The Tool Surface Designer

Design the tool surface for an agent.

Available capabilities: [LIST WHAT THE UNDERLYING SYSTEMS CAN DO]
Tasks the agent must accomplish: [LIST]
Systems of record: [WHAT IS AUTHORITATIVE FOR WHAT]

Do NOT map capabilities to tools one-to-one.

Produce:
1. The minimum tool set covering every task
2. For each tool: name, purpose, which tasks it serves, what it
   composes internally
3. Capabilities deliberately excluded, with reasoning
4. Composite tools — sequences that always occur together and
   should be a single call
5. Tools that should be split — anything doing two unrelated things
6. Which tools are read-only and which have side effects, clearly
   separated
7. Which side-effecting tools are reversible and which are not
8. Tasks that cannot be accomplished with this surface

Design principles to apply:
- Fewer tools with clear boundaries beat many tools with
  overlapping ones
- A tool should map to something the agent decides to do, not to
  an implementation detail
- Read and write should be separable so read-only agents are
  possible
- Irreversible operations should be individually identifiable so
  they can be gated