Three surfaces. One callback contract.

Same tool, ≥3 cosine-similar calls in 60s

Agent stuck on a tool it's not getting unstuck from

Prompt-tokens-per-turn slope > 1500/turn

ReAct context window growing without bound

Exact-hash repeat embedding within session

Same vector look-up paid for repeatedly

No user-facing output for 5 min, calls firing

Background agent that nobody is reading

Classification prompt sent to a frontier model

Opus answering a yes/no — Haiku would suffice

≥5 identical calls in 30s

Provider 529 + no exponential backoff

≥30 tool defs OR ≥30KB of tool JSON

The MCP problem — 55K tokens before user types

≥3 retrieval calls with overlapping queries in 120s

RAG pipeline that keeps re-fetching

Same image SHA-256 (or phash distance ≤6) across calls

Same screenshot re-uploaded turn after turn

High-detail flag on low-detail-suitable image

Paying 4× for detail you don't need

Vision dollars concentrated in <5% of sessions

One agent eating the vision budget

Deepgram channels≥2 with multichannel=true

Per-channel billing doubled without reason

Same text_hash across ≥3 ElevenLabs voice_ids in 10s

A/B test scaffold leaked into production

Same Cohere rerank request_hash across ≥2 calls in 30s

Reranks fired without a cache layer

≥2 correction-shaped user turns + similar regenerations within 10 turns

Cost-aligned proxy for hallucination/grounding failures