The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to generate and execute a Python script named data_cleaning.py. This script is used to process raw data and output cleaned results to a project directory.
[COMMAND_EXECUTION]: The skill requires all tool calls to be executed via run_sdk_snippet by passing code to the python_code parameter, establishing a pattern of arbitrary code execution for standard operations.
[DATA_EXFILTRATION]: The provided template for data_cleaning.py explicitly uses parent directory referencing (..) to access data sources. This pattern could be exploited to read sensitive files located outside the intended project directory within the workspace environment.
[REMOTE_CODE_EXECUTION]: The skill processes untrusted external data (CSV, JSON, Excel, Web-sourced data) and uses it to dynamically generate logic for dashboard cards (getCardData functions) and data cleaning scripts. This constitutes a risk of injecting malicious logic into the generated code.
[PROMPT_INJECTION]: The skill includes an 'Indirect Prompt Injection' surface where malicious instructions could be embedded in the processed data files.
Ingestion points: Reads from FILE_DATA_SOURCES, user-uploaded files, and web-sourced data.
Boundary markers: None identified in the instructions for separating data from instructions during processing.
Capability inventory: Subprocess execution via run_sdk_snippet, file system write access for project creation and data cleaning, and JS code generation for dashboard cards.
Sanitization: The skill instructions focus on data quality (missing values, types) rather than security sanitization to prevent instruction injection into generated scripts.

develop-data-analysis-dashboard