SkillTotal AI Component Security Scan

August 21, 2026

Version updated for https://github.com/pezhik/skilltotal to version v0.42.0.

This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

SkillTotal is an open-source CLI tool that analyzes AI-related components, including agent skills/plugins, MCP servers, npm packages, Python packages, repositories, and AI-generated projects. It statically scans these components to identify security risks, dangerous capabilities, prompt-injection surfaces, and data-exfiltration paths before installation or trust. SkillTotal avoids running components on the user’s machine and provides deterministic, evidence-anchored reports that are free of false positives.

What’s Changed

Added

Credential-only files are detected (ruleset 46). A file that is a credential rather than one that contains one had no pattern to match: no assignment, no vendor prefix. The MCP publisher’s login artifacts (.mcpregistry_github_token, .mcpregistry_registry_token) are now read as the credentials they are. The registry token grants republish rights over the server, so shipping one in a package is a standing supply-chain takeover of that component.

Fixed

A long secret could survive redaction in the report. Evidence snippets are capped at 240 characters before redaction ran, so snippet.replace(value, marker) silently did nothing whenever the secret was longer than the cap — leaving most of a live credential (a JWT, a PEM body, a token on a minified line) in the output. Redaction now removes the longest prefix of the secret that is actually present. A security report must never re-publish what it found.