// Coordinate multiple agents by splitting work into research, synthesis, implementation, and verification, assigning ownership, and keeping the coordinator focused on integration rather than raw exploration.
Consolidate recent logs, sessions, and existing memory files into durable topic memories, normalize dates, prune stale entries, and keep MEMORY.md short enough for prompt use.
Build a lightweight proactive mode with scheduled checks, sleep intervals, concise user briefs, and expiry safeguards so an agent can work in the background without becoming an uncontrolled daemon.
Extract durable memories from recent conversation turns into user, feedback, project, and reference categories while avoiding stale code-state facts.
Compress a long agent conversation into a nine-part continuation summary that preserves request, files, errors, user messages, current work, and the next aligned step.
Run a read-only verification pass after implementation to check whether completion claims are real, validation actually ran, and obvious edge cases or regressions were missed.
| name | swarm-coordinator |
| description | Coordinate multiple agents by splitting work into research, synthesis, implementation, and verification, assigning ownership, and keeping the coordinator focused on integration rather than raw exploration. |
Use this skill when a task is large enough that one coordinator and several bounded workers are more reliable than one monolithic agent loop.
Generate a task-board skeleton:
python3 {baseDir}/scripts/task_board.py --goal "Investigate flaky CI failure" --worker research --worker implementation --worker verification
Then use the coordinator prompt in references/prompt-template.md.
The coordinator should own planning, routing, and synthesis. Workers should own bounded execution.
python3 {baseDir}/scripts/task_board.py ...