Design: Multi-agent coordination layer (Mode 3 DAG-as-Protocol)

[AVIDS2]

Context

Architecture discussion from v0.2 planning session.

Problem

When 3+ agents collaborate on a shared .vek repo, who coordinates merges, how are task dependencies expressed, and how do you get a global view?

Proposed Approach: DAG-as-Protocol (Mode 3)

Use special tool name conventions to express agent coordination within the existing DAG no new mechanisms needed.

Convention

• __delegate__ assign task to another agent
• __report__ return results from delegated task
• __sync__ checkpoint across agents

Possible new APIs

• vek.watch(branch, callback) subscribe to new nodes on a branch
• Per-branch log already supported via vek.log()

Why not now

Need real user feedback first. Ship demo, get users, see how they actually use vek in multi-agent setups, then design based on evidence.

Alternatives considered

1. Coordinator Pattern central agent merges all branches
2. Event Bus agents subscribe to each other's branches
3. DAG-as-Protocol convention-based, no new mechanism (preferred)

Status

Parked. Revisit after first user feedback from HN launch.