quota-exhaustion-failoverlisted

# Skill: Quota / token exhaustion failover (HARD GATE) Use when any coding agent or surface returns a quota, rate-limit, weekly/monthly cap, spend-limit, or "tokens exhausted" error. **Rule:** never silent-fail and never burn the same exhausted surface with blind retries. Try **param/effort/scope levers** before vendor failover. Then traverse the vendor-neutral tier tree + closed-world lineup before naming a substitute SKU. - **Hard gate tree:** [`../../knowledge/ai-coding-quota-exhaustion-decision-tree.md`](../../knowledge/ai-coding-quota-exhaustion-decision-tree.md) - **Deep layer (playbook):** [`../../knowledge/coding-agent-levers-playbook.md`](../../knowledge/coding-agent-levers-playbook.md) — lever encyclopedia, interaction matrix, missing-lever failover, worked examples. Tree remains the hard gate; do not duplicate the encyclopedia here. ## Ladder 1. **Classify** — context overflow | rate limit | soft quota | hard weekly/monthly/spend 2. **Record** — surface, model, error, reset time, task state, remaining headroom 3. **Levers on usable surface (before vendor hop)** 1. Lower effort / thinking / fast flags to stretch quota if quality OK 2. Raise reasoning dial on same model before bigger SKU 3. Tighten prompt/scope; decompose into smaller runs 4. Shrink context; avoid max window params unless required 4. **Failover order (same tier first)** 1. Same vendor, different surface (CLI ↔ cloud agent) 2. Same tier, different vendor (closed-world lineup)