Skip to content

Security: Infiland/cursor-opencode-auth

Security

docs/SECURITY.md

Security Notes

This project intentionally uses only documented Cursor surfaces:

  • Cursor CLI (agent)
  • Cursor Cloud Agents API (https://api.cursor.com/v0/...)

It does not reverse-engineer private Cursor endpoints.

Cursor CLI risks

  • Cursor CLI can read your repository and run shell commands depending on your Cursor CLI permissions.
  • In print mode, --force enables file edits without confirmation.

Mitigation:

  • Configure Cursor CLI permissions (~/.cursor/cli-config.json or <project>/.cursor/cli.json) to allow only what you need.
  • Deny reads of sensitive files like .env*.

OpenCode plugin tool risks

  • OpenCode custom tools run on your machine.
  • Even if OpenCode permissions are strict, a buggy/malicious plugin could still execute code.

Mitigation:

  • Only install plugins you trust.
  • Gate tool usage in OpenCode with permission rules (require ask).

cursor_cli_patch behavior

cursor_cli_patch:

  • Creates a temporary git worktree directory.
  • Runs Cursor CLI inside it with --force.
  • Produces a git diff patch.
  • Removes the temp worktree (may use git worktree remove --force on the temp directory).

This is designed to prevent Cursor from editing your main working tree directly.

Cloud Agents risks

Cloud Agents run in a remote Ubuntu environment with internet access and auto-run terminal commands.

Mitigation:

  • Only use Cloud Agents on repos where remote execution is acceptable.
  • Avoid passing secrets in prompts.
  • Prefer using Cursor’s Cloud Agent Secrets management (Cursor Settings → Cloud Agents → Secrets) instead of committing .env files.

There aren’t any published security advisories