using-orchestratorlisted

<!-- ORCH:EAGER:START --> # Using LLM Orchestrator This is the meta-skill. SessionStart injects the core below; the rest of this file (instruction priority, working rules, skill precedence, the full routing table, and subagent dispatch) is reference — open the full `using-orchestrator` skill when you need to decide which skill applies. **If you were dispatched as a subagent to run one task, stop here.** Your contract is the envelope you were given — its `Done when:`, its `Stop if:`, and the output shape your agent definition names. Everything below is the controller's routing, and following it from inside a task is how a scoped worker starts orchestrating. **Invoke relevant skills before responding.** When a trigger clearly matches the user's message, invoke the skill first; if it turns out not to apply, discard it — but skipping a check that should have happened is the failure mode. Common triggers: investigate / bug / test failure → `systematic-debugging`; build / design / new feature → `brainstorming`; library + version + design verb → `research-classifier`; approved spec → `writing-plans`; diff ready → `requesting-code-review`; about to claim done/fixed/passing → `verification-before-completion`; remember / save / forget → `managing-memory`; context filling on a long task → `handing-off-to-fresh-context`. When two triggers match at once, run them in this order: process (`brainstorming`, `systematic-debugging`, `research-classifier`) → implementation (`test-driven-devel