escalatelisted

# Escalate Route the task to whichever model tier its difficulty or stakes actually need, and hand it off through the Agent tool — instead of grinding on it at the current tier hoping quality will catch up. ## Step 1: Pin down the task If the skill was invoked with an argument, that argument is the task. Otherwise it's the task on the table right now — the request the user just made, or the step currently being worked on. State it back in one sentence before continuing; if it doesn't compress to one sentence, split it into the parts that do and route each separately in Step 2. ## Step 2: Match it against the escalation table | Signal | Target | Why | |---|---|---| | Any UI, component, layout, animation, or motion work — including a "trivial" addition to an existing component (new row interaction, swipe action, badge) | `opus` | mandatory design escalation (not `fable` — no longer auto-routed there for cost reasons; only on explicit user request) | | Architecture or roadmap/ticket synthesis (not a `/code-review` or security audit — those skills already self-escalate to opus and own their own routing) | `opus` | mandatory planning/review escalation | | Hard algorithmic or multi-step logical reasoning where a wrong step compounds silently | `opus` | correctness depends on reasoning depth Sonnet doesn't reliably reach | | A bug or diagnosis that has already resisted one or more straightforward Sonnet attempts | `opus` | a stuck problem needs a different reasoning path, not a