research-systematic-literature-review
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external research papers and metadata from sources such as OpenReview, PubMed, and DOI resolvers. While this represents a surface for indirect prompt injection, the skill incorporates structured synthesis phases and an 'Adversarial Stress Test' (red-teaming) that mitigate the risk of the agent following instructions potentially embedded in the source documents.
- Ingestion points: External research papers and preprints retrieved via web browsing tools.
- Boundary markers: Use of structured Markdown templates and standardized evidence tables.
- Capability inventory: File system write access and execution of local Python scripts.
- Sanitization: Reasoning-based synthesis and methodological critique protocols are used to filter and validate evidence.
- [COMMAND_EXECUTION]: The skill utilizes several internal Python scripts (
init_review_pack.py,prisma_flow_md.py,validate_review_pack.py,paper_context_artifacts.py) to manage the review lifecycle. These scripts perform deterministic operations on Markdown files and do not invoke arbitrary shell commands or external subprocesses. - [EXTERNAL_DOWNLOADS]: The skill explicitly directs the agent to retrieve literature from established academic platforms and DOI resolvers. These are well-known services within the research domain, and the skill implements deduplication and version resolution to prioritize canonical published versions over preprints.
Audit Metadata