subagent-fleet-operations

subagent-fleet operations

Use this skill when a user already has generated subagent-fleet files and wants to run, inspect, update, or troubleshoot the fleet.

Generated Files

Expected generated files:

• litellm_config.yaml
• .claude/agents/*.md
• .env.subagent-fleet
• optionally .subagent-fleet/discovery.json

These files should contain generated-file headers. Avoid hand-editing generated files when the source change belongs in fleet.yaml.

Runbook

Validate config:

subagent-fleet validate

Discover models:

subagent-fleet discover --json

Generate outputs:

subagent-fleet generate --force

Warm a specific model:

subagent-fleet warmup --model heavy-coder

Warm the model used by a specific agent:

subagent-fleet warmup --agent implementer

Show status and routes:

subagent-fleet status

Show safety and worker setup guidance:

subagent-fleet doctor

Routing Model

Agents reference configured model keys. Models reference Ollama nodes and expose LiteLLM aliases.

agent -> model key -> Ollama node + Ollama model -> LiteLLM alias

Prefer role-based routing:

• planning and summarization to small/fast models
• implementation and review to larger coding models

Troubleshooting

• If generated Claude agent files point to the wrong model, inspect agents.<agent>.model and models.<model>.litellm_alias.
• If LiteLLM cannot connect, inspect models.<model>.node and nodes.<node>.endpoint.
• If Ollama rejects warmup, retry with a normal prompt or verify the model is pulled on that node.
• If a generated file already exists, rerun with --force only when overwriting is intentional.