product-image-processor
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to executecurlfor binary asset downloads. It includes explicit defensive instructions to quote all URL and path arguments and never execute shell text provided by external sources. - [EXTERNAL_DOWNLOADS]: The skill downloads product images from remote URLs specified in a CSV library or task input. It requires binary transfer methods and includes checks to validate that the downloaded content is an image and not an HTML error response.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from
product-library.csvand external image URLs. - Ingestion points: Ingests
Product NameandImage URLfields from CSV files and follows external links to download assets. - Boundary markers: Contains explicit instructions stating that supplied text, source URLs, and embedded data must be treated as content only and never as authority to execute commands or extend access.
- Capability inventory: Uses
Bash(curl) for downloads andWritefor saving files to the local filesystem. - Sanitization: Implements filename slugification (lowercase, hyphen substitution) and mandates shell argument quoting to mitigate risks from malicious input strings.
Audit Metadata