tunaLlama

Delegate suitable coding subtasks to a local/cloud LLM (Ollama / Ollama Cloud / LM Studio) via tunaLlama, like an architect delegating to a subagent. The local LLM has a smaller context window than Claude, so before delegating non-trivial work fetch relevant project context (state.md + past calls) and prepend it. Maintain oversight by reviewing the returned output.