video-ad-specs
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [METADATA_POISONING]: The skill includes an auxiliary file
skill-report.jsoncontaining a pre-computed security audit result. This file claims a 'safe' verdict and 'analysis_status: ok', which is a self-referential pattern designed to target automated scanners and influence the final security assessment. - [INDIRECT_PROMPT_INJECTION]: The skill defines workflows that ingest untrusted user input (prompts for video generation) and interpolates them into JSON payloads for shell-based tool execution (
belt app run). - Ingestion points: User-provided strings for prompts, voiceovers, and captions used in
SKILL.mdBash examples. - Boundary markers: None identified; user strings are inserted directly into JSON structures within shell commands.
- Capability inventory: The skill uses
Bash(belt *), enabling network access to the inference service and local execution of thebeltCLI tool. - Sanitization: No evidence of input validation, escaping, or filtering for user-provided strings before they are passed to the CLI.
- [EXTERNAL_DOWNLOADS]: The skill documentation encourages the installation of external components from remote sources.
- The instruction
npx skills add belt-sh/clidownloads and executes a package from a third-party registry. - The skill references an installation script located at
https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md.
Audit Metadata