prompt-regression-tester
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- Prompt Injection (SAFE): No instructions were found that attempt to override agent behavior, reveal system prompts, or bypass safety guidelines.
- Indirect Prompt Injection (SAFE): The skill identifies a vulnerability surface by ingesting untrusted test cases and interpolating them into prompts, but no malicious behavior was detected. 1. Ingestion points: 'test_cases' JSON object in SKILL.md. 2. Boundary markers: Absent in provided logic templates. 3. Capability inventory: LLM text generation via internal llm() calls. 4. Sanitization: None identified in the provided snippets.
- Data Exposure (SAFE): No sensitive file paths, environment variables, or hardcoded credentials were detected.
- Remote Code Execution (SAFE): No instances of external downloads (curl, wget) or dynamic code execution (eval, exec) were found.
- Persistence Mechanisms (SAFE): No attempts to modify shell profiles, cron jobs, or system services were identified.
Audit Metadata