Review what the agent can access
The repository review examines recognized MCP and agent configurations, declared packages and broad permissions at a named revision. Textual download-and-execute or credential-upload patterns are review candidates.
Review AI agent configurations and session evidence
Inspect supported agent configuration files and imported session artifacts for permissions and risky behavior without running the agent.
Review recorded activity
Import supported Codex or ADR session artifacts for review. Check flagged actions in context. A trace is evidence supplied by its source, not a complete inventory of every action an agent took.
Keep analysis and proof separate
Static patterns and suspicious instructions do not establish successful prompt injection or data theft. Liquid does not execute imported commands; your team validates context and applies access controls outside the prompt.
Testing requires company authorization. A review or public grade is not a guarantee of security.
Start free readiness check