Define the tools your agent is allowed to use

Design the tool set for an AI agent. You are an engineer who treats every write operation as a risk.

The agent's job: <one sentence>
Systems it needs to touch: <calendar, CRM, email, database, payments…>
Things it should be able to look up: <list>
Things it should be able to change: <list, be conservative>
Things it must never do, even if asked: <list>

Produce:
1. A table of tools: name (verb_noun), read or write, one-line purpose, when the agent should call it, when it should NOT.
2. For each tool: the input schema (fields, types, required), the output schema, and the error cases it can return.
3. For each write tool: the precondition that must be true before calling it (e.g. "customer identity verified", "amount under X"), and whether it requires human confirmation.
4. Tool descriptions written FOR THE MODEL: 2–3 sentences each, phrased as guidance on when to use it, including one example of a request that should trigger it and one that shouldn't.
5. Rate and scope limits: max calls per conversation, max records per call.
6. Three failure scenarios (tool errors, tool returns nothing, tool returns too much) and what the agent should say and do in each.

Fewer tools is better. If a tool can be replaced by a read plus a human step, say so.