B
Offline
Step 1 of 10

Teardown and Review

10of 10~5 min
  • Proper session cleanup
  • How to review agent performance
  • Best practices for future sessions
What you'll do
  • Stop all agents
  • Archive the session
  • Review what worked

Congratulations on completing your first multi-agent research session! Before moving on, let's properly clean up and extract lessons for future sessions.

1Stop All Agents

Gracefully stop each agent in their ntm panes:

Terminal
# In each agent pane, press Ctrl+C or type:
exit
# Or use ntm to kill panes
ntm kill codex
ntm kill claude
ntm kill gemini
# Kill the entire session (optional)
ntm kill $SESSION_ID
# Verify no agents are running
ntm list

2Archive the Session

Package all session files for future reference:

Terminal
# Create archive
brenner session archive \
  --thread-id "$SESSION_ID" \
  --output archives/$SESSION_ID.tar.gz
# Contents of archive:
# - kickoff.md
# - all agent messages (DELTA, REVISION, etc.)
# - compiled artifact.json
# - score.json
# - session metadata
# View archive contents
tar -tzf archives/$SESSION_ID.tar.gz

3Review the Session

Take a few minutes to reflect on what worked and what didn't:

Which agent performed best?

Note their model/config for future sessions with similar research questions.

Where did you need to intervene?

Frequent interventions might indicate unclear kickoff prompts or role definitions.

What hypotheses surprised you?

The best sessions produce ideas you wouldn't have generated alone.

What would you change next time?

Capture specific improvements while the session is fresh in your mind.

Export Session Log

Create a human-readable log for your research notes:

Terminal
# Export as markdown
brenner session export \
  --thread-id "$SESSION_ID" \
  --format markdown \
  --output sessions/$SESSION_ID/log.md
# View the log
cat sessions/$SESSION_ID/log.md

What's Next?

Run More Sessions

Practice makes perfect. Run sessions on different research questions to develop your orchestration intuition.

Customize Roles

Create custom role definitions beyond the standard three. Experiment with specialist roles for your domain.

Build Your Corpus

Add domain-specific excerpts to your corpus. Better context leads to better hypotheses.

Automate Workflows

Use the brenner CLI in scripts to automate common patterns. Create templates for different research types.

Pro Tip
Start a "session notes" file. After each multi-agent session, jot down one thing that worked well and one thing to improve. This compounds over time.
🎉

Tutorial Complete!

You've learned the full multi-agent cockpit workflow.

You now know how to:

  • • Set up the multi-agent infrastructure
  • • Configure and assign agent roles
  • • Write effective kickoff prompts
  • • Launch and monitor sessions
  • • Compile and score research artifacts
  • • Iterate toward high-quality outputs