2026-05-19 Agent Swarm With Hard Scope

An agent swarm is not a group of AI tools editing the same main branch at the same time.

For Prox OS, a useful swarm means:

• One task graph.
• One branch or worktree per worker.
• One clear ownership area per worker.
• Allowed paths and forbidden paths.
• Run logs for every meaningful execution.
• Integrator review before merge.
• Human final decision.

Why This Matters

AI can make parallel work tempting. But without scope, parallel agents create the fastest possible path to codebase cognitive debt: design intent gets lost, hidden coupling grows, and later agents cannot tell which behavior is deliberate.

Hard scope turns AI parallelism into a system:

• Docs Agent can update docs.
• UI Agent can work in os-ui or UI Workshop.
• API Agent can work on API contracts.
• Shell Agent can work on shell runtime.
• Integrator Agent reads all diffs before merge.

The swarm is useful only when coordination is stronger than generation speed.