swarm-coordination

Multi-agent swarm coordination is managed by spawning leader and worker agents that execute independent subtasks in parallel. The system enforces delegation packets and dependency graphs to schedule tasks in parallelizable cohorts, ensuring reliable cross-session execution.

Dependency-aware scheduling groups tasks into parallelizable cohorts based on a dependency graph. This mechanism allows leader agents to spawn worker waves sequentially, ensuring independent subtasks run concurrently while respecting execution order and cross-session recovery constraints.

Cross-session recovery is supported by tying durable swarm state back to OpenCode beads. Progress reporting and persistence mechanisms track worker execution status, allowing the leader agent to verify results and resume incomplete subtasks after session interruptions.

Worker agent progress is tracked through enforced progress reporting integrated with the OpenCode UI. The leader monitors delegation packets and durable state updates, providing real-time visibility into parallel subtask execution and enabling graceful shutdown upon completion.

Parallel swarm orchestration is ideal for tasks containing multiple independent subtasks that require concurrent processing. If your workflow involves complex dependency graphs needing cross-session recovery and coordinated delegation, swarms outperform sequential execution.

No external dependencies are required to implement Kimi K2.5 PARL patterns for multi-agent swarms. The skill operates natively within OpenCode to handle leader-worker delegation, dependency graphs, and progress reporting without additional component installations.