The Agent Skills Directory

[COMMAND_EXECUTION]: Uses shell commands (mkdir -p) to manage output directories based on dataset names. This pattern relies on the agent to ensure dataset names resolved from user input do not contain shell metacharacters.
[COMMAND_EXECUTION]: Executes a local Python script to process files. The script uses standard pandas methods for reading CSV, Excel, and JSON data and printing summary statistics.
[PROMPT_INJECTION]: The skill ingests untrusted data from external files, which constitutes an indirect prompt injection surface. Malicious content within a dataset could potentially influence the agent when it reads the resulting summary.
Ingestion points: Local files in input/ processed by the Python script.
Boundary markers: Output is formatted in markdown tables, but no explicit 'ignore instructions' warnings are provided around interpolated data.
Capability inventory: Access to Read, Write, Bash, Glob, and Grep tools.
Sanitization: No content-level validation or sanitization is performed on the data being profiled.
[SAFE]: The core functionality uses well-known data science libraries (pandas, openpyxl) for intended analytical purposes with no network exfiltration or credential access.

profile