The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses an HTML template (assets/explanation_template.html) to present model explanations. This template interpolates potentially untrusted data such as model descriptions, input data, and prediction explanations into a format that could be rendered by a user.
Ingestion points: The skill processes user-provided models and datasets (e.g., referenced in assets/example_explanation.json as customer_churn_dataset.csv) to generate explanations based on user requests.
Boundary markers: The provided templates and instructions lack explicit delimiters or "ignore embedded instructions" warnings to prevent the AI from following malicious instructions hidden within dataset values or model metadata.
Capability inventory: The skill is granted Bash(cmd:*), Write, and Edit permissions, which allows it to generate, save, and potentially execute code or reports based on processed data.
Sanitization: There is no evidence of sanitization, HTML escaping, or validation logic in the provided assets to handle untrusted input before it is interpolated into the final HTML output.

explaining-machine-learning-models