The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied financial data from a JSON file and includes identifiers (such as opportunity IDs and cohort IDs) directly in its generated reports. This creates an attack surface where an attacker could embed malicious instructions in the input data to influence the AI agent's behavior when it reviews the forecast output.
Ingestion points: Data is loaded from intake.json by scripts/bookings_forecaster.py, scripts/cohort_arr_projector.py, and scripts/funnel_confidence_scorer.py.
Boundary markers: The skill lacks explicit boundary markers or instructions to the AI to ignore content within the generated reports.
Capability inventory: The Python scripts are restricted to the standard library and do not have network access or file system write permissions.
Sanitization: There is no sanitization or escaping of the string values imported from the JSON file before they are interpolated into the Markdown output.

commercial-forecaster