Agent policy analysis

Lint the instructions behind the assistant.

Static analysis for system prompts: objectives, tool-call conditions, trust boundaries, sensitive-data handling, uncertainty behavior and conflicting policies.

No model requiredClause-level conflictsClient-side only

System prompt / agent policy

Processed in this tab

Results

Deterministic checks

Ready to lint

The tool checks policy structure and known contradictions without executing the instructions.

Scope

Static analysis, not a safety guarantee

This linter looks for visible policy gaps and contradictions. It cannot prove that an agent is safe, aligned or secure, and it does not simulate model behavior.

Current checks

Identity and objective signals, tool invocation conditions, external-content trust boundaries, sensitive-data handling, fallback behavior, output policy and clause-level positive/negative conflicts.