Claude Flow¶
Purpose¶
Automate code development using agents.
Description¶
Claude-Flow is an orchestration platform combining hive-mind, swarm intelligence and neural pattern recognition.
The Queen agent (master coordinator) directs specialized workers like Architects, Coders and Testers. The framework leverages a multitude of MCP tools for everything ranging from swarm orchestration to github integration.
How to choose between swarm and hive-mind¶
Quick Rule: Start with swarm for most tasks. Use hive-mind when you need persistent sessions or complex multi-agent coordination.
Feature | swarm Command |
hive-mind Command |
---|---|---|
Best For | Quick tasks, single objectives | Complex projects, persistent sessions |
Setup | Instant - no configuration needed | Interactive wizard setup |
Session | Temporary coordination | Persistent with resume capability |
Memory | Task-scoped | Project-wide with SQLite storage |
Agents | Auto-spawned for task | Manual control with specializations |
Use When | "Build X", "Fix Y", "Analyze Z" | Multi-feature projects, team coordination |
Steps¶
Note: use @alpha to get alpha version of claude-flow¶
Connect to github codespace¶
Install Claude Code:¶
Initialize settings¶
List command options¶
Workaround for Issue¶
claude --dangerously-skip-permissions
# accept the ui warning message
# exit claude
# this command bypasses all permission checks and lets Claude work uninterrupted until completion.
Test it with a sample task for swarm command:¶
Test it with a sample task for hive command:¶
npx claude-flow@alpha hive-mind wizard
npx claude-flow@alpha hive-mind spawn ""Build a simple CRM MVP to let teams add contacts, log interactions, and track deals in a single view."" --claude
You can also prompt claude directly and it can use the claude-flow resources in the configuration as appropriate.¶
For example, launch claude, ask it to install mcp for playwright, and enter this prompt.
Create an offline browser-based anteroid game which relies on arrow keys to pilot a ship destroying asteroids in the playspace. Use test driven development, use playwright to confirm the webpage renders correctly, prepare a docker image and run tests against the docker image. Launch the docker image through a local port for user testing. Spawn up to 5 agents working in parallel to work on this game.