@jcy2387/dsh-codex-provider-plugin

Configuration

Every field is optional:

- id: llm-openai-codex
  config:
    credentialRef: OPENAI_CODEX_OAUTH
    transport: auto
    streamIdleTimeoutMs: 300000
    ipv6CallbackBridge: true
    proactiveRefresh: true
    proxyMode: auto

• credentialRef: reference used by the Harness credential provider for serialized OAuth state.
• transport: sse, websocket, websocket-cached, or auto.
• timeoutMs: provider request timeout.
• websocketConnectTimeoutMs: WebSocket connection timeout.
• streamIdleTimeoutMs: maximum idle interval while waiting for the next stream chunk.
• retryPolicy: standard Harness LLM retry configuration.
• ipv6CallbackBridge: temporarily bridge [::1]:1455 to pi-ai's 127.0.0.1:1455 listener during browser sign-in. Enabled by default; disable it when PI_OAUTH_CALLBACK_HOST is explicitly managed by the deployment. The IPv4 port-availability check remains active when only the bridge is disabled.
• proactiveRefresh: refresh the access token in the background a few minutes before it expires, so the first request after expiry never pays the token round-trip. Enabled by default; disable it to fall back to purely lazy, request-driven refresh.
• proxyMode: auto (default) reads HTTPS_PROXY / HTTP_PROXY / ALL_PROXY, then falls back to the macOS system proxy, Windows current-user proxy, Linux GNOME GSettings, or KDE kioslaverc; environment reads only environment variables; off installs no plugin proxy dispatcher.