The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface where untrusted data processed by the agent can be injected into generated HTML reports without proper sanitization.
Ingestion points: Data enters the analysis pipeline via the load_data method in scripts/funnel_analyzer.py, which accepts arbitrary pandas.DataFrame objects.
Boundary markers: The export_report method in scripts/funnel_analyzer.py generates HTML reports without using delimiters or warnings to separate system instructions from user-provided data.
Capability inventory: The skill uses the Write tool to save these reports locally to the filesystem using standard file operations.
Sanitization: The export_report function in scripts/funnel_analyzer.py directly interpolates step names and insight strings into HTML templates using f-strings (e.g., <td>{row['step']}</td> and <li>{insight}</li>). This lack of escaping or sanitization creates a Cross-Site Scripting (XSS) risk if the input data contains malicious script tags.

funnel-analysis