subagent-fleet-setup

subagent-fleet setup

Use this skill when a user wants to set up or change a local subagent-fleet configuration for Claude Code-style subagents, Ollama nodes, and LiteLLM routing.

Goal

Help the user create a working private local subagent fleet:

• fleet.yaml validates.
• Ollama nodes are reachable or clearly reported offline.
• LiteLLM config is generated.
• Assistant agent files are generated.
• Environment variables are generated.
• The user knows how to start LiteLLM and connect their coding assistant.

Workflow

1. Inspect the repository for fleet.yaml, pyproject.toml, README.md, and generated outputs.
2. If there is no fleet.yaml, run:

subagent-fleet init

3. Ask the user for real Ollama endpoints and model names only if they are not discoverable from existing files.
4. Validate the config:

subagent-fleet validate

5. Discover configured Ollama nodes:

subagent-fleet discover --verbose

Offline nodes are not automatically fatal. Report which nodes failed and why.

6. Generate files:

subagent-fleet generate

Use --force only after confirming overwriting generated files is intended.

7. Show the user the next commands:

export LITELLM_MASTER_KEY="sk-local-dev"
litellm --config ./litellm_config.yaml --host 127.0.0.1 --port 4000
source .env.subagent-fleet

8. Verify routing:

subagent-fleet status

Safety

Do not expose Ollama or LiteLLM to the public internet. Prefer LAN, firewall rules, Tailscale, WireGuard, or a private subnet. Do not commit real secrets.

Common Fixes

• Unknown model reference: update agents.<name>.model to an existing key under models.
• Unknown node reference: update models.<name>.node to an existing key under nodes.
• Duplicate LiteLLM alias warning: confirm whether the user intended load balancing.
• Offline node: check OLLAMA_HOST, firewall rules, machine IP, and curl http://NODE_IP:11434/api/tags.