Research Team

2. Initialize memory files

For each coauthor, create their memory file if it doesn't exist:

# {Role} Memory
**Last updated:** YYYY-MM-DD round 000 (initialized)

## Lessons
(none yet)

## Running Context
- Awaiting first assignment from PI

3. Initialize team log

# Research Team Log
**Project:** [from CLAUDE.md]
**Initialized:** YYYY-MM-DD

## Rounds
(none yet — PI coordinator will begin Round 001)

4. Launch PI Coordinator

Spawn the pi-coordinator agent (see .claude/agents/pi-coordinator.md) with:

• Path to the approved research plan
• Current pipeline stage from PIPELINE_STATE.md
• Instruction to begin Round 001

Usage

/research-team                            # Initialize and start round 1
/research-team resume                     # Resume from last completed round
/research-team round N                    # Start specific round N
/research-team status                     # Show current team status (reads state.json)
/research-team --max-rounds 5             # Limit to 5 rounds (default: 10)
/research-team --max-budget-usd 8.00      # Cap total agent spend

Budget and Round Limits

The PI coordinator tracks cumulative cost and round count in quality_reports/team/state.json. When either limit is reached, the PI saves state and stops gracefully. Default max rounds: 10. Default budget: unlimited.

Resume Mode

When resuming:

1. Read quality_reports/team/team-log.md for last completed round
2. Read the last sync-meeting.md for carried-over items
3. Launch PI coordinator to continue from the next round

Status Mode

When checking status, read quality_reports/team/state.json and display:

1. Current round number and pipeline stage
2. Each agent's last score and status
3. Open REVISE/REJECT items (convergence status)
4. Budget spent vs. limit (if set)