DeepInspect Guardrails
PassAudited by ClawScan on May 1, 2026.
Overview
This is a small local command-risk classifier that does not execute commands, access credentials, or use the network.
This skill appears safe to install as a local advisory classifier. Before relying on it for enforcement, review and tune the policy and do not let an allow result replace human approval for high-impact commands.
Findings (1)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If another agent treats an allow result as a complete safety guarantee, risky commands not covered by the baseline patterns could proceed without extra review.
The classifier only escalates commands when its pattern checks produce reasons; with the bundled policy's default action set to allow, commands outside the configured patterns receive an allow decision.
let decision = policy.actions.default; if (reasons.includes('REMOTE_EXEC_PATTERN') || reasons.includes('OUTSIDE_WORKSPACE_PATH')) { decision = policy.actions.critical; } else if (reasons.length > 0) { decision = policy.actions.highRisk; }Use this as an advisory preflight check, tune the policy for your environment, and keep separate human approval for destructive, privileged, or unfamiliar commands.