inkos-long-story-analysis

Pass

Audited by Gen Agent Trust Hub on Aug 16, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted content (novels, PDFs, URLs) using the ingest_material tool. This is a standard attack surface where text from a novel could contain instructions intended to mislead the agent.
  • Ingestion points: The skill accepts content from files, PDFs, URLs, and uploads via ingest_material.
  • Boundary markers: The instructions explicitly require the agent to anchor claims to chapter excerpts and separate observation from inference, which provides a natural structure for identifying source material versus instructions.
  • Capability inventory: The skill is restricted to analyzing content and managing book-related metadata (manage_book_reference); it does not have access to shell execution or non-whitelisted network domains.
  • Sanitization: No explicit content sanitization is described, but the analytical framework provided in references/analysis-lens.md promotes objective deconstruction over instruction following.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 16, 2026, 10:11 AM
Security Audit — agent-trust-hub — inkos-long-story-analysis