Skip to content

SECURITY.md

Latest commit

 

History

History
28 lines (18 loc) · 1.65 KB

SECURITY.md

File metadata and controls

28 lines (18 loc) · 1.65 KB

Security

Trust Model

This plugin runs inside your AI coding agent (Claude Code, Codex, Cursor, etc.) with the same permissions as the agent itself. By installing it, you trust:

  1. CopilotKit/skills repository — The SKILL.md files and reference docs are loaded into the agent's context and influence its behavior. We maintain this repo and review all changes.

  2. MCP server at mcp.copilotkit.ai — Skills can query this server for live documentation via search-docs and search-code tools. We host and control this server. It indexes the CopilotKit and AG-UI public repositories.

  3. Auto-approval hooks — The plugin includes a PreToolUse hook (hooks/auto-approve-copilotkit.sh) that auto-approves specific read-only shell commands without user confirmation:

    • npx copilotkit --help/--version/info/doctor
    • pnpm list @copilotkit/* / npm ls @copilotkit/*
    • nx run @copilotkit/*:test / vitest run (test execution)
    • nx run @copilotkit/*:build (build verification)

    The hook never auto-approves: install, add, remove, delete, publish, deploy, push, or any other destructive operation. Unrecognized commands are deferred to the user for manual approval.

What This Plugin Cannot Do

  • It cannot execute code outside of what the agent already has permission to do
  • It cannot access files or networks beyond the agent's sandbox
  • It cannot auto-approve destructive operations
  • The hooks only fire for Bash commands matching specific patterns

Reporting Vulnerabilities

If you find a security issue, please report it to security@copilotkit.ai or open a GitHub issue at https://github.com/CopilotKit/skills/issues.