The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill executes a local Python script using the --sql flag, which accepts arbitrary SQL strings. This allows for the execution of complex DuckDB logic that can interact with the host filesystem.
[DATA_EXFILTRATION] (HIGH): DuckDB's built-in functions such as read_csv_auto(), read_parquet(), or glob() can be used within the --sql parameter to access sensitive system files (e.g., /etc/passwd) if the process has sufficient permissions.
[PROMPT_INJECTION] (HIGH): Category 8: Indirect Prompt Injection. 1. Ingestion points: Untrusted Excel and CSV files uploaded to /mnt/user-data/uploads/. 2. Boundary markers: Absent; there are no delimiters or instructions to ignore embedded commands in the processed data. 3. Capability inventory: Ability to execute arbitrary SQL, read any file via DuckDB, and write output files to arbitrary paths via the --output-file parameter. 4. Sanitization: Absent; the skill is designed to support arbitrary SQL, making it highly susceptible to malicious payloads embedded in data files that could influence the generated SQL queries.
[DATA_EXFILTRATION] (MEDIUM): The --output-file parameter allows the agent to write query results to potentially sensitive locations on the filesystem, which could lead to file overwriting or unauthorized data placement.

data-analysis