copilot-harness
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides detailed instructions for executing the
copilotCLI andwingetinstaller to manage the GitHub Copilot environment on Windows. - [EXTERNAL_DOWNLOADS]: The skill documents the installation of the
GitHub.Copilotpackage using thewingetpackage manager, which fetches software from GitHub's official distribution channels. - [INDIRECT_PROMPT_INJECTION]: The skill defines a command-line interface for passing natural language prompts to the Copilot CLI, which represents a potential injection surface.
- Ingestion points: The
-por--promptflag in thecopilotcommand accepts arbitrary text input from the agent or user (SKILL.md). - Boundary markers: There are no specific delimiters or instruction-isolation markers described to separate data from instructions within the prompt.
- Capability inventory: The
copilotCLI is capable of file system writes (e.g., plan mode, tool-using prompts), executing tools via theskilltool, and performing network operations, especially when high-privilege flags like--allow-all-toolsor--yoloare used (SKILL.md). - Sanitization: The skill does not describe any sanitization or validation of the input before it is passed to the underlying model for execution.
Audit Metadata