analyze-variants
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill processes external data from process mining tools, which represents an indirect prompt injection surface. The agent is instructed to interpret metrics and traces, potentially exposing it to instructions embedded within the processed data. This is a common characteristic of data-processing skills.
- Ingestion points: Data enters the agent context through the outputs of get_variants_with_metrics_v2, get_cases_with_metrics_v2, and get_edges_with_metrics_v2 (in SKILL.md).
- Boundary markers: The skill provides structural guardrails, such as using shared contracts and an 'investigation method' to distinguish observed presence from tested outcomes, which helps prevent misinterpretation of data.
- Capability inventory: The skill is constrained to a minimal toolset (Read, AskUserQuestion), lacking high-risk capabilities like shell execution, network access, or file modification.
- Sanitization: The instructions require the agent to validate business outcome definitions and available attributes rather than inferring success from event names, providing a logical layer of validation for external input.
Audit Metadata