litreview
Pass
Audited by Gen Agent Trust Hub on Jul 18, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill ingests research question data and source metadata from user-provided JSON files and renders them into markdown reports, creating a surface for indirect prompt injection.\n
- Ingestion points: The scripts
scripts/search_strategy_builder.py,scripts/source_quality_scorer.py, andscripts/thematic_synthesis_builder.pyread input directly from external JSON files specified at runtime.\n - Boundary markers: The markdown reports generated by the scripts do not include explicit delimiters or "ignore embedded instructions" warnings to isolate processed data from the agent's instruction context.\n
- Capability inventory: The skill's scripts are restricted to the Python standard library for data processing and do not possess network, subprocess execution, or sensitive file access capabilities.\n
- Sanitization: No sanitization or escaping logic is implemented in the scripts to prevent instructions embedded within research titles, abstracts, or findings from influencing the agent's behavior.
Audit Metadata