single-cell-research-planner
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONINGNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided biomedical directions to generate research plans (Step 1). While this creates an ingestion surface, the skill lacks executable capabilities such as shell access, network operations, or file system writes across all files. It implements boundary controls via a scope-redirection mechanism and strict rules against data fabrication, which mitigates the risk of the agent following malicious instructions embedded in input data.
- [METADATA_POISONING]: An evaluation report (
eval_report_single-cell-research-planner_result.json) is included within the skill package which makes various safety and performance claims. These were identified as self-authored assertions and were disregarded as authoritative conclusions during this analysis. - [NO_CODE]: The skill is strictly instruction-based and does not contain any executable scripts, binary files, or automated dependency management files.
Audit Metadata