5 streamlined commands for multi-agent coordination located in ~/code/jat/commands/jat/
One agent = one session = one task. Each Claude session handles exactly one task from start to completion.
Core Workflow:
/jat:start [agent-name | task-id]- Main command: handles registration, task selection, conflict detection, and work start/jat:complete [task-id]- Finish work, verify, commit, close task, end session/jat:commit- Create well-organized commits with automatic documentation updates
Escalation & Planning:
/jat:verify [url]- Escalatory: browser verification when user wants deeper testing/jat:tasktree [prd-path]- Convert PRD/spec into structured tasks with dependencies
Maintenance:
jat-doctor- Bash script to diagnose installation issues (run anytime)
Key behaviors:
- All commands register your agent identity in the Agent Registry
- File reservations prevent conflicts between agents
- Memory system provides cross-session context
Quick Start:
# Simple workflow
/jat:start # Create agent, show available tasks
/jat:start task-abc # Create agent, start specific task
/jat:complete # Complete task, end session
# With specific agent (IDE spawn)
/jat:start MyAgent task-abc # Use MyAgent, start taskEscalatory Verification:
# Agent says "READY FOR REVIEW"
# User wants deeper testing...
/jat:verify # Open browser, test the feature, check console
/jat:verify /tasks # Verify specific pageSession Lifecycle:
spawn agent → work on task → review → /jat:complete → session ends
│
(optional escalation)
│
/jat:verify → browser test → back to review
Session-Aware:
Each command automatically updates .claude/sessions/agent-{session_id}.txt for statusline display. Supports multiple concurrent agents in different terminals.
See project CLAUDE.md for detailed documentation.