Audit-Verlauf

The static Ruby backtick and hardcoded URL findings are false positives caused by Markdown code spans, fenced command examples, model IDs, and documentation links. No prompt injection or hidden malicious behavior was found in SKILL.md. The remaining concern is expected external processing of prompts and media URLs through the inference.sh belt CLI.

The static Ruby backtick and hardcoded URL findings are false positives caused by Markdown code spans, fenced command examples, model IDs, and documentation links. No prompt injection or hidden malicious behavior was found in SKILL.md. The remaining concern is expected external processing of prompts and media URLs through the inference.sh belt CLI.

The static backtick and weak-crypto findings are mostly false positives caused by Markdown code fences, inline model identifiers, and the word Description. The real risk is that the skill authorizes Bash(belt *) and instructs agents to submit prompts and user media URLs to external inference.sh model services, which is legitimate for the skill but requires user awareness.

The skill is a legitimate AI video generation tool using the inference.sh CLI (belt). Static findings flagged 59 external_command locations and 19 network locations, but evaluation confirms all are FALSE POSITIVES. The backtick syntax in SKILL.md markdown code blocks is documentation only (not executable Ruby), and hardcoded URLs reference legitimate service documentation and user-provided media URLs. No command injection vectors exist. The skill only invokes the belt CLI with hardcoded app IDs and JSON parameters, no user input is interpolated into shell commands.