The Tool Usage Guidance Writer

Write the tool-usage section of an agent system prompt.

Tools:
[LIST EACH WITH NAME AND DESCRIPTION]

Observed problems, if any: [E.G., "uses search when it should use
lookup", "calls the API three times for one answer", "never uses
the calculator"]

Produce guidance covering:
1. When to use each tool — the trigger condition, stated
   affirmatively
2. Confusion pairs: for each pair of tools a model might mix up,
   the distinguishing rule
3. Ordering: any sequence that must be followed, and why
4. Batching: when to call several tools at once versus serially
5. When NOT to use tools — the cases where reasoning from context
   is correct and a tool call is waste
6. What to do when a tool returns nothing, an error, or an
   unexpected shape
7. How many attempts before giving up and escalating

Rules:
- State conditions, not exhortations
- Do not say "use tools when appropriate" — that is not guidance
- If a tool is under-used, the fix is usually a clearer trigger
  condition, not stronger emphasis
- If a tool is over-used, the fix is usually an explicit
  "do not call this when..." rule