Openclaw Plugin

Precision decisioning, agentic trust, and verifiable identity for autonomous agents

This plugin appears to implement what it claims — it will intercept high-risk tool calls and POST verification/attestation data to governance.taskhawktech.com. Before installing: (1) Confirm you trust the external gateway (review their privacy policy, source repo, and the gateway URL) because agent inputs/outputs (possibly sensitive) are transmitted off-host; (2) Test in advisory mode (mode: "advisory") first to avoid fail-closed blocking and to see what data would be sent; (3) Prefer providing your own KEVROS_API_KEY in config rather than allowing the plugin's auto-signup if you want control over provisioning and rate limits; (4) Consider restricting highRiskTools to only the truly sensitive tools for your deployment; (5) If you need stronger guarantees, verify the gateway implementation (server-side handling of payload hashing and storage) before sending production data.

Type: OpenClaw Skill Name: openclaw-plugin Version: 0.3.9 The Kevros plugin intercepts high-risk tool calls (e.g., bash, exec, write_file) and sends tool inputs and truncated outputs (up to 500 characters) to an external endpoint (governance.taskhawktech.com) for 'governance' and 'attestation.' While this aligns with its stated purpose of auditing, the code in dist/index.js and src/index.ts explicitly transmits actual tool results, which contradicts the claim in SKILL.md that 'actual data is never stored' and only hashes are used. This creates a significant data leakage risk if the agent handles sensitive information like credentials or private keys, though no evidence of intentional malware or backdoors was found.