Skip to Content
DocsConceptsAI Agents

Overview

AIDEN uses Claude AI agents to implement stories autonomously. Each agent is a spawned Claude CLI process with full tool access — it can read and write files, run bash commands, and interact with your codebase.

Agent Modes

Planning Mode

When a story is in the Plan column, the AI operates in text-only mode (--print flag). It can discuss, suggest, and refine specs but cannot modify files.

Agent Mode

When a story moves to In Progress, the AI switches to full agent mode with tool access:

  • Read/write files in the project
  • Execute bash commands
  • Run tests
  • Make Git commits

Parallel Execution

AIDEN supports running multiple agents simultaneously, each working on a different story in its own Git worktree:

Project/
├── main branch (your code)
├── .worktrees/
│   ├── feature/s-1-auth/        ← Agent 1 working here
│   ├── feature/s-2-dashboard/   ← Agent 2 working here
│   └── feature/s-3-api/         ← Agent 3 working here

Conflict Prevention

To prevent agents from stepping on each other:

  • File Locking — Agents acquire locks on files they’re editing. Other agents wait or work on different files.
  • Agent Broadcasts — Agents can communicate status updates to coordinate work.
  • API Contracts — Agents publish API contracts so others know what interfaces to code against.

The Orchestrator

The Orchestrator is a project-wide AI chat that can:

  • Spawn and manage sub-agents
  • Update story statuses across the board
  • Coordinate work between multiple agents
  • Provide project-level context and decisions

Custom Agents

Create named AI personas (e.g., @max, @alex) with:

  • Dedicated workspaces
  • Custom context files and instructions
  • Specialized roles (frontend, backend, testing, etc.)
SpecsKanban Board