Prompt Review Policy
This page is a policy starter for reviewing shared prompts before team-wide use.
Policy Objective
Prevent unsafe or low-signal shared prompts from spreading across teams.
Scope
- shared prompt libraries
- repo instruction prompts
- automation prompts
- prompts that can run commands, modify code, or access MCP servers
Audience
- maintainers of prompt libraries
- reviewers
- team leads
Mandatory Requirements
- shared prompts must name their objective and expected output
- prompts that can edit code, run commands, deploy, or use MCP must include human review checkpoints
- prompts must not encourage secret sharing or unsafe destructive actions
Recommended Controls
- keep prompts versioned in-repo
- label prompts by risk level
- pair prompts with example outputs and validation commands
Approval Process
Review prompts before publishing them for broad team use when they:
- edit code
- run commands
- access external systems
- touch production or deployment workflows
Audit Evidence
- prompt source file
- reviewer sign-off
- example output
Enforcement
Unsafe shared prompts should be removed or downgraded to private experimentation.
Copy-Ready Policy Template
Shared prompts that edit code, run commands, deploy, or access MCP servers must include:
- objective
- required context
- validation steps
- human review checkpoint
- security note
Adoption Checklist
- prompts versioned in repo
- risk-labeled
- reviewer assigned
Verification Note
Template only. Teams should align prompt review depth with their change risk.