The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the agent with the ability to execute SQL commands through the sqlite3 CLI and a Python wrapper script (scripts/health_query.py). This allows for direct interaction with local database files.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests data from a local SQLite database (~/data/health.db) which contains millions of records from Apple Health.
Boundary markers: No specific delimiters or instructions are provided to the agent to mitigate risks from potentially malicious data stored within the database records.
Capability inventory: The skill provides the agent with capabilities for arbitrary SQL execution and file system access to the database.
Sanitization: The run_query function in scripts/health_query.py executes raw SQL strings without using parameterization or validation, creating a vulnerability surface for SQL injection.
[DYNAMIC_EXECUTION]: The Python utility scripts/health_query.py implements dynamic execution of SQL code via its query subcommand. This allows the agent to construct and run SQL statements at runtime, which is a high-capability feature requiring careful input handling.

health-data