spritecook-polish-sprites
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill incorporates security best practices by explicitly instructing the agent to keep sensitive signed URLs out of persistent manifests and local files to prevent accidental leakage.
- [SAFE]: The bundled helper script (scripts/compare_alpha.py) implements resource protection by limiting image processing to a maximum of 16 million pixels, preventing potential denial-of-service through resource exhaustion during local processing.
- [SAFE]: The helper script includes strict input validation for alpha cutoffs and metadata fields, and it deliberately omits sensitive source URLs from its generated JSON reports to ensure that temporary credentials are not logged or stored.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external asset metadata and image files, creating an attack surface for indirect prompt injection or file-parsing exploits. Ingestion points: asset.json (metadata), preserved.png, and current.png. Boundary markers: The agent uses a dedicated local script with a fixed CLI, isolating the data processing. Capability inventory: File system writes to a local comparison directory and tool calls to update asset metadata. Sanitization: The implementation includes pixel count limits, RGBA conversion checks, and specific schema validation for metadata fields.
Audit Metadata