Config schema

Topchester config is JSONC. This is the public shape most users need:

interface TopchesterConfig {
  models?: {
    "default"?: string | ModelSlotConfig;
    "fast"?: string | ModelSlotConfig;
    "kb.summarize"?: string | ModelSlotConfig;
    "choices"?: string[];
    "providers"?: Record<string, ModelProviderConfig> & { default?: string };
  };
  ignore?: {
    paths?: string[];
  };
  tools?: {
    bash?: {
      allow?: string[][];
      allowExact?: string[];
      deny?: string[][];
    };
  };
  projectInstructions?: {
    enabled?: boolean;
    files?: string[];
    fallbackFiles?: string[];
  };
  mcp?: Record<string, McpStdioServerConfig>;
  hooks?: Record<string, HookHandlerConfig[]>;
}

Arrays such as ignore.paths, tools.bash.allow, tools.bash.allowExact, tools.bash.deny, and hook handler arrays concatenate across config layers. Later scalar and object fields override or merge according to the config loader.

Model providers use OpenAI-compatible provider config:

interface ModelProviderConfig {
  type: "openai-compatible";
  baseURL: string;
  apiKeyEnv?: string;
  apiKey?: string;
  headers?: Record<string, string>;
  supportsStructuredOutputs?: boolean;
  service_tier?: "flex" | "priority";
  includeUsage?: boolean;
  promptCaching?: boolean;
  toolProtocol?: "auto" | "native" | "text-json" | "text-xml";
  openRouterToolRouting?: "auto" | "force" | "off";
}

The known codex provider is OAuth-backed. Its config stays non-secret:

{
  "providers": {
    "codex": {
      "type": "openai-compatible",
      "baseURL": "https://chatgpt.com/backend-api",
    },
  },
}

Codex OAuth tokens live in ~/.config/topchester/auth.json, not in project config.