video-ad-specs
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install the
belt-sh/clivianpxand recommends adding several extensions from theinference-shGitHub organization. These are external dependencies originating from sources outside the recognized trusted ecosystem. - [COMMAND_EXECUTION]: The skill uses the
Bashtool to executebeltCLI commands. These commands trigger remote execution of AI models and media processing workflows on theinference.shplatform. - [INDIRECT_PROMPT_INJECTION]: The skill contains templates for shell commands that take natural language prompts as input (e.g.,
belt app run ... --input '{"prompt": "..."}'). This creates a surface where the agent might interpolate untrusted data into command arguments without validation. - Ingestion points: Prompt fields within the JSON payload of
beltcommands inSKILL.md. - Boundary markers: None; input strings are passed directly into CLI arguments.
- Capability inventory: Shell command execution via
Bashtool. - Sanitization: No evidence of prompt sanitization or validation logic is provided.
Audit Metadata