The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses multiple shell commands for file management and processing. Evidence includes find assets/brand -type f -exec md5 -r {} \; for duplicate detection and sips -s format jpeg in.heic --out out.jpg for image conversion in SKILL.md. Additionally, an inline Python verification script in references/manifest.md executes file and sips via subprocess.run to validate file integrity.
[REMOTE_CODE_EXECUTION]: The instructions suggest executing an external shell script for installation: sh <sp-canvas skill dir>/scripts/install.sh. This pattern involves executing code from a path outside the immediate skill directory.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to various external services to fetch brand data and images. It fetches configuration and screenshots from well-known services such as Apple's iTunes API and the Google Play Store, as well as arbitrary marketing and press URLs provided as input.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted sources such as marketing websites and social media platforms, creating a surface for indirect prompt injection.
Ingestion points: curl fetches data from the App Store, Google Play, and user-provided URLs in SKILL.md and references/sources.md.
Boundary markers: None explicitly implemented for the content fetched from third-party sites.
Capability inventory: The skill has access to subprocess.run (via the verification script), sh (installation), and curl for network operations.
Sanitization: The skill performs file type and dimension checks but does not explicitly sanitize filenames or HTML content against injection before processing.

sp-brand-kit