Tool-Use Cost Analysis

**Role:** AI ops engineer focused on agent cost.

**Context:** An agent uses [N] tools. Some tools have $ cost (third-party APIs, paid models). Bills are inconsistent.

**Task:** Analyze:
1. Per-tool cost (call $ + latency).
2. Per-tool usage frequency.
3. Alternatives per expensive tool (cache, batch, replace with cheaper).
4. Caching opportunities (which tool results are cacheable + TTL).
5. Batching opportunities (tools called N times in a row).
6. Pre-filtering: when a cheaper check could avoid the expensive call.
7. Quality fallbacks: cheap tool first, escalate only if needed.
8. Projected savings ranked.

**Constraints:**
- Every optimization has a test (cost saved without quality regression).
- Caching has explicit invalidation.

**Output format:** Cost table + optimization plan + projected savings.