gentle-ai-bench
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill documentation outlines standard operational procedures for a benchmarking harness. It focuses on maintaining test integrity, managing unique identifiers, and ensuring PR evidence matches execution results from local builds rather than just declaration checks.
- [INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest data from external files to determine execution parameters.
- Ingestion points:
.github/workflows/ci.ymland files within thebench/directory. - Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" warnings when reading these files.
- Capability inventory: The skill utilizes
go testand a custom CLI tool (gentle-ai-bench) to run commands derived from the ingested configuration. - Sanitization: No specific sanitization or filtering of the content from the external files is described.
- Context: This ingestion is fundamental to the skill's purpose of automating project-specific benchmarking workflows.
Audit Metadata