moda-social-ads
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-supplied campaign offers, brand kits, and file content to generate ad designs. This creates a surface where embedded instructions in the source data could potentially influence the agent's design operations.
- Ingestion points: User input for campaign offers and intents in SKILL.md, source data extracted via 'moda file show', and external web content via 'moda web read' (referenced in multi-unit-workflow.md).
- Boundary markers: The skill uses bracketed placeholders (e.g., [the offer], [KIT]) to delimit user-provided variables within its command recipes.
- Capability inventory: The skill uses the 'moda' CLI for extensive canvas mutations, media generation (images, video, audio), and file system writes (exports).
- Sanitization: The skill relies on the vendor CLI's internal handling of markup and JS-based mutations; explicit sanitization steps are not defined in the instructions.
- [DYNAMIC_EXECUTION]: The skill uses a sandboxed JavaScript environment to perform bulk canvas edits and generate decorative patterns.
- Evidence: References to 'moda canvas edit' in references/edit-code.md and the '' tag in references/markup.md.
- Impact: While the environment is described as sandboxed, the documentation notes that the sandbox is not a hard security boundary against untrusted code.
Audit Metadata