prompt-engineering
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- [Prompt Injection] (SAFE): No malicious instructions, jailbreak attempts, or safety filter bypasses were detected. The content is purely instructional.
- [Data Exposure & Exfiltration] (SAFE): No hardcoded credentials, sensitive file paths, or network exfiltration patterns are present.
- [Indirect Prompt Injection] (LOW): The skill demonstrates processing external data such as code diffs. 1. Ingestion points: 'diff' and 'context' parameters in the Python prompt template (SKILL.md). 2. Boundary markers: The template uses markdown code blocks (triple backticks) to delimit untrusted input. 3. Capability inventory: The prompt instructs the model to analyze code and generate JSON findings. 4. Sanitization: No explicit sanitization or input validation is shown in the documentation snippet. This represents a known surface for indirect injection but is aligned with the primary purpose of the skill.
- [Unverifiable Dependencies & Remote Code Execution] (SAFE): No external packages are installed and no remote scripts are downloaded or executed.
Audit Metadata