The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads and processes video metadata sidecars (.metadata.json) that may be generated by external tools like video-downloader. This data is formatted and uploaded to the Immich server as asset descriptions, creating a surface for indirect injection if the source content is malicious.
Ingestion points: scripts/immich/metadata.py reads data from JSON sidecar files.
Boundary markers: No delimiters or explicit warnings are provided to the agent or the server API to distinguish untrusted content.
Capability inventory: The skill uses scripts/immich/client.py to perform network operations (POST/PATCH) to a remote API.
Sanitization: No sanitization or filtering is applied to the metadata fields before they are uploaded.
[COMMAND_EXECUTION]: The skill utilizes the uv run tool to execute its internal Python logic, which is the expected execution pattern for this environment.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to a user-configured Immich server address using the httpx library to upload media and manage assets.

immich