The Failure Mode Enumerator

Enumerate the failure modes of this agent design.

Design: [PASTE OR SUMMARISE]

For each failure mode, give:
1. What goes wrong
2. What causes it
3. How it presents — does it fail loudly, fail silently, or produce
   a plausible wrong answer?
4. Who notices, and how long it takes
5. What it costs
6. Prevention: guardrail, validation, tool design, or prompt
7. Detection: what signal would surface it
8. Recovery: what happens after

Group by category:
- Model does the wrong thing (wrong tool, wrong arguments, wrong
  reasoning)
- Model does nothing (gives up, loops, stalls)
- Tool fails (unavailable, slow, wrong result, partial success)
- State problems (stale context, lost state, contaminated memory)
- Runaway behaviour (infinite loop, cost explosion, cascading
  retries)
- Environment problems (rate limits, auth expiry, downstream
  outages)

Rank all findings by expected cost — probability multiplied by
severity — not by how interesting they are.

Then: which three would you fix before shipping anything?