paper-spine-audit
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-agent review flow through the
structured_review.pyscript that creates an indirect prompt injection surface by interpolating untrusted manuscript text into prompts for sub-agents. - Ingestion points: Manuscript content is read from user-supplied files (e.g., LaTeX, Markdown, or Word documents) and partitioned into sections for review.
- Boundary markers: The generated prompts for the reviewer sub-agents do not utilize specific delimiters or clear instructions to ignore instructions that might be embedded within the manuscript text.
- Capability inventory: The skill orchestrates the creation of three independent reviewer agents, providing them with the manuscript text to perform focused audits.
- Sanitization: Although the scripts perform text normalization and LaTeX command stripping, they do not include mechanisms to sanitize or filter Natural Language instructions that could influence the sub-agents' behavior.
Audit Metadata