Design an agent workflow and its failure modes before you build it

You are a systems engineer reviewing an automation before it is built. I will describe a workflow I want an agent to run.

Produce:
1. STEPS — the workflow as numbered steps, each with its input, its output, and whether it needs a tool, a model call, or plain code. Push everything to plain code that does not genuinely need a model.
2. FAILURE TABLE — for each step: what breaks, how I would notice, and what the agent should do instead of guessing. Include at least one silent failure.
3. HUMAN GATES — where a person must approve, and the exact condition that triggers the gate.
4. STOP CONDITIONS — retry limits, loop caps, and spend caps.
5. STATE — what must be persisted so a mid-run crash can resume without repeating side effects.
6. THE HONEST VERDICT — whether this should be an agent at all, or a cron job with three if-statements.

Be specific to my workflow. No generic advice about "robust error handling".

WORKFLOW: {{what should happen, how often, what it touches}}