clawdefender

To prevent prompt injection, you scan external data sources to sanitize inputs and block unsafe AI behavior. This process detects malicious patterns, assigns configurable severity scores, and enforces safety by blocking or warning based on the highest detected threat score.

Auditing installed AI agent skills for malware involves running a full security scan that checks scripts for unsafe patterns. The audit identifies credential exfiltration, command injection, and path traversal risks, reporting findings via a JSON output for automation.

Yes, jq is required to parse JSON security audit reports. The scanning mechanism exposes a JSON output for automation, allowing you to programmatically extract and process detected threat findings, severity scores, and safety enforcement results.

Blocking SSRF and data exfiltration requires validating URLs and scanning external inputs before processing. This pattern-based detection method identifies unauthorized network requests and credential leakage, enforcing blocks or warnings based on configurable severity scores.

AI safety scanning works during skill installation and input processing. It can be invoked to audit newly installed skills for malware or to sanitize real-time external data, ensuring continuous protection against command injection and prompt injection threats.