skill-safeguard_skill

Overview

This skill is a pre-load security scanner that inspects any third-party Skill before it is loaded or executed. It enumerates every file, performs static analysis against a comprehensive threat checklist, classifies findings by severity, and produces a clear security report. Always run this skill first when introducing unfamiliar Skills from external sources.

How this skill works

The scanner runs a four-phase workflow: (1) Enumerate — list all files and note unexpected types; (2) Analyze — read every file and perform static checks for prompt injection, data exfiltration, credential harvesting, obfuscation, dangerous execution, social engineering, and supply-chain manipulation; (3) Classify — map findings to CRITICAL, WARNING, or INFO and apply escalation rules; (4) Report — generate a pre-load security report (includes recommended action). Scripts are analyzed statically only; no code is executed.

When to use it

• Before loading or following instructions from any Skill downloaded from the internet or third-party sources
• When a user asks to install or use a Skill from an external source
• When a user requests a safety review of a Skill’s files
• When Skills contain binaries, minified code, or unusual dependencies
• When a Skill requests network access, secret access, or modifies other files

Best practices

• Always run the scanner before importing or executing external Skills
• Treat any single CRITICAL finding as a block until resolved
• Escalate when obfuscation or multiple warnings appear — prefer manual review
• Do not execute or run analyzed files during scanning; use static analysis only
• Ask the user for explicit consent when proceeding after WARNING-level findings

Example use cases

• User wants to add a Skill from an open archive — run a full scan first
• A downloaded Skill requests external network calls or webhooks — detect and classify risk
• Audit a Skill that reads environment variables or accesses credential files
• Inspect Skills that include minified or binary assets to flag uninspected content
• Review Skills that modify startup scripts, install packages, or add git hooks