llm-wiki-ingestion-stack
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a robust architecture for ingesting untrusted external data from various sources (PDFs, web clips, code repos). It explicitly mitigates indirect prompt injection risks by mandating structured manifests, boundary markers, and human review gates before content is indexed for retrieval.
- [EXTERNAL_DOWNLOADS]: The documentation references a wide variety of third-party tools for document conversion and extraction, such as Docling, MarkItDown, Playwright, and Whisper. These are presented as architectural recommendations, and the skill includes instructions to verify official documentation and security notes before implementation.
- [DATA_EXFILTRATION]: The architecture incorporates specific safety gates for PII/sensitive data redaction and secret scanning (using tools like gitleaks and trufflehog) to prevent the accidental ingestion or exposure of credentials during the ETL process.
Audit Metadata