`TASK-TEMPLATE-WITH-RALPH-GATES.md`

71 lines
# YOLO Task Template — With Ralph Clarifying Gates

Use this template when creating new YOLO tasks. The Ralph gate ensures
the autonomous agent investigates before acting — preventing destructive
changes based on assumptions.

---

```markdown
# [Task Title — Action-Oriented]

## Ralph Clarifying Questions (Self-Check Before Executing)

Before making ANY changes, investigate and answer these questions.
Print your answers before proceeding. If any answer is unclear or
contradicts the task assumptions, STOP and log the issue instead
of proceeding blindly.

1. **[Current State Question]**
   - What is the current state of the thing we're about to change?
   - Commands to investigate...

2. **[Impact Assessment Question]**
   - What will be affected by this change?
   - How many products/pages/users does this touch?

3. **[Dependency Question]**
   - Does anything else depend on the thing we're changing?
   - Will other systems break if we modify this?

4. **[Reversibility Question]**
   - Can this change be reversed if something goes wrong?
   - What's the rollback plan?

5. **[Validation Question]**
   - How do we verify the change worked correctly?
   - What does "success" look like? What does "failure" look like?

## What To Do (ONLY after answering above)

### Step 1: [First action]
- Details and commands...

### Step 2: [Second action]
- Details and commands...

## DO NOT (Safety Rails)
- List actions that should NOT be taken
- Common mistakes to avoid
- Destructive operations that require human approval

## Verification (MANDATORY)
- Specific checks to confirm success
- Expected output/status codes
- Report format: what to include in the log

## Credentials
- [Only include what's needed for this specific task]
```

---

## Key Principles

1. **Investigate before acting** — The Ralph gate forces the agent to
   read current state before making changes
2. **Print answers** — Makes the investigation auditable in logs
3. **STOP on confusion** — If answers contradict assumptions, log and halt
4. **Verify after acting** — Every task ends with mandatory verification
5. **Safety rails** — Explicit "DO NOT" section prevents common mistakes