---

name: cli-and-projects description: "Routes CrewAI CLI, project scaffolding, JSONC projects, run/train/test/replay/chat/checkpoint/deploy commands, and project template troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: MIT

CrewAI CLI and Projects

Use this sub-skill when a task involves the crewai command line, scaffolding a CrewAI crew or flow project, understanding JSON-first versus classic project layouts, validating deployment readiness, or diagnosing project-root command failures.

Read First

• CLI reference for command groups, safe/default behavior, and command options.
• Project templates for JSON-first crew, classic crew, and flow layouts plus pyproject.toml expectations.
• Deployment for crewai deploy routing, validation checks, lockfiles, and credential boundaries.
• Troubleshooting for missing project roots, placeholder inputs, JSONC/classic confusion, unsafe custom references, chat_llm, deploy credentials, and uv wrapper issues.
• CLI inspector script for safe local introspection of the installed crewai_cli command tree and lightweight project-layout checks.

Route Across CrewAI Skills

• Use ../../SKILL.md first when the user needs overall CrewAI package routing or is not clearly asking about CLI/project files.
• Use ../core-runtime/SKILL.md for Agent, Task, Crew, guardrails, callbacks, hierarchical process semantics, and JSONC object field meaning beyond project layout.
• Use ../flows-and-events/SKILL.md for Flow, @start, @listen, routers, plotting semantics, persistence, and event graph design beyond CLI invocation.
• Use ../tools-and-mcp/SKILL.md for built-in tool classes, BaseTool, MCP adapters, and custom tool implementation details referenced by project templates.
• Use ../memory-knowledge-and-rag/SKILL.md for memory stores, knowledge sources, RAG loaders, and reset-memory data behavior beyond CLI flags.
• Use ../observability-and-hooks/SKILL.md for tracing, hooks, log-tasks-outputs interpretation, and telemetry configuration beyond command syntax.

Safe Operating Defaults

• Prefer read-only checks first: crewai --help, crewai create --help, crewai version, crewai deploy validate, and this sub-skill's inspector script.
• Do not run crewai run, crewai train, crewai test, crewai replay, crewai chat, crewai checkpoint resume, crewai deploy create, or crewai deploy push on an untrusted project without user approval; they may execute local Python, custom tools, callbacks, LLM calls, subprocesses, deployment actions, or destructive state changes.
• Treat custom:<name> tool references and JSONC {"python": "module.attribute"} references as executable local code; inspect before running.
• Keep exact installed-version assumptions fresh with crewai version or the bundled inspector when command signatures matter.