The Agent Skills Directory

[SAFE]: The skill provides transparent auditing tools for AI model validation. The included Python scripts rely exclusively on the standard library and perform local analysis on provided data. No malicious behavior or safety guideline violations were detected during analysis.\n- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes external, potentially untrusted data such as split assignment tables and performance reports.\n
Ingestion points: Data is ingested via the --splits argument in check_split_leakage.py (CSV) and the --report or --manifest arguments in check_metric_reporting.py (Markdown/JSON).\n
Boundary markers: There are no explicit boundary markers or instructions provided to the agent to isolate these inputs from the model's instructions.\n
Capability inventory: The skill executes local Python scripts to process these inputs and can write results back to the project directory. It does not have network access or privilege escalation capabilities.\n
Sanitization: Data is processed using standard library parsers (csv, json) and analyzed through regular expression matching and set arithmetic. The logic prevents the input data from being executed as code or influencing the agent's behavior in an adversarial manner.

model-assessment