Configuration

All keys live under the plugin's settings namespace. With openai-codex scoped globally, you can set defaults directly:

openai-codex:
  # Reasoning effort applied when a request omits one.
  reasoningEffort: high   # off | low | medium | high | max
  # Default output budget.
  maxTokens: 128000
  # Advisory context window for models without an exact catalog value.
  defaultContextWindow: 400000
  # Text verbosity.
  verbosity: low          # low | medium | high
  # Refresh the access token this far before expiry.
  refreshBeforeMs: 300000
  # Drop an in-flight request after this much provider idle time.
  streamIdleTimeoutMs: 30000
  # Strip model-emitted redundant `sandbox_permissions: "workspace-write"` from
  # tool calls before the harness sees them (kept when a sandbox denial in the
  # conversation makes it a legitimate retry; genuine escalations preserved).
  sanitizeSandboxPermissions: true
  # Override the codex Responses endpoint (advanced/debugging).
  baseURL: https://chatgpt.com/backend-api/codex/responses

Unset optional keys (reasoningEffort, verbosity, retryPolicy) fall back to provider defaults.

dsh-openai-codex

同名套件的其他儲存庫

安裝

說明文件

Configuration