analyze-results
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an entry point for untrusted external data processing, creating a potential indirect prompt injection surface.
- Ingestion points: Locates and parses JSON and CSV experimental result files within the project directories (
figures/,results/, etc.). - Boundary markers: Absent; there are no clear isolation techniques or explicit delimiter instructions to distinguish the metric data from potential natural language instructions embedded within the files.
- Capability inventory: Possesses wide-ranging capabilities through allowed tools like
Bash(*),Read,Grep,Glob,Write, andEdit. - Sanitization: There is no mention of sanitizing or validating fields in JSON/CSV files prior to processing and generating analysis insights.
Audit Metadata