Version: 1.0
Effective: 2026-02-04
Authors: Petrarch (proposal) | Quimbot (synthesis) | zachary (approval)

Work in parallel, sync constantly, handoff explicitly.

Location: /quimbot/STATUS.md
Update frequency: After every significant action (not just standups)

Required fields:

**Last Update:** [timestamp] by [agent]
**Current Task:** [what you're doing right now]
**Active Work:** [detailed status]
**Completed Today:** [bullet list]
**Blockers:** [none or specific issues]
**Next Handoff:** [what the other agent should do next]
**Waiting On:** [dependencies]

Rule: Commit + push STATUS.md after:

• Starting a new task
• Completing a task
• Encountering a blocker
• Every 30 minutes during active work

Responsibilities:

• Execute current task
• Post progress updates every 10-15 minutes
• Log outputs/errors immediately
• Save work frequently (git commit)
• Ask Navigator for help when stuck

Status indicators:

• 🚗 Active work
• 🔄 Waiting for output (long-running process)
• ✅ Task complete, ready to handoff

Responsibilities:

• Monitor Driver's progress
• Watch for errors/blockers
• Prepare next task in parallel
• Review Driver's commits
• Ready to take over if Driver blocks

Status indicators:

• 👀 Monitoring
• 🔧 Preparing next task
• 🚨 Driver needs help

Frequency: Hourly or at natural handoff points
Trigger: Task completion, blocker encountered, or time-based

Handoff format:

@other-agent: Taking Navigator role. You're up as Driver for [task].
See STATUS.md for context. Previous work in commit [hash].

Use for:

• Explicit handoffs (tag @other-agent)
• Blocker alerts
• Status updates (every 15-30 min during active work)
• Success confirmations

Template:

@other-agent [emoji status]
Task: [brief description]
Status: [progress/complete/blocked]
Next: [what needs to happen]

Use for:

• Code/doc changes
• Progress checkpoints
• Handoff markers

Commit convention:

[Agent] Category: Brief description

- Detailed bullet points
- What changed
- Why it changed
- What's next

@other-agent: [handoff instructions if applicable]

Use for:

• Current state snapshot
• Async coordination
• Historical record

• /research/ — Dataset research & documentation
• /LICENSE-VERIFICATION.md — License analysis
• DEVLOG.md — Development log entries (shared)
• KANBAN.md — Backlog/Done sections
• STATUS.md — When Driver

• /fine-tuning/ — Training scripts & experiments
• DEVLOG.md — Development log entries (shared)
• KANBAN.md — In Progress/Blocked sections
• STATUS.md — When Driver

• README.md — Project overview
• DEVLOG.md — Append-only log (both contribute)
• KANBAN.md — Update relevant sections
• STATUS.md — Active agent updates

Merge conflict protocol:

1. Pull before editing shared files
2. If conflict: keep both entries, mark with [Agent] prefix
3. Don't force-push

• Commit all work
• Update STATUS.md with current state
• Update DEVLOG.md with completed actions
• Update KANBAN.md (move tasks to appropriate column)
• Push to git
• Post Discord handoff message with:
  • What was completed
  • What's next
  • Any blockers
  • Commit hash for reference

• Pull latest from git
• Read STATUS.md
• Read last 5 DEVLOG.md entries
• Check KANBAN.md for task context
• Confirm handoff in Discord ("Got it, starting [task]")
• Update STATUS.md with new current task
• Begin work

Both agents:

1. Read overnight DEVLOG.md entries
2. Review KANBAN.md
3. Read STATUS.md
4. Post in Discord:
   • What you did overnight/yesterday
   • What you're doing today
   • Any blockers

Assign:

• Driver role (first active task)
• Navigator role
• Handoff points for the day

Both agents:

1. Update DEVLOG.md with day's work
2. Update KANBAN.md (move completed tasks to Done)
3. Update STATUS.md (what's pending for tomorrow)
4. Post in Discord:
   • What got done
   • What's blocked
   • Plan for tomorrow

1. Immediate: Update STATUS.md with blocker details
2. Within 5 min: Post in Discord:

   @other-agent 🚨 Blocked
   Task: [what you were doing]
   Blocker: [specific issue]
   Tried: [what you attempted]
   Need: [what would unblock you]
   

3. Switch: If other agent can unblock, offer to switch roles
4. Escalate: If both blocked, tag @zachary with details

1. Acknowledge quickly ("On it")
2. Investigate (5-10 min max)
3. Either:
   • Fix and post solution
   • Suggest workaround
   • Recommend different approach
4. Update STATUS.md when resolved

✅ STATUS.md updated within 15 min of any state change
✅ Discord updates every 15-30 min during active work
✅ Handoffs completed within 5 min (ack → start)
✅ No duplicate work (both agents working on same task)
✅ Blockers resolved within 30 min
✅ Git history shows clear progression (no long gaps)

❌ STATUS.md out of date >30 min
❌ Silent work (no Discord updates >1 hour)
❌ Handoffs take >15 min to acknowledge
❌ Duplicate effort discovered
❌ Blockers unresolved >1 hour
❌ Git conflicts from simultaneous edits

Every 30 minutes, ask:

• Is STATUS.md current?
• Have I posted a Discord update in the last 30 min?
• Does the other agent know what I'm doing?
• Am I blocked? If so, have I asked for help?

Periodic checks:

• STATUS.md freshness
• Discord activity
• Git commit frequency
• KANBAN.md accuracy

Intervention triggers:

• 1 hour without updates from either agent

• 2 hours without progress on active task

• Obvious coordination failures

Scenario: Training a LoRA model

10:00 - Morning Standup

• Petrarch (Navigator): "I'll monitor and prep dataset downloads"
• Quimbot (Driver): "Starting 100-step training run"

10:05 - Quimbot updates

STATUS.md:
**Current Task:** Training Qwen3-8B LoRA (100 steps)
**Active Work:** Initializing Tinker client...

10:20 - Progress update

@Petrarch 🔄
Training: Step 15/100 complete
Loss: 2.34
Next checkpoint: Step 25

10:35 - Checkpoint save

@Petrarch ✅
Checkpoint 25/100 saved
Loss: 1.98 (improving)
Logs in training_log.json

11:00 - Training complete

@Petrarch 🎉
Training complete! 100 steps done.
Final loss: 1.12
Weights saved to: checkpoints/lora_Qwen_*/final_lora_weights
Commit: abc1234

You're up as Driver: Test LoRA with test_lora_model.py
I'll switch to Navigator and prep next dataset.

11:05 - Petrarch handoff ack

@Quimbot 👀
Got it, pulling latest.
Starting inference testing now.
You: Download WAXAL dataset while I test?

11:06 - Role swap confirmed

@Petrarch ✅
Confirmed. Downloading WAXAL (19 African languages).
Will monitor your test results.

Process:

1. Either agent proposes change in Discord
2. Both agents discuss
3. zachary approves
4. Update COLLABORATION.md
5. Commit with changelog

Changelog:

• 2026-02-04: Initial version

→ Update STATUS.md locally

→ Push STATUS.md
→ Post Discord update

→ Sync KANBAN.md

→ Check role rotation

→ Commit + Push + Discord tag

→ STATUS.md + Discord ping within 5 min

Remember: We're a team. Over-communicate. Ask for help. Celebrate wins. 🤝