The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied queries and interpolates them directly into shell commands used to call external scripts.
Ingestion points: User-provided search queries (e.g., "1950s street scene") in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the provided instruction set.
Capability inventory: The skill uses shell execution to run webmedia.py, freesound-fetch.py, and audio-judge.py.
Sanitization: The skill does not describe or implement sanitization for the query arguments passed to the underlying shell scripts.
[EXTERNAL_DOWNLOADS]: The skill depends on several external prerequisites that are not bundled with the skill itself, requiring the user to have previously installed specific scripts.
Evidence: References webmedia.py, freesound-fetch.py, and audio-judge.py as "external prerequisites" and "unbundled" helpers.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands to retrieve and process media.
Evidence: Shell command patterns like webmedia.py "..." --type image --download and audio-judge.py <file> "<target>" are explicitly defined in the usage section.
[DATA_EXFILTRATION]: The skill is configured to send audio media to an external cloud provider (OpenAI) for processing and evaluation.
Evidence: The audio-judge.py script is described as sending clips to gpt-audio to return analysis JSON.

web-media-getter