# Human Approval Gates Template

By John & Dexter / The AI Fixer

Use this template before letting an AI agent move from drafting into action.

## Workflow name

Name:
Owner:
Date:
Agent/model:
System or channel:

## Intended job

The agent may:

1.
2.
3.

The agent must not:

1.
2.
3.

## Risk class

Choose one:

- LOW: read-only, no external side effects.
- MEDIUM: edits/drafts with review before public/customer impact.
- HIGH: public/customer/business impact, requires explicit approval.
- NEVER-AUTO: not allowed for this agent.

Why:

## Approval gates

| Gate | Trigger | Who approves? | What evidence is required? | What happens if rejected? |
|---|---|---|---|---|
| Draft gate | Agent creates a proposed output | | | |
| Send/publish gate | Output leaves private workspace | | | |
| Spend gate | Money, bids, offers or purchases | | | |
| Production gate | Deploy/restart/live config change | | | |
| Sensitive-data gate | Private/customer/internal data used | | | |
| Safety/legal gate | Advice could affect safety, warranty or rights | | | |

## Tool permissions

Allowed tools:

- 

Disallowed tools:

- 

Time or scope limits:

- 

## Verification

Before approval, check:

- [ ] output matches the task;
- [ ] source/evidence is visible;
- [ ] no secret/private data leaked;
- [ ] claims are not stronger than proof;
- [ ] rollback path exists if anything changes;
- [ ] logs are enough to audit later.

## Rollback / stop path

If this goes wrong, the owner can:

1.
2.
3.

Where the previous state lives:

## Final decision

- [ ] Approved
- [ ] Approved with edits
- [ ] Rejected
- [ ] Needs more evidence

Approved by:
Time:
Notes: