| name | ruflo-orchestration |
| description | Use when the user explicitly asks for swarm, multi-agent, orchestration, parallel agents, federated agents, autopilot builds, or Ruflo/Claude Flow style coordination. Codex-native distillation of ruvnet/ruflo. |
Ruflo-Style Orchestration
Codex may use subagents only when the user explicitly requests delegation, subagents, swarm work, or parallel agent work.
When To Orchestrate
Good fits:
- Independent research questions.
- Disjoint implementation slices with non-overlapping files.
- Parallel verification while local implementation continues.
- Large reviews where security, architecture, and tests can be assessed separately.
Poor fits:
- The immediate next step depends on the delegated result.
- The task is tightly coupled across the same files.
- The subtask is vague or requires constant judgment.
Swarm Plan
Before spawning agents:
- Define the shared goal and done criteria.
- Split work by ownership boundary.
- Assign each agent a concrete output and file scope.
- Tell workers they are not alone in the codebase and must not revert others' changes.
- Continue local non-overlapping work while agents run.
Integration
- Review returned changes before trusting them.
- Merge concepts, not blindly.
- Run unified verification after integration.
- Close unused agents once their result is no longer needed.
Source Notes
Distilled from github.com/ruvnet/ruflo swarm and coordination skills, adapted to Codex’s explicit subagent rules.
