pseo-scale
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill architects a system that ingests external data from databases, CMSs, and APIs to provide context for automated content enrichment via LLMs.
- Ingestion points: External data sources identified in the 'Content Enrichment Pipeline' section (SKILL.md) ingest raw data for processing.
- Boundary markers: The architecture focuses on grounding prompts in real data attributes but does not explicitly define delimiters or boundary markers to prevent embedded instructions in the source data from affecting the LLM generation.
- Capability inventory: The pipeline is designed to generate introductions, FAQs, and content summaries automatically based on the ingested data.
- Sanitization: Mitigation strategies are suggested, including a 5-10% human review sampling per batch and the application of an external quality guard tool ('pseo-quality-guard') prior to publishing.
Audit Metadata