← Back to Guides

AI Agent Prompt Security

A starting checklist for reviewing AI-agent prompts, tool access, risky instructions, and CI-friendly prompt checks. Reduce prompt injection risk, tighten trust boundaries, and make LLM pipelines safer to operate.

Last updated 27 Jul 2026 · maintained by Myles Agnew

🤖

In progress

Only the scope below is published so far. Join updates to hear when the checklist is finished.

Join Security Tool Updates →

Preview the scope

Audit system prompts for sensitive data leakage or overly broad permissions
Review tool access — does the agent need all the tools it has access to?
Check for prompt injection vectors in user-controlled inputs
Verify trust boundaries between agent roles and external data sources
Review output handling — are responses validated before acting on them?
Assess whether the agent can be instructed to ignore its system prompt
Add CI-friendly prompt lint checks to catch risky patterns pre-deployment
Log agent interactions for auditability and anomaly detection
+ more in the finished checklist

Subscribe for updates

Get new security tools, checklists and guides by email. No spam — unsubscribe anytime.

Please don't include passwords, secrets, or client data.