Hierarchy attacks
Override attempts, privileged role impersonation, and policy-evasion language.
STATIC DEFENSE FOR AGENT INPUTS
Catch the hidden claw before your agent does.
Paste an instruction, retrieved document, tool output, or message. ClawDefender returns a deterministic risk verdict and practical guardrails before your agent acts.
DeterministicProduction service$0.01 / scanNo model call
Static analysis reduces obvious agent risk but cannot prove content is safe. Keep least-privilege tools, domain allowlists, and human approval for destructive actions.
INTERACTIVE PAYWALL CONSOLE
Scan the instruction boundary.
The free quote validates your payload. A paid run scores suspicious signals and returns an allow, review, or block verdict after settlement.
WHAT THE CLAW CATCHES
Small signatures.
Useful friction.
Secret grabs
Requests to reveal tokens, private keys, recovery phrases, passwords, or environment files.
Tool hazards
Destructive commands, internal network targets, opaque payloads, and external transfers.
FOR AUTONOMOUS CLIENTS
One scan between retrieval and action.
Start with a free quote. Confirm the spend in your own policy, satisfy the x402 or MPP challenge, then consume the structured verdict.
FREE POST /quote
$0.01 POST /v1/run
OUTPUT allow · review · block
curl
TypeScript
Quote
Validate the exact JSON body and read the fixed price.
Confirm
Apply your agent's own spend limit and approval policy.
Pay
Retry the hostname-bound challenge with an idempotency key.
Decide
Read the receipt, risk score, findings, and verdict.