Claw Patrol - Security Firewall for Agents

This is an agent-only security firewall designed to prevent the security risks that arise when granting AI agents production access. It manages agents so they don't directly hold credentials, and parses all traffic at the wire level to gate actions according to user-defined rules.

Core Features:

• Precise Action Control: Going beyond the limitations of existing IAM or RBAC, it inspects down to granular units of behavior, such as SQL verbs (e.g., blocking DROP TABLE), Kubernetes resources, and HTTP methods.
• Intelligent Approval Flow: For ambiguous requests, it can be configured so that an LLM Judge makes the decision, or so that it goes through Human In The Loop approval via Slack, etc.
• Security and Auditing: Keys are stored outside the agent so that credentials are not leaked in the event of prompt injection, and it provides integrated audit logs for various systems such as Postgres, GitHub, and Slack.
• Regression Testing: Real actions are recorded as JSON fixtures, allowing verification in a CI environment of whether existing behavior is affected when policies change.

While existing tools tend to focus on either LLM call monitoring or simple sandboxing, Claw Patrol's distinguishing feature is that it integrates protocol-layer monitoring, secrets management, risky call routing, and full logging all together.