The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill interacts with the Unsplash API to search for and download images and attribution metadata.
Evidence: The script scripts/generate_blog_image.py makes network requests to api.unsplash.com and related Unsplash image domains using the Python standard library urllib.
[COMMAND_EXECUTION]: The skill instructions involve executing a local Python script to perform image manipulation (cropping, dithering, pixelation).
Evidence: Instructions in README.md and SKILL.md direct the user/agent to run python3 scripts/generate_blog_image.py with various arguments.
[INDIRECT_PROMPT_INJECTION]: The skill processes external metadata from the Unsplash API, which represents a potential surface for indirect prompt injection if an attacker were to poison Unsplash metadata fields.
Ingestion points: scripts/generate_blog_image.py processes JSON responses from the Unsplash API (request_json and unsplash_metadata functions).
Boundary markers: None explicitly present for isolating external metadata in the output JSON files, though the data is handled as structured JSON.
Capability inventory: The skill has the capability to write files to the local filesystem and perform network operations to Unsplash endpoints.
Sanitization: The script utilizes standard JSON parsing and implements a scrub function to ensure the UNSPLASH_ACCESS_KEY is not leaked in error messages or logs.

blog-image-generator