# Day 04 — Agent Job Card

Manual: Guide 04 / Agent Workflow Builder  
Output: one narrow review-only AI job with an approved plan and proof loop.

## 10-year-old mode

What boring prep job can AI do while a human checks the result?

## Fill this card

- Agent job name:
- Trigger / when it runs:
- Input it may read:
- Output it must create:
- Allowed actions:
- Forbidden actions:
- Source-of-truth file or folder:
- Business rule AI must not guess:
- Edge case to handle:
- Acceptance criteria:
- Human review step:
- Proof-loop check:

## Safe AI help

Ask AI to draft the job card, approved-plan fields, and proof-loop checklist.

## Human approval lock

No sending, posting, deploying, contacting people, buying, deleting, account changes, credential use, or private identity/payment/tax/KYC handling.

## Done means

The agent has one job, one output, one review gate, and no permission to create real-world risk alone.