paw-cra-quality-control
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to use
ffprobe(part of ffmpeg) to extract metadata (codec, resolution, duration, audio) programmatically from video assets. The file paths for these assets are sourced from external manifest files (video-manifest.json,clip-manifest.json). There is a potential risk of command injection if the agent passes these unsanitized paths directly to a shell, as a malicious manifest could contain filenames with command delimiters (e.g., backticks or semicolons). - [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests and processes untrusted data from multiple external sources.
- Ingestion points: Campaign manifests in JSON format (
design-manifest.json,video-manifest.json, etc.), brand guidelines in Markdown, and local configuration files in YAML (config.yaml,config.user.yaml). - Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are specified for the processing of manifest content.
- Capability inventory: The skill has capabilities to read/write files, append to project logs, and execute system commands (
ffprobe). - Sanitization: No sanitization, validation, or escaping procedures for the external content are mentioned in the instructions.
Audit Metadata