Security checks across malware telemetry and agentic risk
This OpenClaw plugin transparently adds a local tool-call approval and audit layer, with a real but disclosed limitation that its heuristic classifier can miss nonstandard or evasive tools.
Install only if you want an OpenClaw dispatcher-level approval and audit layer. Treat it as best-effort policy enforcement, not a complete security boundary; review blockOn and approvalOn defaults, keep timeout behavior set to deny for sensitive work, and avoid relying on it against custom or adversarial tools that may not match its classifier patterns.
params: Record<string, unknown>,
): ClassificationResult | null {
const lower = toolName.toLowerCase();
const isDelete = /delete|remove|unlink|rm\b/.test(lower);
const isWrite = /write|edit|put|patch|create|move|rename/.test(lower);
const isRead = /read|get|cat|stat|list|ls\b/.test(lower);
if (!isDelete && !isWrite && !isRead) return null;62/62 vendors flagged this plugin as clean.