The Agent Skills Directory

[Data Processing and Indirect Prompt Injection]: The skill processes user-supplied CSV files to generate its dashboard. Ingesting external data introduces a potential surface for indirect prompt injection.
Ingestion points: Four CSV files located in input/**/*.csv, representing Microsoft 365 admin center activity exports.
Boundary markers: The agent is instructed to map specific headers and use a dedicated script for processing, which limits the scope of ingestion.
Capability inventory: The build_dashboard.py script performs file reading, data aggregation, and HTML generation, with no network or arbitrary code execution capabilities.
Sanitization: The script utilizes html_escape for text and json.dumps for structured data before injection into the HTML template. Additionally, the template includes a runtime helper function to escape strings during rendering, reducing the risk of malicious scripts being executed.
[Local Script Execution]: The skill executes a bundled Python script (scripts/build_dashboard.py) to perform the data analysis and report generation. This is a standard functional pattern for this utility. The script is restricted to the Python standard library and explicitly avoids network operations, minimizing the risk of unauthorized data transmission or external dependency risks.

create-copilot-adoption-dashboard