OpenClaw OpenAI Multi Account

Manage multiple OpenAI OAuth login accounts inside OpenClaw, including OpenAI Codex OAuth account snapshots, switching, real active-account detection, 5-hour...

This skill legitimately manipulates your OpenClaw account files (reading/writing openclaw.json and agent auth-profiles.json) to snapshot and switch accounts. The red flag is the default quota-probe URL in the script (CODEX_USAGE_URL) pointing at chatgpt.com—a non-official host—which the script may call while sending Authorization tokens. Before installing: 1) Inspect the full scripts yourself (you have them bundled); search for network calls and confirm the endpoint(s). 2) If you want to use quota probing, set OPENCLAW_CODEX_USAGE_URL to a trusted internal/mock server or remove/disable probing. 3) Run the tool in an isolated/test environment first (use a disposable OPENCLAW_HOME with test credentials). 4) Backup your OpenClaw configs (openclaw.json and agents/*/agent/auth-profiles.json) before use. 5) If you cannot verify the endpoint or author, treat this as untrusted code and avoid giving it access to real credentials.

Type: OpenClaw Skill Name: openclaw-openai-multi-account Version: 1.0.2 The skill manages OpenAI OAuth tokens by reading/writing sensitive files in `~/.openclaw/` and transmitting them to a remote endpoint (`chatgpt.com`) to monitor usage. While these actions align with the stated purpose of account switching and the script uses restricted file permissions (0o600), the inherent risk of credential handling and the ability to override the usage URL via environment variables (`OPENCLAW_CODEX_USAGE_URL`) qualify it as suspicious under the review criteria. Furthermore, the script contains a `NameError` in `cmd_import_codex` (undefined `CODEX_ACCOUNTS_DIR`) and references specific local developer paths in `SKILL.md`, indicating potential quality issues or an incomplete state.