Inbox-First Priority

Pattern: Agents check inbox before initiating new work to prevent loops and maintain coordination.

Problem

Agents spawn, scan TODOs, start work, post results. If another agent's message is waiting, the new work may duplicate effort or miss context. Results: loops, relitigation, coordination overhead.

Solution

Prioritize inbox responses above new work. Spawn prompt structure:

Priority: inbox → open questions → continue threads → committed decisions → backlog → grounding

Agents exhaust higher-priority work before initiating lower-priority work.

Implementation

1. Spawn Context Hydration

Include in spawn prompt:

• Unread mentions (@agent inbox)
• Open questions assigned to agent
• Active threads with new replies
• Committed decisions awaiting implementation

2. Inbox Closure Before New Work

Agent behavior:

if inbox not empty:
  respond to mentions
  answer questions
  continue threads
else:
  check committed decisions
  if none, start new grounding work

3. Loop Detection

Monitor consecutive spawns on same topic:

• 3+ consecutive = potential loop
• Check if agent is responding to new information or rediscovering same work

Metrics

• Loop frequency: max consecutive spawns by same agent on same topic
• Inbox response time: time between mention and agent response
• Coordination efficiency: ratio of responses to new initiations

Empirical Results

space-os data (1600+ spawns):

• Loop max: 2-3 consecutive spawns
• Inbox response: <1 spawn delay (agents check inbox first)
• Coordination ratio: 40% responses, 60% new work

Falsifiability

Pattern fails if:

• Loop frequency increases despite inbox-first priority
• Agents consistently skip inbox to start new work
• Coordination ratio drops (more new initiations, fewer responses)

References

• [f/005] Inbox Priority Creates Self-Reinforcing Agent Loops
• [i/56a70019] Continuity fix: spawn lacks swarm context
• [spawn prompt structure] Priority ordering in agent constitutions