llm-golden-dataset-builder
Pass
Audited by Gen Agent Trust Hub on Apr 23, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is designed for local development tasks such as searching, reading, and writing project files. No external network requests, remote code execution, or persistence mechanisms were detected. The skill operates within the developer's project context to organize existing logs.- [PROMPT_INJECTION]: The skill ingests untrusted data from production logs, which creates a surface for indirect prompt injection. 1. Ingestion points: Production logs and traces identified in Step 1 of SKILL.md. 2. Boundary markers: No explicit markers are used to delimit user content from instructions. 3. Capability inventory: File system search, read, and write operations (SKILL.md). 4. Sanitization: No content filtering or validation is performed. This risk is mitigated by the mandatory human-in-the-loop confirmation step (Step 4) before any data is saved to the golden dataset.
Audit Metadata