moda-social-youtube
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided prompts (e.g., episode briefs, channel taglines, and photo descriptions) and interpolates them into design layouts, creating a potential surface for prompt injection attacks if that data contains malicious instructions.
- Ingestion points: User-supplied strings in 'SKILL.md' (e.g., '[one-line brief]', '[Channel] — banner') and content retrieved from existing canvas nodes as described in 'references/design-quality.md'.
- Boundary markers: The skill utilizes an XML-like markup structure (e.g., '', '') to encapsulate content, providing a degree of structural isolation.
- Capability inventory: The agent can execute 'moda' bash commands, search for project files, generate remote AI media, and process JavaScript-based canvas mutations.
- Sanitization: The skill provides strict 'Data preservation' rules, instructing the agent to treat source data as a 'canonical record' and relay values 'verbatim' without interpretation.
- [DYNAMIC_EXECUTION]: The skill uses a specialized tool called 'moda canvas edit' that executes JavaScript to batch-update canvas design properties, which is a form of local dynamic code execution.
- Evidence: As documented in 'references/edit-code.md', the agent generates and executes JavaScript snippets to update node colors, positions, and content. The platform employs a sandbox that proactively blocks potentially dangerous operations such as 'eval', 'require', 'fetch', and 'process' to maintain a secure execution boundary as part of the tool's core design.
Audit Metadata