@agent-native/core 0.83.0 → 0.84.1

package/corpus/core/docs/content/locales/pt-BR/cloneable-saas.mdx

@@ -30,9 +30,9 @@ Cada um é um aplicativo real que você pode usar hoje e a plataforma de lançam
 | [**Plan**](/docs/template-plan)           | Planos visuais e recapitulações de relações públicas com diagramas, wireframes e anotações.                                                  |
 | [**Dispatch**](/docs/template-dispatch)   | O plano de controle do espaço de trabalho: segredos compartilhados, integrações reutilizáveis, Slack/Telegram, trabalhos agendados.          |
 
-Não quer um modelo de domínio? Use [Chat](/docs/template-chat) quando quiser um aplicativo básico com o qual os usuários possam conversar imediatamente ou comece a agir primeiro com [Pure-Agent Apps](/docs/pure-agent-apps).
+Não quer um app de domínio? Use [Chat](/docs/template-chat) quando quiser um aplicativo básico com o qual os usuários possam conversar imediatamente, ou comece com uma ação em [Pure-Agent Apps](/docs/pure-agent-apps).
 
-Veja o catálogo completo em [Templates](/templates) ou vá direto para um. Por exemplo, [Dispatch](/docs/template-dispatch) é um ótimo lugar para começar se você quiser um aplicativo estilo espaço de trabalho.
+Veja o catálogo completo em [Apps](/apps) ou vá direto para um. Por exemplo, [Dispatch](/docs/template-dispatch) é um ótimo lugar para começar se você quiser um aplicativo estilo espaço de trabalho.
 
 ## O que você ganha imediatamente {#what-you-get}
 

package/corpus/core/docs/content/locales/pt-BR/evals.mdx

@@ -231,6 +231,14 @@ Códigos de saída:
 | `1`    | Pelo menos uma avaliação com pontuação abaixo do limite ou o conjunto apresentou erro.                     |
 | `2`    | Argumentos incorretos (por exemplo, `--threshold` fora de `[0, 1]`).                                       |
 
+### Ignorar evals gated {#skipping}
+
+Use `skipReason` quando uma suíte de evals for intencionalmente reservada para
+execução manual, noturna ou dependente de secrets. Evals ignorados não executam
+o agente nem os scorers, não fazem a suíte falhar e aparecem no JSON com
+`status: "skipped"` e `skipReason`. Mantenha cobertura determinística barata no
+CI normal de PR.
+
 ### Como porta de implantação de CI {#ci}
 
 Adicione-o ao pipeline executado antes da implantação:

package/corpus/core/docs/content/locales/pt-BR/faq.mdx

@@ -116,7 +116,7 @@ Não. Você não treina modelos, não ajusta ou lida com incorporações. Você
 
 ### Posso migrar um aplicativo existente para agente nativo? {#can-i-use-existing-code}
 
-Você pode, mas o agente nativo funciona melhor quando construído do zero. A arquitetura – banco de dados compartilhado, sincronização de polling, actions, estado do aplicativo – precisa ser totalmente integrada. Começar a partir de um modelo e personalizá-lo é o caminho recomendado. Pense nisso como a mudança do desktop para o mobile: você _pode_ fazer o retrofit, mas construir nativo é melhor.
+Você pode, mas o agente nativo funciona melhor quando construído do zero. A arquitetura – banco de dados compartilhado, sincronização de polling, actions, estado do aplicativo – precisa ser totalmente integrada. Começar a partir de um app e personalizá-lo é o caminho recomendado. Pense nisso como a mudança do desktop para o mobile: você _pode_ fazer o retrofit, mas construir nativo é melhor.
 
 ## Modelos e o que você pode construir {#templates}
 

package/corpus/core/docs/content/locales/pt-BR/template-chat.mdx

@@ -13,7 +13,7 @@ Se você deseja o menor tempo de execução somente de ação sem navegador UI,
   <Screen
     surface="desktop"
     html={
-      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>Como posso ajudar?</h1><p class='wf-muted' style='margin:10px 0 0'>Converse sobre qualquer coisa. Adicione ações, componentes, páginas, jobs ou seu próprio backend.</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>Envie mensagem ao agente...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 4.6 · Auto</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
+      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>Como posso ajudar?</h1><p class='wf-muted' style='margin:10px 0 0'>Converse sobre qualquer coisa. Adicione ações, componentes, páginas, jobs ou seu próprio backend.</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>Envie mensagem ao agente...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 5 · Padrão</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
     }
   />
 </WireframeBlock>

package/corpus/core/docs/content/locales/zh-CN/cloneable-saas.mdx

@@ -30,9 +30,9 @@ description: "分叉一个可用的 SaaS 产品并使其成为您的产品 - 包
 | [**Plan**](/docs/template-plan)           | 带有图表、线框图和注释的视觉计划和公关概述。                                                        |
 | [**Dispatch**](/docs/template-dispatch)   | 工作区控制平面：共享秘密、可重用集成、Slack/Telegram、计划作业。                                    |
 
-不需要域模板？当您想要一个用户可以立即交谈的基本应用程序时，请使用 [Chat](/docs/template-chat)，或者使用 [Pure-Agent Apps](/docs/pure-agent-apps) 首先开始操作。
+不需要域应用？当您想要一个用户可以立即交谈的基本应用程序时，请使用 [Chat](/docs/template-chat)，或者从 [Pure-Agent Apps](/docs/pure-agent-apps) 中的一个操作开始。
 
-查看 [Templates](/templates) 下的完整目录，或直接跳到其中一个 - 例如，如果您想要一款工作区风格的应用，[Dispatch](/docs/template-dispatch) 是一个很好的起点。
+查看 [Apps](/apps) 下的完整目录，或直接跳到其中一个 - 例如，如果您想要一款工作区风格的应用，[Dispatch](/docs/template-dispatch) 是一个很好的起点。
 
 ## 开箱即用的东西 {#what-you-get}
 

package/corpus/core/docs/content/locales/zh-CN/evals.mdx

@@ -229,6 +229,13 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1`  | 至少有一项评估得分低于阈值，或者套件出错。            |
 | `2`  | 错误的参数（例如 `--threshold` 位于 `[0, 1]` 之外）。 |
 
+### 跳过受控评估 {#skipping}
+
+当某个评估套件只应在手动、夜间或具备密钥的环境中运行时，请使用
+`skipReason`。被跳过的评估不会运行代理或 scorer，不会让套件失败，并会
+在 JSON 输出中显示 `status: "skipped"` 和 `skipReason`。廉价的确定性覆
+盖应保留在普通 PR CI 中。
+
 ### 作为 CI 部署门 {#ci}
 
 将其添加到部署之前运行的管道：

package/corpus/core/docs/content/locales/zh-CN/faq.mdx

@@ -114,7 +114,7 @@ Anthropic Claude、OpenAI（GPT-5 系列）、Google Gemini 以及使用 OpenAI
 
 ### 我可以将现有应用迁移到代理原生吗？ {#can-i-use-existing-code}
 
-可以，但是从头开始构建原生代理效果最好。架构——共享数据库、轮询同步、actions、应用程序状态——需要始终集成。从模板开始并自定义它是推荐的路径。可以把它想象成从桌面优先到移动优先的转变：你*可以*改造，但构建原生更好。
+可以，但是从头开始构建原生代理效果最好。架构——共享数据库、轮询同步、actions、应用程序状态——需要始终集成。从应用开始并自定义它是推荐的路径。可以把它想象成从桌面优先到移动优先的转变：你*可以*改造，但构建原生更好。
 
 ## 模板以及您可以构建的内容 {#templates}
 

package/corpus/core/docs/content/locales/zh-CN/template-chat.mdx

@@ -13,7 +13,7 @@ description: "一个最小的聊天优先代理本机应用程序：持久的聊
   <Screen
     surface="desktop"
     html={
-      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>我能帮什么？</h1><p class='wf-muted' style='margin:10px 0 0'>聊任何事情。添加 actions、组件、页面、任务或你自己的后端。</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>给代理发消息...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 4.6 · Auto</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
+      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>我能帮什么？</h1><p class='wf-muted' style='margin:10px 0 0'>聊任何事情。添加 actions、组件、页面、任务或你自己的后端。</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>给代理发消息...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 5 · 自动</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
     }
   />
 </WireframeBlock>

package/corpus/core/docs/content/locales/zh-TW/cloneable-saas.mdx

@@ -30,9 +30,9 @@ description: "分叉一個可用的 SaaS 產品並使其成為您的產品 - 包
 | [**Plan**](/docs/template-plan)           | 帶有圖表、線框圖和註解的視覺計畫和 PR 概述。                                                    |
 | [**Dispatch**](/docs/template-dispatch)   | 工作區控制平面：共用秘密、可重用整合、Slack/Telegram、計畫作業。                                |
 
-不需要域範本？當您想要一個使用者可以立即交談的基本應用程式時，請使用 [Chat](/docs/template-chat)，或者使用 [Pure-Agent Apps](/docs/pure-agent-apps) 首先開始操作。
+不需要域應用程式？當您想要一個使用者可以立即交談的基本應用程式時，請使用 [Chat](/docs/template-chat)，或者從 [Pure-Agent Apps](/docs/pure-agent-apps) 中的一個操作開始。
 
-檢視 [Templates](/templates) 下的完整目錄，或直接跳到其中一個 - 例如，如果您想要一款工作區風格的應用，[Dispatch](/docs/template-dispatch) 是一個很好的起點。
+檢視 [Apps](/apps) 下的完整目錄，或直接跳到其中一個 - 例如，如果您想要一款工作區風格的應用，[Dispatch](/docs/template-dispatch) 是一個很好的起點。
 
 ## 開箱即用的東西 {#what-you-get}
 

package/corpus/core/docs/content/locales/zh-TW/evals.mdx

@@ -229,6 +229,13 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1`    | 至少有一項評估得分低於閾值，或者套件出錯。            |
 | `2`    | 錯誤的參數（例如 `--threshold` 位於 `[0, 1]` 之外）。 |
 
+### 跳過受控評估 {#skipping}
+
+當某個評估套件只應在手動、夜間或具備密鑰的環境中執行時，請使用
+`skipReason`。被跳過的評估不會執行代理或 scorer，不會讓套件失敗，並會
+在 JSON 輸出中顯示 `status: "skipped"` 和 `skipReason`。廉價的確定性覆
+蓋應保留在普通 PR CI 中。
+
 ### 作為 CI 部署門 {#ci}
 
 將其新增到部署之前執行的管道：

package/corpus/core/docs/content/locales/zh-TW/faq.mdx

@@ -114,7 +114,7 @@ Anthropic Claude、OpenAI（GPT-5 系列）、Google Gemini 以及使用 OpenAI
 
 ### 我可以將現有應用遷移到 Agent-Native 嗎？ {#can-i-use-existing-code}
 
-可以，但是從頭開始建置原生代理效果最好。架構——共用資料庫、輪詢同步、actions、應用程式狀態——需要始終整合。從範本開始並自訂它是推薦的路徑。可以把它想象成從桌面優先到行動優先的轉變：你*可以*改造，但建置原生更好。
+可以，但是從頭開始建置原生代理效果最好。架構——共用資料庫、輪詢同步、actions、應用程式狀態——需要始終整合。從應用程式開始並自訂它是推薦的路徑。可以把它想象成從桌面優先到行動優先的轉變：你*可以*改造，但建置原生更好。
 
 ## 範本以及您可以建置的內容 {#templates}
 

package/corpus/core/docs/content/locales/zh-TW/template-chat.mdx

@@ -13,7 +13,7 @@ description: "一個最小的聊天優先 Agent-Native 應用程式：持久的
   <Screen
     surface="desktop"
     html={
-      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>我能幫什麼？</h1><p class='wf-muted' style='margin:10px 0 0'>聊任何事情。新增 actions、元件、頁面、工作或你自己的後端。</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>給代理發訊息...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 4.6 · Auto</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
+      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>我能幫什麼？</h1><p class='wf-muted' style='margin:10px 0 0'>聊任何事情。新增 actions、元件、頁面、工作或你自己的後端。</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>給代理發訊息...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 5 · 自動</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
     }
   />
 </WireframeBlock>

package/corpus/core/docs/content/template-chat.mdx

@@ -13,7 +13,7 @@ If you want the smallest action-only runtime with no browser UI, start with [Pur
   <Screen
     surface="desktop"
     html={
-      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>How can I help?</h1><p class='wf-muted' style='margin:10px 0 0'>Chat about anything. Add actions, components, pages, jobs, or your own backend.</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>Message the agent...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 4.6 · Auto</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
+      "<div style='min-height:560px;box-sizing:border-box;display:flex;align-items:center;justify-content:center;padding:56px 40px'><div style='display:flex;flex-direction:column;align-items:center;justify-content:center;gap:28px;width:min(700px,92%);min-height:430px'><div style='height:34px'></div><div style='text-align:center'><h1 style='margin:0'>How can I help?</h1><p class='wf-muted' style='margin:10px 0 0'>Chat about anything. Add actions, components, pages, jobs, or your own backend.</p></div><div class='wf-card' style='width:100%;min-height:150px;display:flex;flex-direction:column;gap:18px'><span class='wf-muted'>Message the agent...</span><div style='flex:1'></div><div style='display:flex;align-items:center;gap:10px'><span data-icon='plus' aria-label='Attach'></span><div style='flex:1'></div><span class='wf-pill'>Sonnet 5 · Auto</span><span class='wf-pill'>Act</span><button class='primary'>↑</button></div></div><div style='height:34px'></div></div></div>"
     }
   />
 </WireframeBlock>

package/corpus/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agent-native/core",
-  "version": "0.83.0",
+  "version": "0.84.1",
   "description": "Framework for agent-native application development — where AI agents and UI share SQL state, actions, and context",
   "homepage": "https://github.com/BuilderIO/agent-native#readme",
   "bugs": {

package/corpus/core/src/agent/engine/builtin.ts

@@ -96,7 +96,7 @@ export function registerBuiltinEngines(): void {
       "Claude models through the Vercel AI SDK. Supports thinking and caching via AI SDK providerOptions.",
     openai: "OpenAI GPT models via the Vercel AI SDK. Requires OPENAI_API_KEY.",
     openrouter:
-      "300+ models from Anthropic, OpenAI, Google, Meta, and more routed through a single endpoint. Use model IDs like 'anthropic/claude-sonnet-4.6' or 'openai/gpt-5.5'. Requires OPENROUTER_API_KEY.",
+      "300+ models from Anthropic, OpenAI, Google, Z.ai, and more routed through a single endpoint. Use model IDs like 'anthropic/claude-sonnet-5', 'openai/gpt-5.5', or 'z-ai/glm-5.2'. Requires OPENROUTER_API_KEY.",
     google:
       "Google Gemini models via the Vercel AI SDK. Requires GOOGLE_GENERATIVE_AI_API_KEY.",
     groq: "Groq LPU inference via the Vercel AI SDK. Requires GROQ_API_KEY.",

package/corpus/core/src/agent/engine/output-tokens.ts

@@ -1,5 +1,5 @@
 const MIN_MAX_OUTPUT_TOKENS = 256;
-// Raise the global clamp to 64K to support models like claude-sonnet-4-6
+// Raise the global clamp to 64K to support models like claude-sonnet-5
 // (64K) and GPT-5.x (up to 128K). Callers can still set higher explicit
 // per-call values; this clamp only applies when no explicit value is given.
 const MAX_MAX_OUTPUT_TOKENS = 64_000;

package/corpus/core/src/agent/harness/acp-adapter.ts

@@ -171,6 +171,7 @@ class AcpHarnessSession implements AgentHarnessSession {
   private queue: AsyncEventQueue<AgentHarnessEvent> | null = null;
   private readonly pendingPermissions = new Map<string, PendingPermission>();
   private readonly toolTitles = new Map<string, string>();
+  private readonly toolInputs = new Map<string, Record<string, unknown>>();
   private approvalCounter = 0;
   private stderrTail = "";
   private childExited = false;
@@ -358,12 +359,20 @@ class AcpHarnessSession implements AgentHarnessSession {
     if (update?.sessionUpdate === "tool_call" && update.title) {
       this.toolTitles.set(update.toolCallId, update.title);
     }
+    if (
+      (update?.sessionUpdate === "tool_call" ||
+        update?.sessionUpdate === "tool_call_update") &&
+      update.rawInput
+    ) {
+      this.toolInputs.set(update.toolCallId, update.rawInput);
+    }
     // Updates that arrive without an active turn are history replay from
     // loadSession; the transcript already contains them, so drop them.
     if (!this.queue) return;
-    for (const event of acpUpdateToHarnessEvents(update, (id) =>
-      this.toolTitles.get(id),
-    )) {
+    for (const event of acpUpdateToHarnessEvents(update, {
+      titleFor: (id) => this.toolTitles.get(id),
+      inputFor: (id) => this.toolInputs.get(id),
+    })) {
       this.queue.push(event);
     }
   }
@@ -488,8 +497,17 @@ function messageToText(content: string | unknown[]): string {
  */
 export function acpUpdateToHarnessEvents(
   update: AcpSessionUpdate,
-  titleFor?: (toolCallId: string) => string | undefined,
+  resolvers?:
+    | ((toolCallId: string) => string | undefined)
+    | {
+        titleFor?: (toolCallId: string) => string | undefined;
+        inputFor?: (toolCallId: string) => Record<string, unknown> | undefined;
+      },
 ): AgentHarnessEvent[] {
+  const titleFor =
+    typeof resolvers === "function" ? resolvers : resolvers?.titleFor;
+  const inputFor =
+    typeof resolvers === "function" ? undefined : resolvers?.inputFor;
   switch (update.sessionUpdate) {
     case "agent_message_chunk": {
       const text = acpContentBlockToText(update.content);
@@ -517,6 +535,9 @@ export function acpUpdateToHarnessEvents(
           type: "tool-done",
           id: update.toolCallId,
           name: update.title || titleFor?.(update.toolCallId) || "tool",
+          ...((update.rawInput ?? inputFor?.(update.toolCallId))
+            ? { input: update.rawInput ?? inputFor?.(update.toolCallId) }
+            : {}),
           result: update.rawOutput ?? acpToolContentText(update.content),
         });
       }
@@ -531,6 +552,9 @@ export function acpUpdateToHarnessEvents(
           type: "tool-done",
           id: update.toolCallId,
           name: update.title || titleFor?.(update.toolCallId) || "tool",
+          ...((update.rawInput ?? inputFor?.(update.toolCallId))
+            ? { input: update.rawInput ?? inputFor?.(update.toolCallId) }
+            : {}),
           result: update.rawOutput ?? acpToolContentText(content),
         });
       }

package/corpus/core/src/agent/harness/ai-sdk-adapter.ts

@@ -448,6 +448,9 @@ export function aiSdkHarnessPartToEvents(part: any): AgentHarnessEvent[] {
         type: "tool-done",
         id: part.toolCallId ?? part.id,
         name: part.toolName ?? part.name ?? "tool",
+        ...(part.input !== undefined || part.args !== undefined
+          ? { input: part.input ?? part.args }
+          : {}),
         result: part.output ?? part.result,
       });
       break;

package/corpus/core/src/agent/harness/translate.ts

@@ -32,6 +32,9 @@ export function agentHarnessEventToAgentChatEvents(
         {
           type: "tool_done",
           tool: event.name,
+          ...(event.input !== undefined
+            ? { input: normalizeToolInput(event.input) }
+            : {}),
           result: stringifyResult(event.result),
           ...(event.mcpApp ? { mcpApp: event.mcpApp } : {}),
         },

package/corpus/core/src/agent/harness/types.ts

@@ -100,6 +100,7 @@ export type AgentHarnessEvent =
       type: "tool-done";
       id?: string;
       name: string;
+      input?: unknown;
       result?: unknown;
       mcpApp?: AgentMcpAppPayload;
     }

package/corpus/core/src/agent/model-config.ts

@@ -1,10 +1,9 @@
 /**
  * Central model catalog for built-in agent engines.
  *
- * To bump the framework's managed default, update
- * FRAMEWORK_DEFAULT_OPENAI_MODEL. Builder gateway and OpenRouter IDs are
- * derived from that provider-native OpenAI ID so the usual default bump stays
- * in this one file.
+ * To bump the framework's managed default, update the Anthropic/OpenAI
+ * constants below. Builder gateway and OpenRouter IDs are derived here so the
+ * usual default bump stays in one file.
  */
 
 // ---------------------------------------------------------------------------
@@ -15,6 +14,7 @@
 //  OpenAI     https://developers.openai.com/api/docs/models/gpt-5.5
 //             https://developers.openai.com/api/docs/models/gpt-5.4
 //  Google     https://ai.google.dev/gemini-api/docs/models
+//  OpenRouter https://openrouter.ai/api/v1/models
 //
 // Family defaults (used when a model id isn't listed explicitly):
 //  claude-*        → 200_000  (Haiku 4.5 and earlier models)
@@ -22,7 +22,7 @@
 //  gemini-2* / gemini-3* → 1_048_576
 //  everything else → 128_000  (safe conservative floor)
 //
-// Note: Fable 5, Sonnet 4.6, and Opus 4.6/4.7/4.8 support 1M via the API
+// Note: Fable 5, Sonnet 5, Sonnet 4.6, and Opus 4.6/4.7/4.8 support 1M via the API
 // at standard prices.
 // ---------------------------------------------------------------------------
 
@@ -31,6 +31,7 @@ const MODEL_CONTEXT_WINDOWS: Record<string, number> = {
   "claude-fable-5": 1_000_000,
   "claude-opus-4-8": 1_000_000,
   "claude-opus-4-7": 1_000_000,
+  "claude-sonnet-5": 1_000_000,
   "claude-sonnet-4-6": 1_000_000,
   "claude-haiku-4-5": 200_000,
   "claude-haiku-4-5-20251001": 200_000,
@@ -49,10 +50,12 @@ const MODEL_CONTEXT_WINDOWS: Record<string, number> = {
   "anthropic/claude-fable-5": 1_000_000,
   "anthropic/claude-opus-4.8": 1_000_000,
   "anthropic/claude-opus-4.7": 1_000_000,
+  "anthropic/claude-sonnet-5": 1_000_000,
   "anthropic/claude-sonnet-4.6": 1_000_000,
   "openai/gpt-5.5": 1_050_000,
   "openai/gpt-5.4": 1_050_000,
   "google/gemini-2.5-flash": 1_048_576,
+  "z-ai/glm-5.2": 1_048_576,
 
   // ── AI-SDK native OpenAI IDs ──────────────────────────────────────────────
   "gpt-5.5": 1_050_000,
@@ -85,12 +88,13 @@ export function getContextWindowForModel(modelId: string): number {
   // Family heuristics for unlisted model IDs
   const id = modelId.toLowerCase();
 
-  // Anthropic Fable 5, Opus 4.x, and Sonnet 4.6 = 1M
+  // Anthropic Fable 5, Opus 4.x, Sonnet 5, and Sonnet 4.6 = 1M
   if (
     id === "claude-fable-5" ||
     id.includes("claude-fable-5") ||
     id.startsWith("claude-opus-4") ||
-    id === "claude-sonnet-4-6" ||
+    id.includes("claude-sonnet-5") ||
+    id.includes("claude-sonnet-4-6") ||
     id.includes("claude-sonnet-4.6")
   )
     return 1_000_000;
@@ -110,10 +114,26 @@ export function getContextWindowForModel(modelId: string): number {
   )
     return 1_048_576;
 
+  // Z.ai GLM 5.x on OpenRouter — 1M context.
+  if (id.startsWith("glm-5") || id.includes("/glm-5")) return 1_048_576;
+
   return DEFAULT_CONTEXT_WINDOW;
 }
 
-const ANTHROPIC_DEFAULT_MODEL_ID = "claude-sonnet-4-6";
+const ENABLE_CLAUDE_SONNET_5 = false;
+
+export const CLAUDE_SONNET_MODEL_ID = ENABLE_CLAUDE_SONNET_5
+  ? "claude-sonnet-5"
+  : "claude-sonnet-4-6";
+export const CLAUDE_SONNET_MODEL_LABEL = ENABLE_CLAUDE_SONNET_5
+  ? "Claude Sonnet 5"
+  : "Claude Sonnet 4.6";
+
+const OPENROUTER_CLAUDE_SONNET_MODEL_ID = ENABLE_CLAUDE_SONNET_5
+  ? "anthropic/claude-sonnet-5"
+  : "anthropic/claude-sonnet-4.6";
+
+const ANTHROPIC_DEFAULT_MODEL_ID = CLAUDE_SONNET_MODEL_ID;
 
 function builderGatewayModelId(model: string): string {
   return model.replace(/\./g, "-");
@@ -138,8 +158,8 @@ export const AGENT_MODEL_CONFIG = {
     defaultModel: FRAMEWORK_DEFAULT_BUILDER_MODEL,
     supportedModels: [
       "auto",
-      "claude-opus-4-7",
-      FRAMEWORK_DEFAULT_BUILDER_MODEL,
+      "claude-opus-4-8",
+      CLAUDE_SONNET_MODEL_ID,
       "claude-haiku-4-5",
       FRAMEWORK_DEFAULT_BUILDER_OPENAI_MODEL,
       "gpt-5-4",
@@ -156,7 +176,7 @@ export const AGENT_MODEL_CONFIG = {
     supportedModels: [
       "claude-fable-5",
       "claude-opus-4-8",
-      ANTHROPIC_DEFAULT_MODEL_ID,
+      CLAUDE_SONNET_MODEL_ID,
       "claude-haiku-4-5-20251001",
     ],
   },
@@ -166,7 +186,7 @@ export const AGENT_MODEL_CONFIG = {
       supportedModels: [
         "claude-fable-5",
         "claude-opus-4-8",
-        ANTHROPIC_DEFAULT_MODEL_ID,
+        CLAUDE_SONNET_MODEL_ID,
         "claude-haiku-4-5-20251001",
       ],
     },
@@ -183,11 +203,12 @@ export const AGENT_MODEL_CONFIG = {
       supportedModels: [
         "anthropic/claude-fable-5",
         "anthropic/claude-opus-4.8",
-        "anthropic/claude-sonnet-4.6",
+        OPENROUTER_CLAUDE_SONNET_MODEL_ID,
         FRAMEWORK_DEFAULT_OPENROUTER_MODEL,
         "openai/gpt-5.4",
         // Current stable Gemini on OpenRouter (2.5 Flash is GA)
         "google/gemini-2.5-flash",
+        "z-ai/glm-5.2",
       ],
     },
     google: {

package/corpus/core/src/agent/production-agent.ts

@@ -3134,7 +3134,13 @@ export async function runAgentLoop(opts: {
           tool: toolCall.name,
           input: toolCall.input as Record<string, string>,
         });
-        send({ type: "tool_done", tool: toolCall.name, result });
+        send({
+          type: "tool_done",
+          tool: toolCall.name,
+          input: toolCall.input as Record<string, unknown>,
+          result,
+          completedSideEffect: false,
+        });
         recordToolResult(result, false);
         return {
           type: "tool-result" as const,
@@ -3152,7 +3158,13 @@ export async function runAgentLoop(opts: {
           tool: toolCall.name,
           input: toolCall.input as Record<string, string>,
         });
-        send({ type: "tool_done", tool: toolCall.name, result });
+        send({
+          type: "tool_done",
+          tool: toolCall.name,
+          input: toolCall.input as Record<string, unknown>,
+          result,
+          completedSideEffect: false,
+        });
         recordToolResult(result, false);
         return {
           type: "tool-result" as const,
@@ -3172,7 +3184,14 @@ export async function runAgentLoop(opts: {
           tool: toolCall.name,
           input: toolCall.input as Record<string, string>,
         });
-        send({ type: "tool_done", tool: toolCall.name, result });
+        send({
+          type: "tool_done",
+          tool: toolCall.name,
+          input: toolCall.input as Record<string, unknown>,
+          result,
+          isError: true,
+          completedSideEffect: false,
+        });
         recordToolResult(result, true);
         return {
           type: "tool-result" as const,
@@ -3247,7 +3266,13 @@ export async function runAgentLoop(opts: {
             `Awaiting human approval to run "${toolCall.name}". This action did ` +
             `NOT execute — a human must approve this specific call before it ` +
             `can run. The turn is paused; do not retry.`;
-          send({ type: "tool_done", tool: toolCall.name, result });
+          send({
+            type: "tool_done",
+            tool: toolCall.name,
+            input: toolCall.input as Record<string, unknown>,
+            result,
+            completedSideEffect: false,
+          });
           recordToolResult(result, false);
           requestedActionStop ??= {
             message: `Waiting for your approval to run ${toolCall.name}.`,
@@ -3280,7 +3305,13 @@ export async function runAgentLoop(opts: {
           tool: toolCall.name,
           input: toolCall.input as Record<string, string>,
         });
-        send({ type: "tool_done", tool: toolCall.name, result });
+        send({
+          type: "tool_done",
+          tool: toolCall.name,
+          input: toolCall.input as Record<string, unknown>,
+          result,
+          completedSideEffect: false,
+        });
         recordToolResult(result, false);
         if (repeats >= 3) {
           requestedActionStop ??= {
@@ -3338,7 +3369,13 @@ export async function runAgentLoop(opts: {
             tool: toolCall.name,
             input: toolCall.input as Record<string, string>,
           });
-          send({ type: "tool_done", tool: toolCall.name, result });
+          send({
+            type: "tool_done",
+            tool: toolCall.name,
+            input: toolCall.input as Record<string, unknown>,
+            result,
+            completedSideEffect: false,
+          });
           recordToolResult(result, false);
           return {
             type: "tool-result" as const,
@@ -3389,6 +3426,7 @@ export async function runAgentLoop(opts: {
             send({
               type: "tool_done",
               tool: toolCall.name,
+              input: toolCall.input as Record<string, unknown>,
               result,
               ...(actionEntry.chatUI ? { chatUI: actionEntry.chatUI } : {}),
             });
@@ -3413,7 +3451,14 @@ export async function runAgentLoop(opts: {
             tool: toolCall.name,
             input: toolCall.input as Record<string, string>,
           });
-          send({ type: "tool_done", tool: toolCall.name, result });
+          send({
+            type: "tool_done",
+            tool: toolCall.name,
+            input: toolCall.input as Record<string, unknown>,
+            result,
+            isError: true,
+            completedSideEffect: false,
+          });
           recordToolResult(result, true);
           requestedActionStop ??= {
             message:
@@ -3467,7 +3512,14 @@ export async function runAgentLoop(opts: {
             toolCallSchemaError.error,
           ),
         );
-        send({ type: "tool_done", tool: toolCall.name, result });
+        send({
+          type: "tool_done",
+          tool: toolCall.name,
+          input: toolCall.input as Record<string, unknown>,
+          result,
+          isError: true,
+          completedSideEffect: false,
+        });
         recordToolResult(result, true);
         return {
           type: "tool-result" as const,
@@ -3491,7 +3543,14 @@ export async function runAgentLoop(opts: {
             rawToolInputError,
           ),
         );
-        send({ type: "tool_done", tool: toolCall.name, result });
+        send({
+          type: "tool_done",
+          tool: toolCall.name,
+          input: toolCall.input as Record<string, unknown>,
+          result,
+          isError: true,
+          completedSideEffect: false,
+        });
         recordToolResult(result, true);
         return {
           type: "tool-result" as const,
@@ -3508,7 +3567,14 @@ export async function runAgentLoop(opts: {
         !isPlanModeToolCallAllowed(toolCall.name, toolCall.input, actionEntry)
       ) {
         const result = planModeBlockedMessage(toolCall.name);
-        send({ type: "tool_done", tool: toolCall.name, result });
+        send({
+          type: "tool_done",
+          tool: toolCall.name,
+          input: toolCall.input as Record<string, unknown>,
+          result,
+          isError: true,
+          completedSideEffect: false,
+        });
         recordToolResult(result, true);
         return {
           type: "tool-result" as const,
@@ -3586,6 +3652,14 @@ export async function runAgentLoop(opts: {
           actionPromise
             .then((zombieRaw: unknown) => {
               const zombieMcp = isMcpActionResult(zombieRaw) ? zombieRaw : null;
+              if (
+                zombieMcp &&
+                zombieMcp.raw &&
+                typeof zombieMcp.raw === "object" &&
+                (zombieMcp.raw as Record<string, unknown>).isError === true
+              ) {
+                return;
+              }
               const zombieText = zombieMcp ? zombieMcp.text : zombieRaw;
               const zombieStr =
                 typeof zombieText === "string"
@@ -3721,7 +3795,10 @@ export async function runAgentLoop(opts: {
       send({
         type: "tool_done",
         tool: toolCall.name,
+        input: toolCall.input as Record<string, unknown>,
         result,
+        ...(isError ? { isError: true } : {}),
+        ...(isError ? { completedSideEffect: false } : {}),
         ...(mcpApp ? { mcpApp } : {}),
         ...(actionEntry.chatUI ? { chatUI: actionEntry.chatUI } : {}),
       });