skills/involvex/skills/copilot-sdk/Gen Agent Trust Hub

copilot-sdk

Warn

Audited by Gen Agent Trust Hub on Apr 8, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends the installation of several packages from public registries, such as github-copilot-sdk on PyPI and GitHub.Copilot.SDK on NuGet. These names suggest an official GitHub origin, but they are not established standard libraries, and their provenance cannot be confirmed from the skill content alone.
  • [COMMAND_EXECUTION]: The instructions guide users to execute commands to initialize projects and run the Copilot CLI in server mode (copilot --server --port 4321), which the SDK then connects to via local networking.
  • [PROMPT_INJECTION]: The interactive CLI assistant examples for TypeScript and Python demonstrate a surface for indirect prompt injection by passing raw user input directly to the model session.
  • Ingestion points: User input is collected from stdin via rl.question and input() in the 'Interactive CLI Assistant' section of SKILL.md.
  • Boundary markers: None; the code examples do not implement delimiters or ignore-instructions markers to isolate user input.
  • Capability inventory: The SDK provides capabilities for defining custom tools with executable handlers and connecting to MCP servers with potential repository access.
  • Sanitization: No sanitization or validation of the user-provided input is performed before it is sent to the model.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 8, 2026, 09:27 PM
Security Audit — agent-trust-hub — copilot-sdk