feat: per-model max output tokens and a conversation-time thinking level backed by agent settings (#28)

Co-authored-by: Alice <alice@prismshadow.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

packages/cli/src/commands/config.ts

@@ -2,7 +2,7 @@
  * `penguin config` — manages a Project's model credentials, default model, model list,
  * Agent-level vault environment variables, and UI language.
  *
- *   penguin config model add --model-id <upstream id> --provider <group> [--api-key <key>] [--context-window <n>] [--set-default] [--root <dir>]
+ *   penguin config model add --model-id <upstream id> --provider <group> [--api-key <key>] [--context-window <n>] [--max-tokens <n>] [--set-default] [--root <dir>]
  *   penguin config model default --model-id <upstream id> --provider <group> [--root <dir>]
  *   penguin config model vision --model-id <upstream id> --provider <group> [--root <dir>]
  *   penguin config model list [--root <dir>]
@@ -120,6 +120,7 @@ export function registerConfigCommand(program: Command, t: Messages): void {
     .option("--api-key <key>", t.config.addApiKey)
     .option("--base-url <url>", t.config.addBaseUrl)
     .option("--context-window <n>", t.config.addContextWindow, parseIntArg)
+    .option("--max-tokens <n>", t.config.addMaxTokens, parseIntArg)
     .option("--client-type <type>", t.config.addClientType)
     // Tri-state: --vision marks it supported / --no-vision marks it unsupported / neither given keeps the existing value (defaults to supported).
     .option("--vision", t.config.addVision)
@@ -131,6 +132,15 @@ export function registerConfigCommand(program: Command, t: Messages): void {
     .option("--set-default", t.config.addSetDefault, false)
     .option("--root <dir>", t.common.root)
     .action(async (opts) => {
+      // Output cap: parseIntArg already rejects non-numbers; 0/negative must not reach the config either.
+      const maxTokens: number | undefined = opts.maxTokens;
+      if (maxTokens !== undefined && maxTokens <= 0) {
+        process.stderr.write(
+          `${t.error(`--max-tokens must be a positive integer: got "${maxTokens}".`)}\n`,
+        );
+        process.exitCode = 1;
+        return;
+      }
       const root = resolveRootOption(opts.root);
       // --model-id takes the upstream id, paired with the required --provider as a
       // reference; the group is never guessed, so --api-key can only ever land on the
@@ -164,6 +174,7 @@ export function registerConfigCommand(program: Command, t: Messages): void {
           provider,
           model_id: modelId,
           ...(opts.contextWindow !== undefined ? { context_window: opts.contextWindow } : {}),
+          ...(maxTokens !== undefined ? { max_tokens: maxTokens } : {}),
           ...(clientType !== undefined ? { client_type: clientType } : {}),
           ...(opts.vision !== undefined ? { vision: opts.vision } : {}),
           ...(Object.keys(pricing).length > 0 ? { pricing } : {}),

packages/cli/src/i18n.ts

@@ -39,6 +39,7 @@ export interface Messages {
     addApiKey: string;
     addBaseUrl: string;
     addContextWindow: string;
+    addMaxTokens: string;
     addClientType: string;
     addVision: string;
     addNoVision: string;
@@ -174,6 +175,8 @@ const en: Messages = {
     addApiKey: "API key, stored inline in the Project's hidden .project_config.toml",
     addBaseUrl: "Custom base URL",
     addContextWindow: "Context window size (tokens)",
+    addMaxTokens:
+      "Per-model max output tokens (positive integer); when set it overrides the Agent's max_tokens, omit to inherit — lower it for small-context models",
     addClientType: "AgentHub client type (e.g. openai); defaults by provider group when omitted",
     addVision: "Mark the model as supporting image input (vision)",
     addNoVision: "Mark the model as NOT supporting image input; omit both to keep current",
@@ -287,6 +290,8 @@ const zh: Messages = {
     addApiKey: "API key，内联存入 Project 的隐藏文件 .project_config.toml",
     addBaseUrl: "自定义 base url",
     addContextWindow: "上下文窗口大小（token 数）",
+    addMaxTokens:
+      "该模型的最大输出长度（正整数）；设置后覆盖 Agent 的 max_tokens，缺省沿用——小上下文模型建议调低",
     addClientType: "AgentHub 客户端协议（如 openai）；缺省按 provider 分组的语义取值",
     addVision: "标注该模型支持图片输入（视觉）",
     addNoVision: "标注该模型不支持图片输入；两者都不给则保留原值",