Hello Minimax Security Audit

What Problem It Solves

A Skill is often more than documentation. It may include scripts, install commands, remote URLs, and dependency setup. The real risk is not the presence of curl, sudo, or ~/.ssh; it is whether the skill automatically executes those operations and sends sensitive data to untrusted destinations.

This skill treats the audit boundary as supply-chain poisoning risk: it focuses on skill.md plus related docs, scripts, and programs to detect dangerous behavior such as automatic remote script execution, access to cloud credentials or private keys, writes to sensitive system paths, hidden command execution, and mismatches between the stated description and actual behavior.

How It Works and Where It Applies

• Static audit: It reads files, searches keywords, inspects directory structures, and fetches URL text, but it does not execute commands from the skill under review.
• Dangerous behavior detection: It scans for command execution, network requests, privilege escalation, hidden execution, and file operations, then evaluates whether the surrounding context forms a real risk.
• Sensitive path review: It checks paths such as ~/.ssh, .env, ~/.aws, ~/.kube, and C:\Users, together with read, write, delete, or exfiltration actions.
• Remote script deep analysis: When automatic remote execution is found, it retrieves the script text and evaluates second-stage downloads, data exfiltration, persistence, system damage, privilege escalation, and environment tampering.
• Dependency risk checks: It reviews pip install, npm install -g, non-official indexes, and git+https:// installs, separating environment pollution from supply-chain risk.
• Graded reporting: It outputs P0, P1, or P2 conclusions with line numbers, code snippets, risk behaviors, and review recommendations.

Use it with these limits in mind: it does not grade teaching-code quality, so examples with SQL injection or missing error handling are not reported by themselves. If a remote URL is inaccessible or cannot be statically parsed, the result is high risk because safety cannot be verified. Even when a remote script appears benign, the remote execution pattern may still keep the finding at a caution level because content can be swapped or tampered with in transit.