skillopt-runner
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches the SkillOpt repository from Microsoft's official GitHub organization (github.com/microsoft/SkillOpt).
- [COMMAND_EXECUTION]: Executes shell commands for environment setup, patching, data generation, and benchmark rollouts. This includes modifying the PATH variable to prioritize a local virtual environment and configuring environment variables such as CODEX_HOME for tool authentication.
- [INDIRECT_PROMPT_INJECTION]: The skill processes synthetic benchmark data through an LLM optimization loop, creating a surface for indirect prompt injection where data could influence agent actions.
- Ingestion points: Synthetic split data (items.json) and benchmark item files (SKILL.md).
- Boundary markers: No explicit delimiters or instructions to ignore embedded commands were identified in the workflow.
- Capability inventory: The skill can execute complex commands and manage processes via the codex and pi CLI backends.
- Sanitization: No input validation or sanitization logic is described for the items being benchmarked.
Audit Metadata