The Agent Skills Directory

Data Ingestion Surface: The script scripts/charts.py reads data from external files (CSV, TSV, JSON). While necessary for charting, processing untrusted data files presents a surface for indirect prompt injection if the agent interprets data content as instructions during processing.
Ingestion points: The load_data function in scripts/charts.py accepts file paths to read data into a DataFrame.
Boundary markers: The skill does not provide explicit delimiters or instructions to the agent to treat data content as strictly non-instructional.
Capability inventory: The skill possesses the ability to read local files, write images to the filesystem, and execute Python scripts via the CLI.
Sanitization: The script performs standard data cleaning (dropping missing values) but does not include sanitization of text content to prevent instruction-based attacks.
External Data Fetching: The load_data function utilizes pandas.read_csv and pandas.read_json, which are capable of fetching data from remote URLs. This is a standard feature of the pandas library but allows the skill to perform network requests if a URL is provided as a data source.
Command Execution: The skill is designed to be invoked through a CLI interface (python scripts/charts.py), which is the standard mechanism for the agent to utilize the bundled Python toolkit.
File System Interaction: The skill reads input data files and writes generated chart images (PNG) to the local filesystem. This behavior is transparently documented and consistent with the skill's purpose as a chart builder.

chart-builder