← Back to Checklists

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

🤖

Full checklist coming soon

I'm finalising this checklist. Join updates to receive it when it drops — along with other practical security resources.

Join Security Tool Updates →

What this checklist covers

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 full checklist