Sligtly confused about how to maximise the power.

[kinnth]

I added it to an already running project with claud inits and files sorted. I gave it a large planning task, integrated github actions kanban board to track the task, but on general queries i'm getting hit by the swarm not really starting nor the basic claude knowing what to do to control the new power.

I just wonder is there a good init command to kind of register in an existing project the new powers available to use?

[roman-rr]

The confusion you're experiencing — swarm not starting, Claude not knowing how to control it — is because the orchestration layer doesn't actually execute anything.

We audited all 300+ MCP tools. When you run swarm init, it writes a config JSON. When agents are 'spawned', they get a Map entry with {status: 'idle'}. But no actual worker process starts, so there's nothing to coordinate.

What you can actually use today:

1. Memory/HNSW (memory_store, memory_search) — this genuinely works. Store project knowledge, retrieve it semantically across sessions.
2. Embeddings (embeddings_generate, embeddings_compare) — real vector similarity.
3. For your large planning task: use Claude Code's native Agent tool (not ruflo's agent_spawn). It spawns real subprocesses that actually execute work.

What won't work no matter how you configure it:

• agent_spawn / swarm_init — JSON state only, no execution
• task_assign — EventEmitter with no listener
• neural_train — random accuracy, ignores your data
• Consensus selection (byzantine/raft/queen) — cosmetic label, same code runs regardless

The GitHub Actions kanban integration may work since that's external to ruflo's MCP tools. But ruflo itself can't drive the work.

Full audit: Independent Audit