The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/well.py to execute the open command on macOS, which launches generated HTML search results in the default web browser. This functionality is intended for local user review and is restricted to opening the generated preview files.
[EXTERNAL_DOWNLOADS]: The skill performs HTTP GET requests to fetch images and metadata from numerous third-party APIs, including NASA, Smithsonian, Met Museum, and Openverse. It supports downloading these assets to a local directory via the download_results function in scripts/well.py based on user-supplied queries.
[DYNAMIC_EXECUTION]: The scripts/sources/__init__.py file utilizes importlib.import_module to load internal image source provider classes. Module names are sourced from a hardcoded registry, which prevents arbitrary or unverified module loading at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from various external image APIs.
Ingestion points: Data enters the agent context via API responses processed in search methods across multiple files in scripts/sources/ (e.g., openverse.py, wikimedia.py, nasa.py).
Boundary markers: None present for the raw metadata, but the content is constrained within structured JSON and HTML templates.
Capability inventory: File writing (downloads) and browser execution via subprocess.run in scripts/well.py.
Sanitization: The format_html function in scripts/_well_utils.py mitigates injection risks by applying html.escape to titles, tags, and queries. Furthermore, a _safe_url helper validates URL protocols (restricting to http, https, or data URIs) before rendering content in the HTML preview output.

image-well