ai-prompt-regression-testing
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions promote security-conscious behaviors, including the use of synthetic or redacted data and the avoidance of hardcoding credentials or private prompts in shared registers.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external evaluation runs (test results and judge outputs).
- Ingestion points: The
summarize_regressionfunction inSKILL.mdaccepts aresultslist containing data from external model evaluations. - Boundary markers: The workflow instructions recommend using "deterministic schema/field checks" and "predeclared rules" to validate outputs rather than relying solely on LLM judges.
- Capability inventory: The provided Python script is restricted to local data processing. It does not perform network operations, file system writes, or subprocess execution.
- Sanitization: The script implements robust input validation, checking data types and enforcing an allowed enum of result statuses ('pass', 'fail', 'incomplete').
- [DYNAMIC_EXECUTION]: The skill includes a Python code snippet intended for result aggregation. This code is static, purely logical, and does not involve runtime compilation or execution of untrusted external code.
Audit Metadata