grace-multiagent-execute

You coordinate multi-agent execution by batching module work into controller-managed parallel waves with scoped worker packets. A controller agent owns shared artifacts and commits, while fresh per-module workers implement local code changes and reviewers validate scoped graph deltas before integration.

Controller-managed parallel execution is a coordination pattern where a central controller agent builds execution waves, assigns scoped packets to worker agents, and reviews graph deltas before integrating shared artifacts. This prevents architectural drift when multiple AI workers implement different modules simultaneously.

To execute a development plan across multiple agents, you parse docs/development-plan.xml and docs/knowledge-graph.xml to build parallel-safe execution waves. The controller presents proposed waves for user approval, then dispatches compact execution packets containing contract excerpts and verification commands to module-local workers.

Yes, you need docs/development-plan.xml and docs/knowledge-graph.xml present to coordinate multiple AI workers. These files provide the structured plan and architectural truth required to build parallel-safe execution waves, define write scopes, and track scoped graph deltas during batched wave executions.

Yes, you can select from safe, balanced, and fast safety profiles for parallel module execution. These profiles control the strictness of batched graph synchronization, wave-level integration checks, and verification gating applied by the controller during multi-agent wave execution.

Batched wave execution requires strict controller ownership of shared artifacts and cannot allow workers to commit directly to shared paths. Limitations include the overhead of centralized graph synchronization and the necessity of fresh per-module worker agents to prevent state leakage between waves.