts/validate_crew_definition.py --helpexplains static JSONC crew checks without importing project custom code.python sub-skills/flows-and-events/scripts/validate_flow_graph.py --helpexplains static flow graph checks without callingkickoff().python sub-skills/cli-and-projects/scripts/inspect_crewai_cli.py --helpexplains CLI/project layout checks that avoid running crews or deploy/login flows.python sub-skills/repo-development/scripts/select_native_tests.py --helpsuggests focused maintainer checks from changed paths without running them.
Refresh Triggers
Run refresh-repo-skill before relying on this skill if the current CrewAI checkout has a different commit than Repository provenance, package versions or CLI entry points changed, docs/edge/en/ workflows changed materially, or major source roots such as lib/crewai/src/crewai, lib/cli/src/crewai_cli, lib/crewai-tools/src/crewai_tools, or lib/crewai-files/src/crewai_files moved.
Signals
- GitHub stars
- 266
- Forks
- 21
- Last commit
- Sep 2026
ahel review
K6low
bundled executables the agent is told to runK1binfo
installs-packages (in sub-skills/tools-and-mcp/scripts/list_tool_exports.py)K1binfo
installs-packages (in references/installation-and-dependencies.md)
Automated review, not a security audit. Ruleset v1+k2.
Advanced
- Catalog kind
- skill
- Gateway key
crewai-vectorspacelab- Source
- github.com/vectorspacelab/arex-skill