kelly-portrait-retouch
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/retouch.mjsexecutes a local Swift scriptscripts/detect-faces.swiftusing theswiftinterpreter vianode:child_process.execFile. This is used for face detection on macOS platforms. - [DATA_EXFILTRATION]: The scripts
scripts/sync-candidate.mjsandscripts/lib/portrait-busabase.mjsperform network operations to upload local image files (source, output, and comparison) to a remote Busabase instance configured in the environment. - [CREDENTIALS_UNSAFE]: The skill's synchronization logic in
scripts/lib/portrait-busabase.mjsreads sensitive configuration includingBUSABASE_API_KEYfrom environment variables to facilitate authentication with the Busabase API. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local image files which constitutes an indirect prompt injection attack surface.
- Ingestion points: The
INPUTimage path processed byscripts/retouch.mjs. - Boundary markers: No explicit delimiters or instructions to ignore embedded content are used when processing image data or results.
- Capability inventory: File system access via
sharpandfs, local command execution ofswift, and network exfiltration to Busabase viafetch. - Sanitization: While image processing is performed by the established
sharplibrary, metadata and counts extracted from the images are returned directly to the agent's context.
Audit Metadata