cawcut-ad-video
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill manages several external resources: it requires the installation of the
@ubnt/cawcutCLI via NPM and downloads project media (images and videos) from pre-signed URLs provided by the CawCut API using bundled scripts. - [COMMAND_EXECUTION]: The skill uses the CawCut CLI and runs several bundled local scripts (
poll.sh,script_plan.ps1,storyboard_review.sh, etc.) to automate the video creation pipeline. On Windows, it executes PowerShell scripts with the-ExecutionPolicy Bypassflag to ensure functionality within the host environment. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external product pages (e.g., Amazon or Shopify URLs) to generate video scripts. While this represents an attack surface for indirect prompt injection, it is the primary intended function of the skill and is mitigated by the fact that the resulting analysis is presented to the user for review and confirmation.
- Ingestion points: Product URLs provided by the user in
SKILL.md(Stage 1). - Boundary markers: The skill relies on the back-end API to process and structure the external content into specific JSON fields, limiting the impact of any embedded instructions.
- Capability inventory: The skill possesses shell execution (Bash/PowerShell), file writing (for local script plans), and network access (CLI/curl).
- Sanitization: All data returned from the API is structured as JSON; the skill includes explicit user-confirmation steps for product information, script edits, and credit expenditures.
Audit Metadata