grimdevminion

# Minion ## Routing Rule Prefer the current host agent's built-in subagent system first. ## Long-Running and Parallel Minions When this skill is explicitly triggered, ask the user whether they want: - the host's built-in subagent tool - a direct CLI Minion - an LM Studio local Minion - an Ollama local Minion - or the OpenRouter helper Use an external Minion route only when: - the user asks for a specific CLI/model/harness/agent - the current host has no useful built-in subagent tool - a model/provider comparison is the point of the task - we are trying to get multiple perspectives - we are doing code review and want other opinions - or the work needs an outside runner through OpenRouter - or the work benefits from a local/offline LM Studio or Ollama model Before launching any outside route, make the route explicit: - runner - model - prompt - optional: structured output - whether it may edit files or must stay read-only - whether this is a new conversation or a resumed one When running multiple minions in parallel, do not wait for every model before responding. As soon as one useful review/result comes back, report the current finding to the user and say which minions are still running. Fold slower minion results into a follow-up when they finish. Some high-reasoning models can run silently for a long time. Silence usually means it is still working. Do not interrupt or kill a minion just because it has not printed output yet. If you need to check, poll or check the c...