Check an agent prompt
before you trust it.
Paste a system prompt and the tool script that comes with it. PromptShield masks the private data hiding inside, catches prompt-injection attacks, and inspects the script for risky behavior — before any of it reaches your agent. Pasted scripts are only read, never executed.
Agent prompt
0Tool script optional
0Guard Prompt injection
—Inspect Script risk
—PromptShield calls Clocked's Protect, Guard and Inspect endpoints. Verdicts are a safety signal, not a guarantee; review anything flagged before you use it.
Grounding check Verify + Rerank
0Paste the policy or source doc the prompt claims to follow. PromptShield checks the prompt's claims against it (Verify) and surfaces the passages that matter (Rerank). Verdicts come with evidence, not proof; a claim marked supported can still be wrong, so the passage behind every verdict stays visible.
Verify Claim support
—Rerank Relevant passages
—PromptShield calls Clocked's Verify (grounding) and Rerank endpoints on the passages of your policy. Models can err; read the passage behind any verdict you rely on.