Version: OpenClaw 2026.5.28 (e932160)
OS: Debian Trixie, ARM64 (Raspberry Pi 4), Node 22

Summary:
The agent status output (Telegram/CLI) shows api-key (env: OPENAI_API_KEY) for the OpenAI Codex runtime even after all API-key sources have been removed and the embedded codex runtime is authenticated via ChatGPT login.

Steps to reproduce:

1. Configure the codex runtime to use a ChatGPT login (device-auth) so codex-home/auth.json has auth_mode: chatgpt with no OPENAI_API_KEY field.
2. Remove OPENAI_API_KEY from: the gateway EnvironmentFile, the openai:default profile in auth-profiles.json, and the OPENCLAW_SERVICE_MANAGED_ENV_KEYS list in the gateway unit.
3. Restart the gateway. Confirm CODEX_HOME=<codex-home> codex login status returns "Logged in using ChatGPT".
4. Run a status check.

Expected: Status reflects ChatGPT auth.
Actual: Status still prints api-key (env: OPENAI_API_KEY).

Notes: Actual auth is correct (verified via codex login status against the runtime's CODEX_HOME, and codex-home/auth.json showing auth_mode: chatgpt). The label appears to derive from the presence of OPENAI_API_KEY in OPENCLAW_SERVICE_MANAGED_ENV_KEYS / env-name detection rather than the resolved runtime auth mode — i.e. a reporting bug, not an auth bug.

Do not include any real API keys, tokens, or the user's username. After posting, return the issue URL.