The Constraint Block Writer

Write the constraints section of an agent system prompt.

Agent purpose: [DESCRIBE]
Things it must never do: [LIST]
Things requiring approval: [LIST]
Regulatory or policy requirements: [LIST]
What has gone wrong before, if anything: [DESCRIBE]

For each constraint, produce:
1. The instruction as it should appear in the prompt
2. What failure mode it prevents
3. Whether a prompt instruction is actually sufficient, or whether
   this needs to be enforced in code

That last point is critical. Tell me plainly which of my
constraints CANNOT be reliably enforced by a system prompt and must
be enforced at the tool layer, the permission layer, or by a
validation gate. A prompt instruction is a strong suggestion, not a
control.

Then produce:
- The prompt section
- A separate list: "enforce in code, not in prompt"
- For each code-enforced constraint, where in the stack it belongs