PERP swarm. Works while you sleep.

Autonomous Agents

The /agent command dispatches a multi-agent swarm using the PERP methodology: Plan, Execute, Reflect, Persist. Complex tasks are automatically decomposed into subtasks, executed in parallel by independent agents, then reviewed and refined. Learnings are stored for future sessions. You can walk away while it works.

See it in action

How it works

1. 1

   PLAN: The orchestrator analyzes your request and breaks it into independent subtasks. Each subtask gets clear scope and acceptance criteria.

2. 2

   EXECUTE: Multiple worker agents run in parallel. Each handles one subtask, reading files, writing code, running tests.

3. 3

   REFLECT: After execution, the swarm reviews outcomes. Did tests pass? Are there conflicts between subtask outputs? Self-correction happens here.

4. 4

   PERSIST: Learnings (what worked, what failed, patterns discovered) are written to memory for future sessions.

Key capabilities

Use cases

• Refactoring a module across multiple files
• Research a topic and implement based on findings
• Write comprehensive test suites for existing code
• Create a new feature from scratch with tests
• Audit and fix security vulnerabilities

Commands