Skip to content

Security: ggghhh16/codex-code-explain

Security

SECURITY.md

Security and data handling

This is an independently developed, third-party VS Code extension, not an official OpenAI or Microsoft product.

Data sent to Codex

When you request an explanation or follow-up, the selection, limited nearby code, language-service definitions and types, and current conversation are processed through your local Codex account and service provider. Ordinary hovering only displays cached results. The extension neither stores nor displays sign-in credentials and has no separate telemetry service.

Common credential files are excluded. Common keys, Bearer tokens, JWTs, connection-string passwords, and sensitive field literals are redacted. Labels for files outside the workspace omit full local paths. Pattern matching cannot detect every secret; do not keep production credentials in source code. Installed language services are user-controlled extensions, not a security boundary.

Execution and file operations

  • Codex is launched with direct process arguments and shell: false. Automatic lookup ignores empty or relative PATH entries; a configured executable path must be absolute.
  • The extension uses a restricted named permission profile and disables inherited shell, MCP, app, and other integrations. Model instructions limit it to explaining supplied material. Codex versions or administrator configuration can affect capabilities; connection errors do not trigger a fallback to broader permissions. The extension is not an operating-system sandbox independent of Codex.
  • Model text is not trusted to create executable command links. Hover actions use a separate command allowlist. The Webview uses a content security policy and escaped rendering, without remote scripts.
  • Notes are written only after you click Save. They are appended only to ordinary .md files. Direct symlinks, hard links, directories, and files with unsaved edits are rejected. Directory permissions, concurrent processes, and file-replacement races on Windows remain subject to the host filesystem.
  • The workspace must be trusted. Do not trust unknown projects or run arbitrarily configured executables.

Repository and package

Local test configuration, logs, sign-in files, screenshots containing local paths, generated protocols, and old VSIX files are not committed. An explicit file list controls the package contents and excludes tests, caches, and user settings.

Reporting issues

Do not upload tokens, private keys, sensitive source code, or credential-bearing logs to public issues. For security issues, prefer the repository's private vulnerability-reporting feature if enabled; otherwise open a contact request without exploit details or sensitive content.

Only local Windows VS Code and the documented Codex version have been verified. This is not a third-party penetration test or formal security proof.

There aren't any published security advisories