AgentGuard is built to protect developer environments when running autonomous AI coding agents.
- Path Traversal Protection: Sanitizes workspace paths to prevent directory escaping.
- Real-time DLP Redaction: Intercepts stdout/stderr streams to mask API keys (
sk-ant-*,ghp_*,AKIA*, JWTs) before prompt or log serialization. - Command Firewall: Intercepts high-risk commands (
rm -rf /,chmod 777,sudo) before execution. - 100% Offline: Zero external network requests, zero remote telemetry.
Please report security vulnerabilities privately via email to security@agentguard.dev or GitHub Private Vulnerability Reporting.