suede-product-marketing
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the repository (README, landing pages, marketing copy) to auto-draft marketing context.
- Ingestion points: Reads local project files (README, landing pages, package.json, etc.) during the auto-drafting phase defined in SKILL.md.
- Boundary markers: The instructions require the agent to cite the source file for every drafted field (e.g., [src: path/to/file.md]) and mark missing information as [unverified], which helps the user track the provenance of the content.
- Capability inventory: The skill utilizes file system read access to analyze the codebase and write access to create or update the .agents/product-marketing.md file.
- Sanitization: The skill lacks programmatic sanitization of the files it reads, relying instead on instructional constraints (e.g., 'Draft only what the sources actually say') and user review to mitigate the risk of processing malicious instructions embedded in project files.
Audit Metadata