The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation includes integration guides that suggest running local Python scripts, such as scripts/import_apple_health.py, to process user data. While metadata limits the agent's tools, these instructions promote the use of shell commands for script execution.
[DATA_EXFILTRATION]: The skill processes highly sensitive health data and handles API tokens for fitness platforms. Although the instructions state that data is stored locally, it involves network interactions with vendor infrastructure and well-known third-party APIs.
[EXTERNAL_DOWNLOADS]: The skill references multiple external knowledge base articles on the vendor domain and interacts with official fitness tracker APIs to retrieve user health data.
[PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted external files (XML, ZIP, CSV, JSON) provided by users. This exposes a surface for indirect prompt injection where malicious instructions embedded in the health data could influence the agent's behavior. Ingestion points include health exports and mapping files. The skill lacks explicit boundary markers and sanitization for instruction prevention, although it performs basic data validation.

wellally-tech