The Agent Skills Directory

[DATA_EXFILTRATION]: The validate_model.py script includes a validate_model_api function that utilizes urllib.request to interact with remote API endpoints provided as input. This function supports an optional api_key which is sent via an Authorization: Bearer header, presenting a mechanism for network activity to non-whitelisted domains.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection during report generation.
Ingestion points: The generate_report.py script loads validation results (findings, issues, and warnings) from JSON files provided via CLI arguments.
Boundary markers: No boundary markers or instructions to ignore embedded content are used when aggregating data into the final report.
Capability inventory: The skill allows file system access (Read, Write, Edit, Grep, Glob) and command execution (Bash(python:*)).
Sanitization: The script performs direct string interpolation of findings and issues into the Markdown report without sanitization, which could allow malicious metadata from a dataset or model file to influence the agent's output context.
[EXTERNAL_DOWNLOADS]: The SKILL.md documentation identifies Fairlearn and AI Fairness 360 (AIF360) as prerequisite Python libraries for fairness assessment. While the provided scripts do not contain automated installation commands, the skill's core functionality relies on these external dependencies being present in the environment.

validating-ai-ethics-and-fairness