Bump version from 4.1.2 to 4.1.3

Merge pull request #352 from shern-point/fix/tool-string-schema-protection
Fix/tool type schema protection
2026-05-03 16:05:26 +08:00 · 2026-04-29 07:55:48 +08:00 · 2026-04-29 07:51:21 +08:00 · 2026-04-29 07:50:54 +08:00 · 2026-04-28 12:18:37 -07:00 · 2026-04-29 02:00:20 +08:00
56 changed files with 1630 additions and 232 deletions
--- a/12
+++ b/12
@@ -28,6 +28,8 @@ FROM debian:bookworm-slim AS runtime-base
 WORKDIR /app
 RUN apt-get update \
    && apt-get install -y --no-install-recommends ca-certificates \
+    && groupadd -r ds2api && useradd -r -g ds2api -d /app -s /sbin/nologin ds2api \
+    && mkdir -p /app/data && chown -R ds2api:ds2api /app \
    && rm -rf /var/lib/apt/lists/*
 COPY --from=busybox-tools /bin/busybox /usr/local/bin/busybox
 EXPOSE 5001
@@ -36,8 +38,9 @@ CMD ["/usr/local/bin/ds2api"]
 FROM runtime-base AS runtime-from-source
 COPY --from=go-builder /out/ds2api /usr/local/bin/ds2api

-COPY --from=go-builder /app/config.example.json /app/config.example.json
-COPY --from=webui-builder /app/static/admin /app/static/admin
+COPY --from=go-builder --chown=ds2api:ds2api /app/config.example.json /app/config.example.json
+COPY --from=webui-builder --chown=ds2api:ds2api /app/static/admin /app/static/admin
+USER ds2api

 FROM busybox-tools AS dist-extract
 ARG TARGETARCH
@@ -60,7 +63,8 @@ RUN set -eux; \
 FROM runtime-base AS runtime-from-dist
 COPY --from=dist-extract /out/ds2api /usr/local/bin/ds2api

-COPY --from=dist-extract /out/config.example.json /app/config.example.json
-COPY --from=dist-extract /out/static/admin /app/static/admin
+COPY --from=dist-extract --chown=ds2api:ds2api /out/config.example.json /app/config.example.json
+COPY --from=dist-extract --chown=ds2api:ds2api /out/static/admin /app/static/admin
+USER ds2api

 FROM runtime-from-source AS final
--- a/README.MD
+++ b/README.MD
@@ -31,6 +31,30 @@
 >
 > 请勿将本项目用于违反服务条款、协议、法律法规或平台规则的场景。商业使用前请自行确认 `LICENSE`、相关协议以及你是否获得了作者的书面许可。

+## 目录
+
+- [架构概览（摘要）](#架构概览摘要)
+- [核心能力](#核心能力)
+- [平台兼容矩阵](#平台兼容矩阵)
+- [模型支持](#模型支持)
+  - [OpenAI 接口](#openai-接口get-v1models)
+  - [Claude 接口](#claude-接口get-anthropicv1models)
+  - [Gemini 接口](#gemini-接口)
+- [快速开始](#快速开始)
+  - [方式一：下载 Release 构建包](#方式一下载-release-构建包)
+  - [方式二：Docker 运行](#方式二docker-运行)
+  - [方式三：Vercel 部署](#方式三vercel-部署)
+  - [方式四：本地源码运行](#方式四本地源码运行)
+- [配置说明](#配置说明)
+- [鉴权模式](#鉴权模式)
+- [并发模型](#并发模型)
+- [Tool Call 适配](#tool-call-适配)
+- [本地开发抓包工具](#本地开发抓包工具)
+- [文档索引](#文档索引)
+- [测试](#测试)
+- [Release 自动构建（GitHub Actions）](#release-自动构建github-actions)
+- [免责声明](#免责声明)
+
 ## 架构概览（摘要）

 ```mermaid
--- a/README.en.md
+++ b/README.en.md
@@ -28,6 +28,30 @@ Documentation entry: [Docs Index](docs/README.md) / [Architecture](docs/ARCHITEC
 >
 > Do not use this project in ways that violate service terms, agreements, laws, or platform rules. Before any commercial use, review the `LICENSE`, the relevant terms, and confirm that you have the author's written permission.

+## Table of Contents
+
+- [Architecture Overview (Summary)](#architecture-overview-summary)
+- [Key Capabilities](#key-capabilities)
+- [Platform Compatibility Matrix](#platform-compatibility-matrix)
+- [Model Support](#model-support)
+  - [OpenAI Endpoint](#openai-endpoint-get-v1models)
+  - [Claude Endpoint](#claude-endpoint-get-anthropicv1models)
+  - [Gemini Endpoint](#gemini-endpoint)
+- [Quick Start](#quick-start)
+  - [Option 1: Download Release Binaries](#option-1-download-release-binaries)
+  - [Option 2: Docker / GHCR](#option-2-docker--ghcr)
+  - [Option 3: Vercel](#option-3-vercel)
+  - [Option 4: Local Run](#option-4-local-run)
+- [Configuration](#configuration)
+- [Authentication Modes](#authentication-modes)
+- [Concurrency Model](#concurrency-model)
+- [Tool Call Adaptation](#tool-call-adaptation)
+- [Local Dev Packet Capture](#local-dev-packet-capture)
+- [Documentation Index](#documentation-index)
+- [Testing](#testing)
+- [Release Artifact Automation (GitHub Actions)](#release-artifact-automation-github-actions)
+- [Disclaimer](#disclaimer)
+
 ## Architecture Overview (Summary)

 ```mermaid
--- a/2
+++ b/2
@@ -1 +1 @@
-4.1.1
+4.1.3
--- a/cmd/ds2api/main.go
+++ b/cmd/ds2api/main.go
@@ -35,8 +35,9 @@ func main() {
 	}

 	srv := &http.Server{
-		Addr:    "0.0.0.0:" + port,
-		Handler: app.Router,
+		Addr:              "0.0.0.0:" + port,
+		Handler:           app.Router,
+		ReadHeaderTimeout: 5 * time.Second,
 	}
 	localURL := fmt.Sprintf("http://127.0.0.1:%s", port)
 	lanIP := detectLANIPv4()
--- a/docs/prompt-compatibility.md
+++ b/docs/prompt-compatibility.md
@@ -152,7 +152,9 @@ OpenAI Chat / Responses 在标准化后、current input file 之前，会默认

 工具调用正例现在优先示范官方 DSML 风格：`<|DSML|tool_calls>` → `<|DSML|invoke name="...">` → `<|DSML|parameter name="...">`。
 兼容层仍接受旧式纯 `<tool_calls>` wrapper，但提示词会优先要求模型输出官方 DSML 标签，并强调不能只输出 closing wrapper 而漏掉 opening tag。需要注意：这是“兼容 DSML 外壳，内部仍以 XML 解析语义为准”，不是原生 DSML 全链路实现；DSML 标签会在解析入口归一化回现有 XML 标签后继续走同一套 parser。
-数组参数使用 `<item>...</item>` 子节点表示；当某个参数体只包含 item 子节点时，Go / Node 解析器会把它还原成数组，避免 `questions` / `options` 这类 schema 中要求 array 的参数被误解析成 `{ "item": ... }` 对象。若模型把完整结构化 XML fragment 误包进 CDATA，兼容层会在保护 `content` / `command` 等原文字段的前提下，尝试把非原文字段中的 CDATA XML fragment 还原成 object / array。
+数组参数使用 `<item>...</item>` 子节点表示；当某个参数体只包含 item 子节点时，Go / Node 解析器会把它还原成数组，避免 `questions` / `options` 这类 schema 中要求 array 的参数被误解析成 `{ "item": ... }` 对象。若模型把完整结构化 XML fragment 误包进 CDATA，兼容层会在保护 `content` / `command` 等原文字段的前提下，尝试把非原文字段中的 CDATA XML fragment 还原成 object / array。不过，如果 CDATA 只是单个平面的 XML/HTML 标签，例如 `<b>urgent</b>` 这种行内标记，兼容层会保留原始字符串，不会强行升成 object / array；只有明显表示结构的 CDATA 片段，例如多兄弟节点、嵌套子节点或 `item` 列表，才会触发结构化恢复。
+在 assistant 最终回包阶段，如果某个 tool 参数在声明 schema 中明确是 `string`，兼容层会在把解析后的 `tool_calls` / `function_call` 重新序列化成 OpenAI / Responses / Claude 可见参数前，递归把该路径上的 number / bool / object / array 统一转成字符串；其中 object / array 会压成紧凑 JSON 字符串。这个保护只对 schema 明确声明为 string 的路径生效，不会改写本来就是 `number` / `boolean` / `object` / `array` 的参数。这样可以兼容 DeepSeek 输出了结构化片段、但上游客户端工具 schema 又严格要求字符串参数的场景（例如 `content`、`prompt`、`path`、`taskId` 等）。
+工具 schema 的权威来源始终是**当前请求实际携带的 schema**，而不是同名工具在其他 runtime（Claude Code / OpenCode / Codex 等）里的默认印象。兼容层现在会同时兼容 OpenAI 风格 `function.parameters`、直接工具对象上的 `parameters` / `input_schema`、以及 camelCase 的 `inputSchema` / `schema`，并在最终输出阶段按这份请求内 schema 决定是保留 array/object，还是仅对明确声明为 `string` 的路径做字符串化。该规则同样适用于 Claude 的流式收尾和 Vercel Node 流式 tool-call formatter，避免不同 runtime 因 schema shape 差异而出现同名工具参数类型漂移。
 正例中的工具名只会来自当前请求实际声明的工具；如果当前请求没有足够的已知工具形态，就省略对应的单工具、多工具或嵌套示例，避免把不可用工具名写进 prompt。
 对执行类工具，脚本内容必须进入执行参数本身：`Bash` / `execute_command` 使用 `command`，`exec_command` 使用 `cmd`；不要把脚本示范成 `path` / `content` 文件写入参数。

@@ -242,7 +244,7 @@ OpenAI 文件相关实现：

 兼容层现在只保留 `current_input_file` 这一种拆分方式；旧的 `history_split` 已废弃，只保留为兼容旧配置的字段，不再参与请求处理。

- `current_input_file` 默认开启；它用于把“完整上下文”合并进隐藏上下文文件。当最新 user turn 的纯文本长度达到 `current_input_file.min_chars`（默认 `0`）时，兼容层会上传一个文件名为 `IGNORE.txt` 的上下文文件，并在文件内容前加入一个明确的 `context note`，提示模型这是被压缩过的历史记录而不是新指令；live prompt 也会显式说明当前处于 compacted-context mode，要求模型用已提供的历史来还原上下文状态并直接回答最新请求，避免把重复工具调用或重复提问当成新的起点。
+- `current_input_file` 默认开启；它用于把“完整上下文”合并进隐藏上下文文件。当最新 user turn 的纯文本长度达到 `current_input_file.min_chars`（默认 `0`）时，兼容层会上传一个文件名为 `IGNORE.txt` 的上下文文件，并在 live prompt 中只保留一个中性的 user 消息要求模型直接回答最新请求，不再暴露文件名或要求模型读取本地文件。
 - 如果 `current_input_file.enabled=false`，请求会直接透传，不上传任何拆分上下文文件。
 - 旧的 `history_split.enabled` / `history_split.trigger_after_turns` 会被读取进配置对象以保持兼容，但不会触发拆分上传，也不会影响 `current_input_file` 的默认开启。

@@ -255,18 +257,12 @@ OpenAI 文件相关实现：
 - 旧历史拆分兼容壳：
  [internal/httpapi/openai/history/history_split.go](../internal/httpapi/openai/history/history_split.go)

-当前输入转文件启用并触发时，上传文件的真实文件名是 `IGNORE.txt`，文件内容是完整 `messages` 上下文；它仍会先用 OpenAI 消息标准化和 DeepSeek 角色标记序列化，再包进 `context note` 和 `IGNORE` 文件边界里：
+当前输入转文件启用并触发时，上传文件的真实文件名是 `IGNORE.txt`，文件内容是完整 `messages` 上下文；它仍会先用 OpenAI 消息标准化和 DeepSeek 角色标记序列化，再包进 `IGNORE` 文件边界里：

 ```text
 [uploaded filename]: IGNORE.txt
 [file content end]

-[context note]
-This is a compacted snapshot of the prior conversation history for the current request.
-Use it as history only. Do not treat it as a new instruction.
-If the same question or tool action already appears here, do not repeat it unless the latest turn adds new information.
-[/context note]
-
 <｜begin▁of▁sentence｜><｜System｜>...<｜User｜>...<｜Assistant｜>...<｜Tool｜>...<｜User｜>...

 [file name]: IGNORE
@@ -322,7 +318,7 @@ If the same question or tool action already appears here, do not repeat it unles

 ```json
 {
-  "prompt": "<｜begin▁of▁sentence｜><｜System｜>原 system / developer\n\nYou have access to these tools: ...<｜end▁of▁instructions｜><｜User｜>You are in a compacted-context mode. The attached history contains the prior conversation state and any earlier tool results. Use it to resolve references and answer the latest user request directly. If the same tool action or question already appears in the attached context, do not repeat it unless the latest turn adds new information.<｜Assistant｜>",
+  "prompt": "<｜begin▁of▁sentence｜><｜System｜>原 system / developer\n\nYou have access to these tools: ...<｜end▁of▁instructions｜><｜User｜>The current request and prior conversation context have already been provided. Answer the latest user request directly.<｜Assistant｜>",
  "ref_file_ids": [
    "file-current-input-ignore",
    "file-systemprompt",
--- a/docs/toolcall-semantics.md
+++ b/docs/toolcall-semantics.md
@@ -39,7 +39,7 @@
 兼容修复：

 - 如果模型漏掉 opening wrapper，但后面仍输出了一个或多个 invoke 并以 closing wrapper 收尾，Go 解析链路会在解析前补回缺失的 opening wrapper。
- 如果模型把 DSML 标签里的分隔符 `|` 写漏成空格（例如 `<|DSML tool_calls>` / `<|DSML invoke>` / `<|DSML parameter>`，或无 leading pipe 的 `<DSML tool_calls>` 形态），或把 `DSML` 与工具标签名直接黏连（例如 `<DSMLtool_calls>` / `<DSMLinvoke>` / `<DSMLparameter>`），Go / Node 会在固定工具标签名范围内归一化；相似但非工具标签名（如 `tool_calls_extra`）仍按普通文本处理。
+- 如果模型把 DSML 标签里的分隔符 `|` 写漏成空格（例如 `<|DSML tool_calls>` / `<|DSML invoke>` / `<|DSML parameter>`，或无 leading pipe 的 `<DSML tool_calls>` 形态），或把 `DSML` 与工具标签名直接黏连（例如 `<DSMLtool_calls>` / `<DSMLinvoke>` / `<DSMLparameter>`），或把最前面的 pipe 误写成全宽竖线（例如 `<｜DSML|tool_calls>` / `<｜DSML|invoke>` / `<｜DSML|parameter>`），Go / Node 会在固定工具标签名范围内归一化；相似但非工具标签名（如 `tool_calls_extra`）仍按普通文本处理。
 - 这是一个针对常见模型失误的窄修复，不改变推荐输出格式；prompt 仍要求模型直接输出完整 DSML 外壳。
 - 裸 `<invoke ...>` / `<parameter ...>` 不会被当成“已支持的工具语法”；只有 `tool_calls` wrapper 或可修复的缺失 opening wrapper 才会进入工具调用路径。

@@ -53,7 +53,7 @@

 在流式链路中（Go / Node 一致）：

- DSML `<|DSML|tool_calls>` wrapper、兼容变体（`<dsml|tool_calls>`、`<｜tool_calls>`、`<|tool_calls>`）、窄容错空格分隔形态（如 `<|DSML tool_calls>`）、黏连形态（如 `<DSMLtool_calls>`）和 canonical `<tool_calls>` wrapper 都会进入结构化捕获
+- DSML `<|DSML|tool_calls>` wrapper、兼容变体（`<dsml|tool_calls>`、`<｜tool_calls>`、`<|tool_calls>`、`<｜DSML|tool_calls>`）、窄容错空格分隔形态（如 `<|DSML tool_calls>`）、黏连形态（如 `<DSMLtool_calls>`）和 canonical `<tool_calls>` wrapper 都会进入结构化捕获
 - 如果流里直接从 invoke 开始，但后面补上了 closing wrapper，Go 流式筛分也会按缺失 opening wrapper 的修复路径尝试恢复
 - 已识别成功的工具调用不会再次回流到普通文本
 - 不符合新格式的块不会执行，并继续按原样文本透传
@@ -64,7 +64,7 @@

 另外，`<parameter>` 的值如果本身是合法 JSON 字面量，也会按结构化值解析，而不是一律保留为字符串。例如 `123`、`true`、`null`、`[1,2]`、`{"a":1}` 都会还原成对应的 number / boolean / null / array / object。
 结构化 XML 参数也会还原为 JSON 结构：如果参数体只包含一个或多个 `<item>...</item>` 子节点，会输出数组；嵌套对象里的 item-only 字段也同样按数组处理。例如 `<parameter name="questions"><item><question>...</question></item></parameter>` 会输出 `{"questions":[{"question":"..."}]}`，而不是 `{"questions":{"item":...}}`。
-如果模型误把完整结构化 XML fragment 放进 CDATA，Go / Node 会先保护明显的原文字段（如 `content` / `command` / `prompt` / `old_string` / `new_string`），其余参数会尝试把 CDATA 内的完整 XML fragment 还原成 object / array；常见的 `<br>` 分隔符会按换行归一化后再解析。
+如果模型误把完整结构化 XML fragment 放进 CDATA，Go / Node 会先保护明显的原文字段（如 `content` / `command` / `prompt` / `old_string` / `new_string`），其余参数会尝试把 CDATA 内的完整 XML fragment 还原成 object / array；常见的 `<br>` 分隔符会按换行归一化后再解析。但如果 CDATA 只是单个平面的 XML/HTML 标签，例如 `<b>urgent</b>` 这种行内标记，兼容层会把它保留为原始字符串，而不会强行升成 object / array；只有明显表示结构的 CDATA 片段，例如多兄弟节点、嵌套子节点或 `item` 列表，才会触发结构化恢复。

 ## 4) 输出结构

--- a/internal/format/openai/render_chat.go
+++ b/internal/format/openai/render_chat.go
@@ -6,12 +6,12 @@ import (
 	"time"
 )

-func BuildChatCompletion(completionID, model, finalPrompt, finalThinking, finalText string, toolNames []string) map[string]any {
+func BuildChatCompletion(completionID, model, finalPrompt, finalThinking, finalText string, toolNames []string, toolsRaw any) map[string]any {
 	detected := toolcall.ParseAssistantToolCallsDetailed(finalText, finalThinking, toolNames)
-	return BuildChatCompletionWithToolCalls(completionID, model, finalPrompt, finalThinking, finalText, detected.Calls)
+	return BuildChatCompletionWithToolCalls(completionID, model, finalPrompt, finalThinking, finalText, detected.Calls, toolsRaw)
 }

-func BuildChatCompletionWithToolCalls(completionID, model, finalPrompt, finalThinking, finalText string, detected []toolcall.ParsedToolCall) map[string]any {
+func BuildChatCompletionWithToolCalls(completionID, model, finalPrompt, finalThinking, finalText string, detected []toolcall.ParsedToolCall, toolsRaw any) map[string]any {
 	finishReason := "stop"
 	messageObj := map[string]any{"role": "assistant", "content": finalText}
 	if strings.TrimSpace(finalThinking) != "" {
@@ -19,7 +19,7 @@ func BuildChatCompletionWithToolCalls(completionID, model, finalPrompt, finalThi
 	}
 	if len(detected) > 0 {
 		finishReason = "tool_calls"
-		messageObj["tool_calls"] = toolcall.FormatOpenAIToolCalls(detected)
+		messageObj["tool_calls"] = toolcall.FormatOpenAIToolCalls(detected, toolsRaw)
 		messageObj["content"] = nil
 	}

--- a/internal/format/openai/render_responses.go
+++ b/internal/format/openai/render_responses.go
@@ -9,19 +9,19 @@ import (
 	"github.com/google/uuid"
 )

-func BuildResponseObject(responseID, model, finalPrompt, finalThinking, finalText string, toolNames []string) map[string]any {
+func BuildResponseObject(responseID, model, finalPrompt, finalThinking, finalText string, toolNames []string, toolsRaw any) map[string]any {
 	// Strict mode: only standalone, structured tool-call payloads are treated
 	// as executable tool calls.
 	detected := toolcall.ParseAssistantToolCallsDetailed(finalText, finalThinking, toolNames)
-	return BuildResponseObjectWithToolCalls(responseID, model, finalPrompt, finalThinking, finalText, detected.Calls)
+	return BuildResponseObjectWithToolCalls(responseID, model, finalPrompt, finalThinking, finalText, detected.Calls, toolsRaw)
 }

-func BuildResponseObjectWithToolCalls(responseID, model, finalPrompt, finalThinking, finalText string, detected []toolcall.ParsedToolCall) map[string]any {
+func BuildResponseObjectWithToolCalls(responseID, model, finalPrompt, finalThinking, finalText string, detected []toolcall.ParsedToolCall, toolsRaw any) map[string]any {
 	exposedOutputText := finalText
 	output := make([]any, 0, 2)
 	if len(detected) > 0 {
 		exposedOutputText = ""
-		output = append(output, toResponsesFunctionCallItems(detected)...)
+		output = append(output, toResponsesFunctionCallItems(detected, toolsRaw)...)
 	} else {
 		content := make([]any, 0, 2)
 		if finalThinking != "" {
@@ -74,12 +74,13 @@ func BuildResponseObjectFromItems(responseID, model, finalPrompt, finalThinking,
 	}
 }

-func toResponsesFunctionCallItems(toolCalls []toolcall.ParsedToolCall) []any {
+func toResponsesFunctionCallItems(toolCalls []toolcall.ParsedToolCall, toolsRaw any) []any {
 	if len(toolCalls) == 0 {
 		return nil
 	}
+	normalizedCalls := toolcall.NormalizeParsedToolCallsForSchemas(toolCalls, toolsRaw)
 	out := make([]any, 0, len(toolCalls))
-	for _, tc := range toolCalls {
+	for _, tc := range normalizedCalls {
 		if strings.TrimSpace(tc.Name) == "" {
 			continue
 		}
--- a/internal/format/openai/render_test.go
+++ b/internal/format/openai/render_test.go
@@ -1,8 +1,11 @@
 package openai

 import (
+	"encoding/json"
 	"strings"
 	"testing"
+
+	"ds2api/internal/toolcall"
 )

 func TestBuildResponseObjectKeepsFencedToolPayloadAsText(t *testing.T) {
@@ -13,6 +16,7 @@ func TestBuildResponseObjectKeepsFencedToolPayloadAsText(t *testing.T) {
 		"",
 		"```json\n{\"tool_calls\":[{\"name\":\"search\",\"input\":{\"q\":\"golang\"}}]}\n```",
 		[]string{"search"},
+		nil,
 	)

 	outputText, _ := obj["output_text"].(string)
@@ -42,6 +46,7 @@ func TestBuildResponseObjectReasoningOnlyFallsBackToOutputText(t *testing.T) {
 		"internal thinking content",
 		"",
 		nil,
+		nil,
 	)

 	outputText, _ := obj["output_text"].(string)
@@ -75,6 +80,7 @@ func TestBuildResponseObjectPromotesToolCallFromThinkingWhenTextEmpty(t *testing
 		`<tool_calls><invoke name="search"><parameter name="q">from-thinking</parameter></invoke></tool_calls>`,
 		"",
 		[]string{"search"},
+		nil,
 	)

 	output, _ := obj["output"].([]any)
@@ -86,3 +92,88 @@ func TestBuildResponseObjectPromotesToolCallFromThinkingWhenTextEmpty(t *testing
 		t.Fatalf("expected function_call output, got %#v", first["type"])
 	}
 }
+
+func TestBuildChatCompletionWithToolCallsCoercesSchemaDeclaredStringArguments(t *testing.T) {
+	toolsRaw := []any{
+		map[string]any{
+			"type": "function",
+			"function": map[string]any{
+				"name": "Write",
+				"parameters": map[string]any{
+					"type": "object",
+					"properties": map[string]any{
+						"content": map[string]any{"type": "string"},
+						"taskId":  map[string]any{"type": "string"},
+					},
+				},
+			},
+		},
+	}
+	obj := BuildChatCompletionWithToolCalls(
+		"chat_test",
+		"gpt-4o",
+		"prompt",
+		"",
+		"",
+		[]toolcall.ParsedToolCall{{
+			Name: "Write",
+			Input: map[string]any{
+				"content": map[string]any{"message": "hi"},
+				"taskId":  1,
+			},
+		}},
+		toolsRaw,
+	)
+	choices, _ := obj["choices"].([]map[string]any)
+	message, _ := choices[0]["message"].(map[string]any)
+	toolCalls, _ := message["tool_calls"].([]map[string]any)
+	fn, _ := toolCalls[0]["function"].(map[string]any)
+	args := map[string]any{}
+	if err := json.Unmarshal([]byte(fn["arguments"].(string)), &args); err != nil {
+		t.Fatalf("decode arguments failed: %v", err)
+	}
+	if args["content"] != `{"message":"hi"}` {
+		t.Fatalf("expected content stringified by schema, got %#v", args["content"])
+	}
+	if args["taskId"] != "1" {
+		t.Fatalf("expected taskId stringified by schema, got %#v", args["taskId"])
+	}
+}
+
+func TestBuildResponseObjectWithToolCallsCoercesSchemaDeclaredStringArguments(t *testing.T) {
+	toolsRaw := []any{
+		map[string]any{
+			"type": "function",
+			"function": map[string]any{
+				"name": "Write",
+				"parameters": map[string]any{
+					"type": "object",
+					"properties": map[string]any{
+						"content": map[string]any{"type": "string"},
+					},
+				},
+			},
+		},
+	}
+	obj := BuildResponseObjectWithToolCalls(
+		"resp_test",
+		"gpt-4o",
+		"prompt",
+		"",
+		"",
+		[]toolcall.ParsedToolCall{{
+			Name:  "Write",
+			Input: map[string]any{"content": []any{"a", 1}},
+		}},
+		toolsRaw,
+	)
+	output, _ := obj["output"].([]any)
+	first, _ := output[0].(map[string]any)
+	args := map[string]any{}
+	if err := json.Unmarshal([]byte(first["arguments"].(string)), &args); err != nil {
+		t.Fatalf("decode response arguments failed: %v", err)
+	}
+	if args["content"] != `["a",1]` {
+		t.Fatalf("expected response content stringified by schema, got %#v", args["content"])
+	}
+}
--- a/internal/httpapi/claude/handler_helpers_misc.go
+++ b/internal/httpapi/claude/handler_helpers_misc.go
@@ -1,6 +1,7 @@
 package claude

 import (
+	"ds2api/internal/toolcall"
 	"fmt"
 	"strings"
 )
@@ -31,30 +32,9 @@ func extractClaudeToolNames(tools []any) []string {
 }

 func extractClaudeToolMeta(m map[string]any) (string, string, any) {
-	name, _ := m["name"].(string)
-	desc, _ := m["description"].(string)
-	schemaObj := m["input_schema"]
-	if schemaObj == nil {
-		schemaObj = m["parameters"]
-	}
-
-	if fn, ok := m["function"].(map[string]any); ok {
-		if strings.TrimSpace(name) == "" {
-			name, _ = fn["name"].(string)
-		}
-		if strings.TrimSpace(desc) == "" {
-			desc, _ = fn["description"].(string)
-		}
-		if schemaObj == nil {
-			if v, ok := fn["input_schema"]; ok {
-				schemaObj = v
-			}
-		}
-		if schemaObj == nil {
-			if v, ok := fn["parameters"]; ok {
-				schemaObj = v
-			}
-		}
+	name, desc, schemaObj := toolcall.ExtractToolMeta(m)
+	if strings.TrimSpace(desc) == "" {
+		desc = "No description available"
 	}
 	return strings.TrimSpace(name), strings.TrimSpace(desc), schemaObj
 }
--- a/internal/httpapi/claude/handler_messages.go
+++ b/internal/httpapi/claude/handler_messages.go
@@ -177,7 +177,7 @@ func stripClaudeThinkingBlocks(raw []byte) []byte {
 	return out
 }

-func (h *Handler) handleClaudeStreamRealtime(w http.ResponseWriter, r *http.Request, resp *http.Response, model string, messages []any, thinkingEnabled, searchEnabled bool, toolNames []string) {
+func (h *Handler) handleClaudeStreamRealtime(w http.ResponseWriter, r *http.Request, resp *http.Response, model string, messages []any, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any) {
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
@@ -205,6 +205,7 @@ func (h *Handler) handleClaudeStreamRealtime(w http.ResponseWriter, r *http.Requ
 		searchEnabled,
 		h.compatStripReferenceMarkers(),
 		toolNames,
+		toolsRaw,
 	)
 	streamRuntime.sendMessageStart()

--- a/internal/httpapi/claude/handler_stream_test.go
+++ b/internal/httpapi/claude/handler_stream_test.go
@@ -81,7 +81,7 @@ func TestHandleClaudeStreamRealtimeTextIncrementsWithEventHeaders(t *testing.T)
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil, nil)

 	body := rec.Body.String()
 	if !strings.Contains(body, "event: message_start") {
@@ -122,7 +122,7 @@ func TestHandleClaudeStreamRealtimeThinkingDelta(t *testing.T) {
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, true, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, true, false, nil, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	foundThinkingDelta := false
@@ -149,7 +149,7 @@ func TestHandleClaudeStreamRealtimeSkipsThinkingFallbackWhenFinalTextExists(t *t
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, true, false, []string{"search"})
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, true, false, []string{"search"}, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	for _, f := range findClaudeFrames(frames, "content_block_start") {
@@ -180,7 +180,7 @@ func TestHandleClaudeStreamRealtimeUpstreamErrorEvent(t *testing.T) {
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	errFrames := findClaudeFrames(frames, "error")
@@ -217,7 +217,7 @@ func TestHandleClaudeStreamRealtimePingEvent(t *testing.T) {

 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)
-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	if len(findClaudeFrames(frames, "ping")) == 0 {
@@ -271,7 +271,7 @@ func TestHandleClaudeStreamRealtimeToolSafetyAcrossStructuredFormats(t *testing.
 			rec := httptest.NewRecorder()
 			req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-			h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, false, false, []string{"Bash"})
+			h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, false, false, []string{"Bash"}, nil)

 			frames := parseClaudeFrames(t, rec.Body.String())
 			foundToolUse := false
@@ -299,7 +299,7 @@ func TestHandleClaudeStreamRealtimeDetectsToolUseWithLeadingProse(t *testing.T)
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, false, false, []string{"write_file"})
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, false, false, []string{"write_file"}, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	foundToolUse := false
@@ -333,7 +333,7 @@ func TestHandleClaudeStreamRealtimeIgnoresUnclosedFencedToolExample(t *testing.T
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "show example only"}}, false, false, []string{"Bash"})
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "show example only"}}, false, false, []string{"Bash"}, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	foundToolUse := false
@@ -365,3 +365,48 @@ func TestHandleClaudeStreamRealtimeIgnoresUnclosedFencedToolExample(t *testing.T
 func TestHandleClaudeStreamRealtimePromotesUnclosedFencedToolExample(t *testing.T) {
 	TestHandleClaudeStreamRealtimeIgnoresUnclosedFencedToolExample(t)
 }
+
+func TestHandleClaudeStreamRealtimeNormalizesToolInputBySchema(t *testing.T) {
+	h := &Handler{}
+	resp := makeClaudeSSEHTTPResponse(
+		`data: {"p":"response/content","v":"<tool_calls><invoke name=\"Write\">{\"input\":{\"content\":{\"message\":\"hi\"},\"taskId\":1}}</invoke></tool_calls>"}`,
+		`data: [DONE]`,
+	)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)
+	toolsRaw := []any{
+		map[string]any{
+			"name": "Write",
+			"inputSchema": map[string]any{
+				"type": "object",
+				"properties": map[string]any{
+					"content": map[string]any{"type": "string"},
+					"taskId":  map[string]any{"type": "string"},
+				},
+			},
+		},
+	}
+
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "write"}}, false, false, []string{"Write"}, toolsRaw)
+
+	frames := parseClaudeFrames(t, rec.Body.String())
+	for _, f := range findClaudeFrames(frames, "content_block_delta") {
+		delta, _ := f.Payload["delta"].(map[string]any)
+		if delta["type"] != "input_json_delta" {
+			continue
+		}
+		partial := asString(delta["partial_json"])
+		var args map[string]any
+		if err := json.Unmarshal([]byte(partial), &args); err != nil {
+			t.Fatalf("decode partial_json failed: %v payload=%s", err, partial)
+		}
+		if args["content"] != `{"message":"hi"}` {
+			t.Fatalf("expected content normalized to string, got %#v", args["content"])
+		}
+		if args["taskId"] != "1" {
+			t.Fatalf("expected taskId normalized to string, got %#v", args["taskId"])
+		}
+		return
+	}
+	t.Fatalf("expected input_json_delta frame, body=%s", rec.Body.String())
+}
--- a/internal/httpapi/claude/standard_request.go
+++ b/internal/httpapi/claude/standard_request.go
@@ -53,6 +53,7 @@ func normalizeClaudeRequest(store ConfigReader, req map[string]any) (claudeNorma
 			ResolvedModel:  dsModel,
 			ResponseModel:  strings.TrimSpace(model),
 			Messages:       payload["messages"].([]any),
+			ToolsRaw:       toolsRequested,
 			FinalPrompt:    finalPrompt,
 			ToolNames:      toolNames,
 			Stream:         util.ToBool(req["stream"]),
--- a/internal/httpapi/claude/standard_request_test.go
+++ b/internal/httpapi/claude/standard_request_test.go
@@ -32,11 +32,39 @@ func TestNormalizeClaudeRequest(t *testing.T) {
 	if len(norm.Standard.ToolNames) == 0 {
 		t.Fatalf("expected tool names")
 	}
+	if norm.Standard.ToolsRaw == nil {
+		t.Fatalf("expected ToolsRaw preserved for downstream normalization")
+	}
 	if norm.Standard.FinalPrompt == "" {
 		t.Fatalf("expected non-empty final prompt")
 	}
 }

+func TestNormalizeClaudeRequestSupportsCamelCaseInputSchemaPromptInjection(t *testing.T) {
+	t.Setenv("DS2API_CONFIG_JSON", `{}`)
+	store := config.LoadStore()
+	req := map[string]any{
+		"model": "claude-sonnet-4-5",
+		"messages": []any{
+			map[string]any{"role": "user", "content": "hello"},
+		},
+		"tools": []any{
+			map[string]any{
+				"name":        "todowrite",
+				"description": "Write todos",
+				"inputSchema": map[string]any{"type": "object", "properties": map[string]any{"todos": map[string]any{"type": "array"}}},
+			},
+		},
+	}
+	norm, err := normalizeClaudeRequest(store, req)
+	if err != nil {
+		t.Fatalf("normalize failed: %v", err)
+	}
+	if !containsStr(norm.Standard.FinalPrompt, `"type":"array"`) {
+		t.Fatalf("expected inputSchema to be injected into prompt, got=%q", norm.Standard.FinalPrompt)
+	}
+}
+
 func TestNormalizeClaudeRequestInjectsToolsIntoExistingSystemMessage(t *testing.T) {
 	t.Setenv("DS2API_CONFIG_JSON", `{}`)
 	store := config.LoadStore()
--- a/internal/httpapi/claude/stream_runtime_core.go
+++ b/internal/httpapi/claude/stream_runtime_core.go
@@ -18,6 +18,7 @@ type claudeStreamRuntime struct {
 	model     string
 	toolNames []string
 	messages  []any
+	toolsRaw  any

 	thinkingEnabled       bool
 	searchEnabled         bool
@@ -47,6 +48,7 @@ func newClaudeStreamRuntime(
 	searchEnabled bool,
 	stripReferenceMarkers bool,
 	toolNames []string,
+	toolsRaw any,
 ) *claudeStreamRuntime {
 	return &claudeStreamRuntime{
 		w:                     w,
@@ -59,6 +61,7 @@ func newClaudeStreamRuntime(
 		bufferToolContent:     len(toolNames) > 0,
 		stripReferenceMarkers: stripReferenceMarkers,
 		toolNames:             toolNames,
+		toolsRaw:              toolsRaw,
 		messageID:             fmt.Sprintf("msg_%d", time.Now().UnixNano()),
 		thinkingBlockIndex:    -1,
 		textBlockIndex:        -1,
--- a/internal/httpapi/claude/stream_runtime_finalize.go
+++ b/internal/httpapi/claude/stream_runtime_finalize.go
@@ -52,6 +52,7 @@ func (s *claudeStreamRuntime) finalize(stopReason string) {
 			detected = toolcall.ParseStandaloneToolCalls(finalThinking, s.toolNames)
 		}
 		if len(detected) > 0 {
+			detected = toolcall.NormalizeParsedToolCallsForSchemas(detected, s.toolsRaw)
 			stopReason = "tool_use"
 			for i, tc := range detected {
 				idx := s.nextBlockIndex + i
--- a/internal/httpapi/openai/chat/chat_history_test.go
+++ b/internal/httpapi/openai/chat/chat_history_test.go
@@ -194,7 +194,7 @@ func TestHandleStreamContextCancelledMarksHistoryStopped(t *testing.T) {
 	rec := httptest.NewRecorder()
 	resp := makeOpenAISSEHTTPResponse(`data: {"p":"response/content","v":"hello"}`, `data: [DONE]`)

-	h.handleStream(rec, req, resp, "cid-stop", "deepseek-v4-flash", "prompt", false, false, nil, session)
+	h.handleStream(rec, req, resp, "cid-stop", "deepseek-v4-flash", "prompt", false, false, nil, nil, session)

 	snapshot, err := historyStore.Snapshot()
 	if err != nil {
@@ -307,19 +307,19 @@ func TestChatCompletionsCurrentInputFilePersistsNeutralPrompt(t *testing.T) {
 	if err != nil {
 		t.Fatalf("expected detail item, got %v", err)
 	}
-	if full.HistoryText != "" {
-		t.Fatalf("expected current input file flow to leave history text empty, got %q", full.HistoryText)
-	}
 	if len(ds.uploadCalls) != 1 {
 		t.Fatalf("expected current input upload to happen, got %d", len(ds.uploadCalls))
 	}
 	if ds.uploadCalls[0].Filename != "IGNORE.txt" {
 		t.Fatalf("expected IGNORE.txt upload, got %q", ds.uploadCalls[0].Filename)
 	}
-	if len(full.Messages) != 1 {
-		t.Fatalf("expected compacted-context prompt to be the only persisted message, got %#v", full.Messages)
+	if full.HistoryText != string(ds.uploadCalls[0].Data) {
+		t.Fatalf("expected uploaded current input file to be persisted in history text")
 	}
-	if !strings.Contains(full.Messages[0].Content, promptcompat.BuildOpenAICurrentInputContextPrompt()) {
-		t.Fatalf("expected compacted-context prompt to be persisted, got %#v", full.Messages[0])
+	if len(full.Messages) != 1 {
+		t.Fatalf("expected neutral prompt to be the only persisted message, got %#v", full.Messages)
+	}
+	if !strings.Contains(full.Messages[0].Content, "Answer the latest user request directly.") {
+		t.Fatalf("expected neutral prompt to be persisted, got %#v", full.Messages[0])
 	}
 }
--- a/internal/httpapi/openai/chat/chat_stream_runtime.go
+++ b/internal/httpapi/openai/chat/chat_stream_runtime.go
@@ -21,6 +21,7 @@ type chatStreamRuntime struct {
 	model        string
 	finalPrompt  string
 	toolNames    []string
+	toolsRaw     any

 	thinkingEnabled       bool
 	searchEnabled         bool
@@ -61,6 +62,7 @@ func newChatStreamRuntime(
 	searchEnabled bool,
 	stripReferenceMarkers bool,
 	toolNames []string,
+	toolsRaw any,
 	bufferToolContent bool,
 	emitEarlyToolDeltas bool,
 ) *chatStreamRuntime {
@@ -73,6 +75,7 @@ func newChatStreamRuntime(
 		model:                 model,
 		finalPrompt:           finalPrompt,
 		toolNames:             toolNames,
+		toolsRaw:              toolsRaw,
 		thinkingEnabled:       thinkingEnabled,
 		searchEnabled:         searchEnabled,
 		stripReferenceMarkers: stripReferenceMarkers,
@@ -142,7 +145,7 @@ func (s *chatStreamRuntime) finalize(finishReason string, deferEmptyOutput bool)
 	if len(detected.Calls) > 0 && !s.toolCallsDoneEmitted {
 		finishReason = "tool_calls"
 		delta := map[string]any{
-			"tool_calls": formatFinalStreamToolCallsWithStableIDs(detected.Calls, s.streamToolCallIDs),
+			"tool_calls": formatFinalStreamToolCallsWithStableIDs(detected.Calls, s.streamToolCallIDs, s.toolsRaw),
 		}
 		if !s.firstChunkSent {
 			delta["role"] = "assistant"
@@ -164,7 +167,7 @@ func (s *chatStreamRuntime) finalize(finishReason string, deferEmptyOutput bool)
 				s.toolCallsEmitted = true
 				s.toolCallsDoneEmitted = true
 				tcDelta := map[string]any{
-					"tool_calls": formatFinalStreamToolCallsWithStableIDs(evt.ToolCalls, s.streamToolCallIDs),
+					"tool_calls": formatFinalStreamToolCallsWithStableIDs(evt.ToolCalls, s.streamToolCallIDs, s.toolsRaw),
 				}
 				if !s.firstChunkSent {
 					tcDelta["role"] = "assistant"
@@ -320,7 +323,7 @@ func (s *chatStreamRuntime) onParsed(parsed sse.LineResult) streamengine.ParsedD
 						s.toolCallsEmitted = true
 						s.toolCallsDoneEmitted = true
 						tcDelta := map[string]any{
-							"tool_calls": formatFinalStreamToolCallsWithStableIDs(evt.ToolCalls, s.streamToolCallIDs),
+							"tool_calls": formatFinalStreamToolCallsWithStableIDs(evt.ToolCalls, s.streamToolCallIDs, s.toolsRaw),
 						}
 						if !s.firstChunkSent {
 							tcDelta["role"] = "assistant"
--- a/internal/httpapi/openai/chat/empty_retry_runtime.go
+++ b/internal/httpapi/openai/chat/empty_retry_runtime.go
@@ -26,14 +26,14 @@ type chatNonStreamResult struct {
 	responseMessageID     int
 }

-func (h *Handler) handleNonStreamWithRetry(w http.ResponseWriter, ctx context.Context, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, historySession *chatHistorySession) {
+func (h *Handler) handleNonStreamWithRetry(w http.ResponseWriter, ctx context.Context, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, historySession *chatHistorySession) {
 	attempts := 0
 	currentResp := resp
 	usagePrompt := finalPrompt
 	accumulatedThinking := ""
 	accumulatedToolDetectionThinking := ""
 	for {
-		result, ok := h.collectChatNonStreamAttempt(w, currentResp, completionID, model, usagePrompt, thinkingEnabled, searchEnabled, toolNames)
+		result, ok := h.collectChatNonStreamAttempt(w, currentResp, completionID, model, usagePrompt, thinkingEnabled, searchEnabled, toolNames, toolsRaw)
 		if !ok {
 			return
 		}
@@ -43,7 +43,7 @@ func (h *Handler) handleNonStreamWithRetry(w http.ResponseWriter, ctx context.Co
 		result.toolDetectionThinking = accumulatedToolDetectionThinking
 		detected := detectAssistantToolCalls(result.text, result.thinking, result.toolDetectionThinking, toolNames)
 		result.detectedCalls = len(detected.Calls)
-		result.body = openaifmt.BuildChatCompletionWithToolCalls(completionID, model, usagePrompt, result.thinking, result.text, detected.Calls)
+		result.body = openaifmt.BuildChatCompletionWithToolCalls(completionID, model, usagePrompt, result.thinking, result.text, detected.Calls, toolsRaw)
 		result.finishReason = chatFinishReason(result.body)
 		if !shouldRetryChatNonStream(result, attempts) {
 			h.finishChatNonStreamResult(w, result, attempts, usagePrompt, historySession)
@@ -72,7 +72,7 @@ func (h *Handler) handleNonStreamWithRetry(w http.ResponseWriter, ctx context.Co
 	}
 }

-func (h *Handler) collectChatNonStreamAttempt(w http.ResponseWriter, resp *http.Response, completionID, model, usagePrompt string, thinkingEnabled, searchEnabled bool, toolNames []string) (chatNonStreamResult, bool) {
+func (h *Handler) collectChatNonStreamAttempt(w http.ResponseWriter, resp *http.Response, completionID, model, usagePrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any) (chatNonStreamResult, bool) {
 	if resp.StatusCode != http.StatusOK {
 		defer func() { _ = resp.Body.Close() }()
 		body, _ := io.ReadAll(resp.Body)
@@ -88,7 +88,7 @@ func (h *Handler) collectChatNonStreamAttempt(w http.ResponseWriter, resp *http.
 		finalText = replaceCitationMarkersWithLinks(finalText, result.CitationLinks)
 	}
 	detected := detectAssistantToolCalls(finalText, finalThinking, finalToolDetectionThinking, toolNames)
-	respBody := openaifmt.BuildChatCompletionWithToolCalls(completionID, model, usagePrompt, finalThinking, finalText, detected.Calls)
+	respBody := openaifmt.BuildChatCompletionWithToolCalls(completionID, model, usagePrompt, finalThinking, finalText, detected.Calls, toolsRaw)
 	return chatNonStreamResult{
 		thinking:              finalThinking,
 		toolDetectionThinking: finalToolDetectionThinking,
@@ -139,8 +139,8 @@ func shouldRetryChatNonStream(result chatNonStreamResult, attempts int) bool {
 		strings.TrimSpace(result.text) == ""
 }

-func (h *Handler) handleStreamWithRetry(w http.ResponseWriter, r *http.Request, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, historySession *chatHistorySession) {
-	streamRuntime, initialType, ok := h.prepareChatStreamRuntime(w, resp, completionID, model, finalPrompt, thinkingEnabled, searchEnabled, toolNames, historySession)
+func (h *Handler) handleStreamWithRetry(w http.ResponseWriter, r *http.Request, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, historySession *chatHistorySession) {
+	streamRuntime, initialType, ok := h.prepareChatStreamRuntime(w, resp, completionID, model, finalPrompt, thinkingEnabled, searchEnabled, toolNames, toolsRaw, historySession)
 	if !ok {
 		return
 	}
@@ -182,7 +182,7 @@ func (h *Handler) handleStreamWithRetry(w http.ResponseWriter, r *http.Request,
 	}
 }

-func (h *Handler) prepareChatStreamRuntime(w http.ResponseWriter, resp *http.Response, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, historySession *chatHistorySession) (*chatStreamRuntime, string, bool) {
+func (h *Handler) prepareChatStreamRuntime(w http.ResponseWriter, resp *http.Response, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, historySession *chatHistorySession) (*chatStreamRuntime, string, bool) {
 	if resp.StatusCode != http.StatusOK {
 		defer func() { _ = resp.Body.Close() }()
 		body, _ := io.ReadAll(resp.Body)
@@ -207,7 +207,7 @@ func (h *Handler) prepareChatStreamRuntime(w http.ResponseWriter, resp *http.Res
 	}
 	streamRuntime := newChatStreamRuntime(
 		w, rc, canFlush, completionID, time.Now().Unix(), model, finalPrompt,
-		thinkingEnabled, searchEnabled, h.compatStripReferenceMarkers(), toolNames,
+		thinkingEnabled, searchEnabled, h.compatStripReferenceMarkers(), toolNames, toolsRaw,
 		len(toolNames) > 0, h.toolcallFeatureMatchEnabled() && h.toolcallEarlyEmitHighConfidence(),
 	)
 	return streamRuntime, initialType, true
--- a/internal/httpapi/openai/chat/handler.go
+++ b/internal/httpapi/openai/chat/handler.go
@@ -144,8 +144,8 @@ func filterIncrementalToolCallDeltasByAllowed(deltas []toolstream.ToolCallDelta,
 	return shared.FilterIncrementalToolCallDeltasByAllowed(deltas, seenNames)
 }

-func formatFinalStreamToolCallsWithStableIDs(calls []toolcall.ParsedToolCall, ids map[int]string) []map[string]any {
-	return shared.FormatFinalStreamToolCallsWithStableIDs(calls, ids)
+func formatFinalStreamToolCallsWithStableIDs(calls []toolcall.ParsedToolCall, ids map[int]string, toolsRaw any) []map[string]any {
+	return shared.FormatFinalStreamToolCallsWithStableIDs(calls, ids, toolsRaw)
 }

 func detectAssistantToolCalls(text, exposedThinking, detectionThinking string, toolNames []string) toolcall.ToolCallParseResult {
--- a/internal/httpapi/openai/chat/handler_chat.go
+++ b/internal/httpapi/openai/chat/handler_chat.go
@@ -109,10 +109,10 @@ func (h *Handler) ChatCompletions(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	if stdReq.Stream {
-		h.handleStreamWithRetry(w, r, a, resp, payload, pow, sessionID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, historySession)
+		h.handleStreamWithRetry(w, r, a, resp, payload, pow, sessionID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, stdReq.ToolsRaw, historySession)
 		return
 	}
-	h.handleNonStreamWithRetry(w, r.Context(), a, resp, payload, pow, sessionID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, historySession)
+	h.handleNonStreamWithRetry(w, r.Context(), a, resp, payload, pow, sessionID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, stdReq.ToolsRaw, historySession)
 }

 func (h *Handler) autoDeleteRemoteSession(ctx context.Context, a *auth.RequestAuth, sessionID string) {
@@ -148,7 +148,7 @@ func (h *Handler) autoDeleteRemoteSession(ctx context.Context, a *auth.RequestAu
 	}
 }

-func (h *Handler) handleNonStream(w http.ResponseWriter, resp *http.Response, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, historySession *chatHistorySession) {
+func (h *Handler) handleNonStream(w http.ResponseWriter, resp *http.Response, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, historySession *chatHistorySession) {
 	if resp.StatusCode != http.StatusOK {
 		defer func() { _ = resp.Body.Close() }()
 		body, _ := io.ReadAll(resp.Body)
@@ -176,7 +176,7 @@ func (h *Handler) handleNonStream(w http.ResponseWriter, resp *http.Response, co
 		writeUpstreamEmptyOutputError(w, finalText, finalThinking, result.ContentFilter)
 		return
 	}
-	respBody := openaifmt.BuildChatCompletionWithToolCalls(completionID, model, finalPrompt, finalThinking, finalText, detected.Calls)
+	respBody := openaifmt.BuildChatCompletionWithToolCalls(completionID, model, finalPrompt, finalThinking, finalText, detected.Calls, toolsRaw)
 	finishReason := "stop"
 	if choices, ok := respBody["choices"].([]map[string]any); ok && len(choices) > 0 {
 		if fr, _ := choices[0]["finish_reason"].(string); strings.TrimSpace(fr) != "" {
@@ -189,7 +189,7 @@ func (h *Handler) handleNonStream(w http.ResponseWriter, resp *http.Response, co
 	writeJSON(w, http.StatusOK, respBody)
 }

-func (h *Handler) handleStream(w http.ResponseWriter, r *http.Request, resp *http.Response, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, historySession *chatHistorySession) {
+func (h *Handler) handleStream(w http.ResponseWriter, r *http.Request, resp *http.Response, completionID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, historySession *chatHistorySession) {
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
@@ -230,6 +230,7 @@ func (h *Handler) handleStream(w http.ResponseWriter, r *http.Request, resp *htt
 		searchEnabled,
 		stripReferenceMarkers,
 		toolNames,
+		toolsRaw,
 		bufferToolContent,
 		emitEarlyToolDeltas,
 	)
--- a/internal/httpapi/openai/chat/handler_toolcall_test.go
+++ b/internal/httpapi/openai/chat/handler_toolcall_test.go
@@ -93,7 +93,7 @@ func TestHandleNonStreamReturns429WhenUpstreamOutputEmpty(t *testing.T) {
 	)
 	rec := httptest.NewRecorder()

-	h.handleNonStream(rec, resp, "cid-empty", "deepseek-v4-flash", "prompt", false, false, nil, nil)
+	h.handleNonStream(rec, resp, "cid-empty", "deepseek-v4-flash", "prompt", false, false, nil, nil, nil)
 	if rec.Code != http.StatusTooManyRequests {
 		t.Fatalf("expected status 429 for empty upstream output, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -112,7 +112,7 @@ func TestHandleNonStreamReturnsContentFilterErrorWhenUpstreamFilteredWithoutOutp
 	)
 	rec := httptest.NewRecorder()

-	h.handleNonStream(rec, resp, "cid-empty-filtered", "deepseek-v4-flash", "prompt", false, false, nil, nil)
+	h.handleNonStream(rec, resp, "cid-empty-filtered", "deepseek-v4-flash", "prompt", false, false, nil, nil, nil)
 	if rec.Code != http.StatusBadRequest {
 		t.Fatalf("expected status 400 for filtered upstream output, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -131,7 +131,7 @@ func TestHandleNonStreamReturns429WhenUpstreamHasOnlyThinking(t *testing.T) {
 	)
 	rec := httptest.NewRecorder()

-	h.handleNonStream(rec, resp, "cid-thinking-only", "deepseek-v4-pro", "prompt", true, false, nil, nil)
+	h.handleNonStream(rec, resp, "cid-thinking-only", "deepseek-v4-pro", "prompt", true, false, nil, nil, nil)
 	if rec.Code != http.StatusTooManyRequests {
 		t.Fatalf("expected status 429 for thinking-only upstream output, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -150,7 +150,7 @@ func TestHandleNonStreamPromotesThinkingToolCallsWhenTextEmpty(t *testing.T) {
 	)
 	rec := httptest.NewRecorder()

-	h.handleNonStream(rec, resp, "cid-thinking-tool", "deepseek-v4-pro", "prompt", true, false, []string{"search"}, nil)
+	h.handleNonStream(rec, resp, "cid-thinking-tool", "deepseek-v4-pro", "prompt", true, false, []string{"search"}, nil, nil)
 	if rec.Code != http.StatusOK {
 		t.Fatalf("expected 200 for thinking tool calls, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -181,7 +181,7 @@ func TestHandleNonStreamPromotesHiddenThinkingDSMLToolCallsWhenTextEmpty(t *test
 	)
 	rec := httptest.NewRecorder()

-	h.handleNonStream(rec, resp, "cid-hidden-thinking-tool", "deepseek-v4-pro", "prompt", false, false, []string{"search"}, nil)
+	h.handleNonStream(rec, resp, "cid-hidden-thinking-tool", "deepseek-v4-pro", "prompt", false, false, []string{"search"}, nil, nil)
 	if rec.Code != http.StatusOK {
 		t.Fatalf("expected 200 for hidden thinking tool calls, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -211,7 +211,7 @@ func TestHandleStreamToolsPlainTextStreamsBeforeFinish(t *testing.T) {
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)

-	h.handleStream(rec, req, resp, "cid6", "deepseek-v4-flash", "prompt", false, false, []string{"search"}, nil)
+	h.handleStream(rec, req, resp, "cid6", "deepseek-v4-flash", "prompt", false, false, []string{"search"}, nil, nil)

 	frames, done := parseSSEDataFrames(t, rec.Body.String())
 	if !done {
@@ -248,7 +248,7 @@ func TestHandleStreamIncompleteCapturedToolJSONFlushesAsTextOnFinalize(t *testin
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)

-	h.handleStream(rec, req, resp, "cid10", "deepseek-v4-flash", "prompt", false, false, []string{"search"}, nil)
+	h.handleStream(rec, req, resp, "cid10", "deepseek-v4-flash", "prompt", false, false, []string{"search"}, nil, nil)

 	frames, done := parseSSEDataFrames(t, rec.Body.String())
 	if !done {
@@ -282,7 +282,7 @@ func TestHandleStreamPromotesThinkingToolCallsOnFinalizeWithoutMidstreamIntercep
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)

-	h.handleStream(rec, req, resp, "cid-thinking-stream", "deepseek-v4-pro", "prompt", true, false, []string{"search"}, nil)
+	h.handleStream(rec, req, resp, "cid-thinking-stream", "deepseek-v4-pro", "prompt", true, false, []string{"search"}, nil, nil)

 	frames, done := parseSSEDataFrames(t, rec.Body.String())
 	if !done {
@@ -319,7 +319,7 @@ func TestHandleStreamPromotesHiddenThinkingDSMLToolCallsOnFinalize(t *testing.T)
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)

-	h.handleStream(rec, req, resp, "cid-hidden-thinking-stream", "deepseek-v4-pro", "prompt", false, false, []string{"search"}, nil)
+	h.handleStream(rec, req, resp, "cid-hidden-thinking-stream", "deepseek-v4-pro", "prompt", false, false, []string{"search"}, nil, nil)

 	frames, done := parseSSEDataFrames(t, rec.Body.String())
 	if !done {
@@ -353,7 +353,7 @@ func TestHandleStreamEmitsDistinctToolCallIDsAcrossSeparateToolBlocks(t *testing
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)

-	h.handleStream(rec, req, resp, "cid-multi", "deepseek-v4-flash", "prompt", false, false, []string{"read_file", "search"}, nil)
+	h.handleStream(rec, req, resp, "cid-multi", "deepseek-v4-flash", "prompt", false, false, []string{"read_file", "search"}, nil, nil)

 	frames, done := parseSSEDataFrames(t, rec.Body.String())
 	if !done {
@@ -390,3 +390,64 @@ func TestHandleStreamEmitsDistinctToolCallIDsAcrossSeparateToolBlocks(t *testing
 		t.Fatalf("expected distinct tool call ids across blocks, got %#v body=%s", ids, rec.Body.String())
 	}
 }
+
+func TestHandleStreamCoercesSchemaDeclaredStringArgumentsOnFinalize(t *testing.T) {
+	h := &Handler{}
+	line := func(v string) string {
+		b, _ := json.Marshal(map[string]any{"p": "response/content", "v": v})
+		return "data: " + string(b)
+	}
+	resp := makeSSEHTTPResponse(
+		line(`<tool_calls><invoke name="Write">{"input":{"content":{"message":"hi"},"taskId":1}}</invoke></tool_calls>`),
+		`data: [DONE]`,
+	)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
+	toolsRaw := []any{
+		map[string]any{
+			"type": "function",
+			"function": map[string]any{
+				"name": "Write",
+				"parameters": map[string]any{
+					"type": "object",
+					"properties": map[string]any{
+						"content": map[string]any{"type": "string"},
+						"taskId":  map[string]any{"type": "string"},
+					},
+				},
+			},
+		},
+	}
+
+	h.handleStream(rec, req, resp, "cid-string-protect", "deepseek-v4-flash", "prompt", false, false, []string{"Write"}, toolsRaw, nil)
+
+	frames, done := parseSSEDataFrames(t, rec.Body.String())
+	if !done {
+		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
+	}
+	for _, frame := range frames {
+		choices, _ := frame["choices"].([]any)
+		for _, item := range choices {
+			choice, _ := item.(map[string]any)
+			delta, _ := choice["delta"].(map[string]any)
+			toolCalls, _ := delta["tool_calls"].([]any)
+			if len(toolCalls) == 0 {
+				continue
+			}
+			call, _ := toolCalls[0].(map[string]any)
+			fn, _ := call["function"].(map[string]any)
+			args := map[string]any{}
+			if err := json.Unmarshal([]byte(asString(fn["arguments"])), &args); err != nil {
+				t.Fatalf("decode streamed tool arguments failed: %v", err)
+			}
+			if args["content"] != `{"message":"hi"}` {
+				t.Fatalf("expected streamed content stringified by schema, got %#v", args["content"])
+			}
+			if args["taskId"] != "1" {
+				t.Fatalf("expected streamed taskId stringified by schema, got %#v", args["taskId"])
+			}
+			return
+		}
+	}
+	t.Fatalf("expected at least one streamed tool call delta, body=%s", rec.Body.String())
+}
--- a/internal/httpapi/openai/chat/vercel_prepare_test.go
+++ b/internal/httpapi/openai/chat/vercel_prepare_test.go
@@ -10,7 +10,6 @@ import (

 	"ds2api/internal/auth"
 	dsclient "ds2api/internal/deepseek/client"
-	"ds2api/internal/promptcompat"
 )

 func TestIsVercelStreamPrepareRequest(t *testing.T) {
@@ -131,8 +130,8 @@ func TestHandleVercelStreamPrepareAppliesCurrentInputFile(t *testing.T) {
 		t.Fatalf("expected payload object, got %#v", body["payload"])
 	}
 	promptText, _ := payload["prompt"].(string)
-	if !strings.Contains(promptText, promptcompat.BuildOpenAICurrentInputContextPrompt()) {
-		t.Fatalf("expected compacted-context prompt, got %s", promptText)
+	if !strings.Contains(promptText, "Answer the latest user request directly.") {
+		t.Fatalf("expected neutral prompt, got %s", promptText)
 	}
 	if strings.Contains(promptText, "first user turn") || strings.Contains(promptText, "latest user turn") {
 		t.Fatalf("expected original turns hidden from prompt, got %s", promptText)
--- a/internal/httpapi/openai/citation_links_test.go
+++ b/internal/httpapi/openai/citation_links_test.go
@@ -26,3 +26,31 @@ func TestReplaceCitationMarkersWithLinksKeepsUnknownIndex(t *testing.T) {
 		t.Fatalf("expected %q, got %q", want, got)
 	}
 }
+
+func TestReplaceCitationMarkersWithLinksSupportsReferenceMarker(t *testing.T) {
+	raw := "新闻摘要[reference:1]，详情[reference:2]。"
+	links := map[int]string{
+		1: "https://example.com/r1",
+		2: "https://example.com/r2",
+	}
+
+	got := replaceCitationMarkersWithLinks(raw, links)
+	want := "新闻摘要[1](https://example.com/r1)，详情[2](https://example.com/r2)。"
+	if got != want {
+		t.Fatalf("expected %q, got %q", want, got)
+	}
+}
+
+func TestReplaceCitationMarkersWithLinksSupportsReferenceZeroBased(t *testing.T) {
+	raw := "来源[reference:0] 与 [reference:1]。"
+	links := map[int]string{
+		1: "https://example.com/first",
+		2: "https://example.com/second",
+	}
+
+	got := replaceCitationMarkersWithLinks(raw, links)
+	want := "来源[0](https://example.com/first) 与 [1](https://example.com/second)。"
+	if got != want {
+		t.Fatalf("expected %q, got %q", want, got)
+	}
+}
--- a/internal/httpapi/openai/history/current_input_file.go
+++ b/internal/httpapi/openai/history/current_input_file.go
@@ -58,6 +58,7 @@ func (s Service) ApplyCurrentInputFile(ctx context.Context, a *auth.RequestAuth,
 	}

 	stdReq.Messages = messages
+	stdReq.HistoryText = fileText
 	stdReq.CurrentInputFileApplied = true
 	stdReq.RefFileIDs = prependUniqueRefFileID(stdReq.RefFileIDs, fileID)
 	stdReq.FinalPrompt, stdReq.ToolNames = promptcompat.BuildOpenAIPrompt(messages, stdReq.ToolsRaw, "", stdReq.ToolChoice, stdReq.Thinking)
@@ -84,5 +85,5 @@ func latestUserInputForFile(messages []any) (int, string) {
 }

 func currentInputFilePrompt() string {
-	return promptcompat.BuildOpenAICurrentInputContextPrompt()
+	return "The current request and prior conversation context have already been provided. Answer the latest user request directly."
 }
--- a/internal/httpapi/openai/history_split_test.go
+++ b/internal/httpapi/openai/history_split_test.go
@@ -67,9 +67,6 @@ func TestBuildOpenAICurrentInputContextTranscriptUsesInjectedFileWrapper(t *test
 	if !strings.HasPrefix(transcript, "[file content end]\n\n") {
 		t.Fatalf("expected injected file wrapper prefix, got %q", transcript)
 	}
-	if !strings.Contains(transcript, "[context note]") || !strings.Contains(transcript, "compacted snapshot of the prior conversation history") {
-		t.Fatalf("expected compacted context note in transcript, got %q", transcript)
-	}
 	if !strings.Contains(transcript, "<｜begin▁of▁sentence｜>") {
 		t.Fatalf("expected serialized conversation markers, got %q", transcript)
 	}
@@ -299,8 +296,8 @@ func TestApplyCurrentInputFileUploadsFirstTurnWithInjectedWrapper(t *testing.T)
 	if strings.Contains(out.FinalPrompt, "CURRENT_USER_INPUT.txt") || strings.Contains(out.FinalPrompt, "IGNORE.txt") || strings.Contains(out.FinalPrompt, "Read that file") {
 		t.Fatalf("expected live prompt not to instruct file reads, got %s", out.FinalPrompt)
 	}
-	if !strings.Contains(out.FinalPrompt, promptcompat.BuildOpenAICurrentInputContextPrompt()) {
-		t.Fatalf("expected compacted-context instruction in live prompt, got %s", out.FinalPrompt)
+	if !strings.Contains(out.FinalPrompt, "Answer the latest user request directly.") {
+		t.Fatalf("expected neutral continuation instruction in live prompt, got %s", out.FinalPrompt)
 	}
 	if len(out.RefFileIDs) != 1 || out.RefFileIDs[0] != "file-inline-1" {
 		t.Fatalf("expected current input file id in ref_file_ids, got %#v", out.RefFileIDs)
@@ -348,14 +345,14 @@ func TestApplyCurrentInputFileUploadsFullContextFile(t *testing.T) {
 		}
 	}
 	if strings.Contains(out.FinalPrompt, "first user turn") || strings.Contains(out.FinalPrompt, "latest user turn") || strings.Contains(out.FinalPrompt, "CURRENT_USER_INPUT.txt") || strings.Contains(out.FinalPrompt, "IGNORE.txt") || strings.Contains(out.FinalPrompt, "Read that file") {
-		t.Fatalf("expected live prompt to stay in compacted-context mode, got %s", out.FinalPrompt)
+		t.Fatalf("expected live prompt to use only a neutral continuation instruction, got %s", out.FinalPrompt)
 	}
-	if !strings.Contains(out.FinalPrompt, promptcompat.BuildOpenAICurrentInputContextPrompt()) {
-		t.Fatalf("expected compacted-context instruction in live prompt, got %s", out.FinalPrompt)
+	if !strings.Contains(out.FinalPrompt, "Answer the latest user request directly.") {
+		t.Fatalf("expected neutral continuation instruction in live prompt, got %s", out.FinalPrompt)
 	}
 }

-func TestApplyCurrentInputFileLeavesHistoryTextEmpty(t *testing.T) {
+func TestApplyCurrentInputFileCarriesHistoryText(t *testing.T) {
 	ds := &inlineUploadDSStub{}
 	h := &openAITestSurface{
 		Store: mockOpenAIConfig{
@@ -380,8 +377,8 @@ func TestApplyCurrentInputFileLeavesHistoryTextEmpty(t *testing.T) {
 	if len(ds.uploadCalls) != 1 {
 		t.Fatalf("expected 1 upload call, got %d", len(ds.uploadCalls))
 	}
-	if out.HistoryText != "" {
-		t.Fatalf("expected current input file flow to leave history text empty, got %q", out.HistoryText)
+	if out.HistoryText != string(ds.uploadCalls[0].Data) {
+		t.Fatalf("expected current input file flow to preserve uploaded text in history, got %q", out.HistoryText)
 	}
 }

@@ -431,8 +428,8 @@ func TestChatCompletionsCurrentInputFileUploadsContextAndKeepsNeutralPrompt(t *t
 		t.Fatal("expected completion payload to be captured")
 	}
 	promptText, _ := ds.completionReq["prompt"].(string)
-	if !strings.Contains(promptText, promptcompat.BuildOpenAICurrentInputContextPrompt()) {
-		t.Fatalf("expected compacted-context prompt, got %s", promptText)
+	if !strings.Contains(promptText, "Answer the latest user request directly.") {
+		t.Fatalf("expected neutral completion prompt, got %s", promptText)
 	}
 	if strings.Contains(promptText, "first user turn") || strings.Contains(promptText, "latest user turn") {
 		t.Fatalf("expected prompt to hide original turns, got %s", promptText)
@@ -477,8 +474,8 @@ func TestResponsesCurrentInputFileUploadsContextAndKeepsNeutralPrompt(t *testing
 		t.Fatal("expected completion payload to be captured")
 	}
 	promptText, _ := ds.completionReq["prompt"].(string)
-	if !strings.Contains(promptText, promptcompat.BuildOpenAICurrentInputContextPrompt()) {
-		t.Fatalf("expected compacted-context prompt, got %s", promptText)
+	if !strings.Contains(promptText, "Answer the latest user request directly.") {
+		t.Fatalf("expected neutral completion prompt, got %s", promptText)
 	}
 	if strings.Contains(promptText, "first user turn") || strings.Contains(promptText, "latest user turn") {
 		t.Fatalf("expected prompt to hide original turns, got %s", promptText)
@@ -613,7 +610,7 @@ func TestCurrentInputFileWorksAcrossAutoDeleteModes(t *testing.T) {
 				t.Fatalf("expected completion payload for mode=%s", mode)
 			}
 			promptText, _ := ds.completionReq["prompt"].(string)
-			if !strings.Contains(promptText, promptcompat.BuildOpenAICurrentInputContextPrompt()) || strings.Contains(promptText, "first user turn") || strings.Contains(promptText, "latest user turn") {
+			if !strings.Contains(promptText, "Answer the latest user request directly.") || strings.Contains(promptText, "first user turn") || strings.Contains(promptText, "latest user turn") {
 				t.Fatalf("unexpected prompt for mode=%s: %s", mode, promptText)
 			}
 		})
--- a/internal/httpapi/openai/responses/empty_retry_runtime.go
+++ b/internal/httpapi/openai/responses/empty_retry_runtime.go
@@ -27,14 +27,14 @@ type responsesNonStreamResult struct {
 	responseMessageID     int
 }

-func (h *Handler) handleResponsesNonStreamWithRetry(w http.ResponseWriter, ctx context.Context, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
+func (h *Handler) handleResponsesNonStreamWithRetry(w http.ResponseWriter, ctx context.Context, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
 	attempts := 0
 	currentResp := resp
 	usagePrompt := finalPrompt
 	accumulatedThinking := ""
 	accumulatedToolDetectionThinking := ""
 	for {
-		result, ok := h.collectResponsesNonStreamAttempt(w, currentResp, responseID, model, usagePrompt, thinkingEnabled, searchEnabled, toolNames)
+		result, ok := h.collectResponsesNonStreamAttempt(w, currentResp, responseID, model, usagePrompt, thinkingEnabled, searchEnabled, toolNames, toolsRaw)
 		if !ok {
 			return
 		}
@@ -43,7 +43,7 @@ func (h *Handler) handleResponsesNonStreamWithRetry(w http.ResponseWriter, ctx c
 		result.thinking = accumulatedThinking
 		result.toolDetectionThinking = accumulatedToolDetectionThinking
 		result.parsed = detectAssistantToolCalls(result.text, result.thinking, result.toolDetectionThinking, toolNames)
-		result.body = openaifmt.BuildResponseObjectWithToolCalls(responseID, model, usagePrompt, result.thinking, result.text, result.parsed.Calls)
+		result.body = openaifmt.BuildResponseObjectWithToolCalls(responseID, model, usagePrompt, result.thinking, result.text, result.parsed.Calls, toolsRaw)

 		if !shouldRetryResponsesNonStream(result, attempts) {
 			h.finishResponsesNonStreamResult(w, result, attempts, owner, responseID, toolChoice, traceID)
@@ -68,7 +68,7 @@ func (h *Handler) handleResponsesNonStreamWithRetry(w http.ResponseWriter, ctx c
 	}
 }

-func (h *Handler) collectResponsesNonStreamAttempt(w http.ResponseWriter, resp *http.Response, responseID, model, usagePrompt string, thinkingEnabled, searchEnabled bool, toolNames []string) (responsesNonStreamResult, bool) {
+func (h *Handler) collectResponsesNonStreamAttempt(w http.ResponseWriter, resp *http.Response, responseID, model, usagePrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any) (responsesNonStreamResult, bool) {
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
@@ -84,7 +84,7 @@ func (h *Handler) collectResponsesNonStreamAttempt(w http.ResponseWriter, resp *
 		sanitizedText = replaceCitationMarkersWithLinks(sanitizedText, result.CitationLinks)
 	}
 	textParsed := detectAssistantToolCalls(sanitizedText, sanitizedThinking, toolDetectionThinking, toolNames)
-	responseObj := openaifmt.BuildResponseObjectWithToolCalls(responseID, model, usagePrompt, sanitizedThinking, sanitizedText, textParsed.Calls)
+	responseObj := openaifmt.BuildResponseObjectWithToolCalls(responseID, model, usagePrompt, sanitizedThinking, sanitizedText, textParsed.Calls, toolsRaw)
 	return responsesNonStreamResult{
 		thinking:              sanitizedThinking,
 		toolDetectionThinking: toolDetectionThinking,
@@ -123,8 +123,8 @@ func shouldRetryResponsesNonStream(result responsesNonStreamResult, attempts int
 		strings.TrimSpace(result.text) == ""
 }

-func (h *Handler) handleResponsesStreamWithRetry(w http.ResponseWriter, r *http.Request, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
-	streamRuntime, initialType, ok := h.prepareResponsesStreamRuntime(w, resp, owner, responseID, model, finalPrompt, thinkingEnabled, searchEnabled, toolNames, toolChoice, traceID)
+func (h *Handler) handleResponsesStreamWithRetry(w http.ResponseWriter, r *http.Request, a *auth.RequestAuth, resp *http.Response, payload map[string]any, pow, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
+	streamRuntime, initialType, ok := h.prepareResponsesStreamRuntime(w, resp, owner, responseID, model, finalPrompt, thinkingEnabled, searchEnabled, toolNames, toolsRaw, toolChoice, traceID)
 	if !ok {
 		return
 	}
@@ -165,7 +165,7 @@ func (h *Handler) handleResponsesStreamWithRetry(w http.ResponseWriter, r *http.
 	}
 }

-func (h *Handler) prepareResponsesStreamRuntime(w http.ResponseWriter, resp *http.Response, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolChoice promptcompat.ToolChoicePolicy, traceID string) (*responsesStreamRuntime, string, bool) {
+func (h *Handler) prepareResponsesStreamRuntime(w http.ResponseWriter, resp *http.Response, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, toolChoice promptcompat.ToolChoicePolicy, traceID string) (*responsesStreamRuntime, string, bool) {
 	if resp.StatusCode != http.StatusOK {
 		defer func() { _ = resp.Body.Close() }()
 		body, _ := io.ReadAll(resp.Body)
@@ -184,7 +184,7 @@ func (h *Handler) prepareResponsesStreamRuntime(w http.ResponseWriter, resp *htt
 	}
 	streamRuntime := newResponsesStreamRuntime(
 		w, rc, canFlush, responseID, model, finalPrompt, thinkingEnabled, searchEnabled,
-		h.compatStripReferenceMarkers(), toolNames, len(toolNames) > 0,
+		h.compatStripReferenceMarkers(), toolNames, toolsRaw, len(toolNames) > 0,
 		h.toolcallFeatureMatchEnabled() && h.toolcallEarlyEmitHighConfidence(),
 		toolChoice, traceID, func(obj map[string]any) {
 			h.getResponseStore().put(owner, responseID, obj)
--- a/internal/httpapi/openai/responses/responses_handler.go
+++ b/internal/httpapi/openai/responses/responses_handler.go
@@ -115,13 +115,13 @@ func (h *Handler) Responses(w http.ResponseWriter, r *http.Request) {

 	responseID := "resp_" + strings.ReplaceAll(uuid.NewString(), "-", "")
 	if stdReq.Stream {
-		h.handleResponsesStreamWithRetry(w, r, a, resp, payload, pow, owner, responseID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, stdReq.ToolChoice, traceID)
+		h.handleResponsesStreamWithRetry(w, r, a, resp, payload, pow, owner, responseID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, stdReq.ToolsRaw, stdReq.ToolChoice, traceID)
 		return
 	}
-	h.handleResponsesNonStreamWithRetry(w, r.Context(), a, resp, payload, pow, owner, responseID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, stdReq.ToolChoice, traceID)
+	h.handleResponsesNonStreamWithRetry(w, r.Context(), a, resp, payload, pow, owner, responseID, stdReq.ResponseModel, stdReq.FinalPrompt, stdReq.Thinking, stdReq.Search, stdReq.ToolNames, stdReq.ToolsRaw, stdReq.ToolChoice, traceID)
 }

-func (h *Handler) handleResponsesNonStream(w http.ResponseWriter, resp *http.Response, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
+func (h *Handler) handleResponsesNonStream(w http.ResponseWriter, resp *http.Response, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
@@ -148,12 +148,12 @@ func (h *Handler) handleResponsesNonStream(w http.ResponseWriter, resp *http.Res
 		return
 	}

-	responseObj := openaifmt.BuildResponseObjectWithToolCalls(responseID, model, finalPrompt, sanitizedThinking, sanitizedText, textParsed.Calls)
+	responseObj := openaifmt.BuildResponseObjectWithToolCalls(responseID, model, finalPrompt, sanitizedThinking, sanitizedText, textParsed.Calls, toolsRaw)
 	h.getResponseStore().put(owner, responseID, responseObj)
 	writeJSON(w, http.StatusOK, responseObj)
 }

-func (h *Handler) handleResponsesStream(w http.ResponseWriter, r *http.Request, resp *http.Response, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
+func (h *Handler) handleResponsesStream(w http.ResponseWriter, r *http.Request, resp *http.Response, owner, responseID, model, finalPrompt string, thinkingEnabled, searchEnabled bool, toolNames []string, toolsRaw any, toolChoice promptcompat.ToolChoicePolicy, traceID string) {
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
@@ -186,6 +186,7 @@ func (h *Handler) handleResponsesStream(w http.ResponseWriter, r *http.Request,
 		searchEnabled,
 		stripReferenceMarkers,
 		toolNames,
+		toolsRaw,
 		bufferToolContent,
 		emitEarlyToolDeltas,
 		toolChoice,
--- a/internal/httpapi/openai/responses/responses_stream_runtime_core.go
+++ b/internal/httpapi/openai/responses/responses_stream_runtime_core.go
@@ -22,6 +22,7 @@ type responsesStreamRuntime struct {
 	model       string
 	finalPrompt string
 	toolNames   []string
+	toolsRaw    any
 	traceID     string
 	toolChoice  promptcompat.ToolChoicePolicy

@@ -72,6 +73,7 @@ func newResponsesStreamRuntime(
 	searchEnabled bool,
 	stripReferenceMarkers bool,
 	toolNames []string,
+	toolsRaw any,
 	bufferToolContent bool,
 	emitEarlyToolDeltas bool,
 	toolChoice promptcompat.ToolChoicePolicy,
@@ -89,6 +91,7 @@ func newResponsesStreamRuntime(
 		searchEnabled:         searchEnabled,
 		stripReferenceMarkers: stripReferenceMarkers,
 		toolNames:             toolNames,
+		toolsRaw:              toolsRaw,
 		bufferToolContent:     bufferToolContent,
 		emitEarlyToolDeltas:   emitEarlyToolDeltas,
 		streamToolCallIDs:     map[int]string{},
--- a/internal/httpapi/openai/responses/responses_stream_runtime_toolcalls.go
+++ b/internal/httpapi/openai/responses/responses_stream_runtime_toolcalls.go
@@ -220,7 +220,8 @@ func (s *responsesStreamRuntime) emitFunctionCallDeltaEvents(deltas []toolstream
 }

 func (s *responsesStreamRuntime) emitFunctionCallDoneEvents(calls []toolcall.ParsedToolCall) {
-	for idx, tc := range calls {
+	normalizedCalls := toolcall.NormalizeParsedToolCallsForSchemas(calls, s.toolsRaw)
+	for idx, tc := range normalizedCalls {
 		if strings.TrimSpace(tc.Name) == "" {
 			continue
 		}
--- a/internal/httpapi/openai/responses/responses_stream_runtime_toolcalls_finalize.go
+++ b/internal/httpapi/openai/responses/responses_stream_runtime_toolcalls_finalize.go
@@ -109,7 +109,8 @@ func (s *responsesStreamRuntime) buildCompletedResponseObject(finalThinking, fin
 		}
 	}

-	for idx, tc := range calls {
+	normalizedCalls := toolcall.NormalizeParsedToolCallsForSchemas(calls, s.toolsRaw)
+	for idx, tc := range normalizedCalls {
 		if strings.TrimSpace(tc.Name) == "" {
 			continue
 		}
--- a/internal/httpapi/openai/responses/responses_stream_test.go
+++ b/internal/httpapi/openai/responses/responses_stream_test.go
@@ -27,7 +27,7 @@ func TestHandleResponsesStreamDoesNotEmitReasoningTextCompatEvents(t *testing.T)
 		Body:       io.NopCloser(strings.NewReader(streamBody)),
 	}

-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, nil, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, nil, nil, promptcompat.DefaultToolChoicePolicy(), "")

 	body := rec.Body.String()
 	if !strings.Contains(body, "event: response.reasoning.delta") {
@@ -57,7 +57,7 @@ func TestHandleResponsesStreamEmitsOutputTextDoneBeforeContentPartDone(t *testin
 		Body:       io.NopCloser(strings.NewReader(streamBody)),
 	}

-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, nil, promptcompat.DefaultToolChoicePolicy(), "")
 	body := rec.Body.String()
 	if !strings.Contains(body, "event: response.output_text.done") {
 		t.Fatalf("expected response.output_text.done payload, body=%s", body)
@@ -91,7 +91,7 @@ func TestHandleResponsesStreamOutputTextDeltaCarriesItemIndexes(t *testing.T) {
 		Body:       io.NopCloser(strings.NewReader(streamBody)),
 	}

-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, nil, promptcompat.DefaultToolChoicePolicy(), "")
 	body := rec.Body.String()

 	deltaPayload, ok := extractSSEEventPayload(body, "response.output_text.delta")
@@ -130,7 +130,7 @@ func TestHandleResponsesStreamEmitsDistinctToolCallIDsAcrossSeparateToolBlocks(t
 		Body:       io.NopCloser(strings.NewReader(streamBody)),
 	}

-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, []string{"read_file", "search"}, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, []string{"read_file", "search"}, nil, promptcompat.DefaultToolChoicePolicy(), "")

 	body := rec.Body.String()
 	doneEvents := extractSSEEventPayloads(body, "response.function_call_arguments.done")
@@ -183,7 +183,7 @@ func TestHandleResponsesStreamRequiredToolChoiceFailure(t *testing.T) {
 		Mode:    promptcompat.ToolChoiceRequired,
 		Allowed: map[string]struct{}{"read_file": {}},
 	}
-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, []string{"read_file"}, policy, "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, []string{"read_file"}, nil, policy, "")

 	body := rec.Body.String()
 	if !strings.Contains(body, "event: response.failed") {
@@ -213,7 +213,7 @@ func TestHandleResponsesStreamFailsWhenUpstreamHasOnlyThinking(t *testing.T) {
 		Body:       io.NopCloser(strings.NewReader(streamBody)),
 	}

-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, nil, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, nil, nil, promptcompat.DefaultToolChoicePolicy(), "")

 	body := rec.Body.String()
 	if !strings.Contains(body, "event: response.failed") {
@@ -251,7 +251,7 @@ func TestHandleResponsesStreamPromotesThinkingToolCallsOnFinalizeWithoutMidstrea
 		Body:       io.NopCloser(strings.NewReader(streamBody)),
 	}

-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, []string{"read_file"}, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, []string{"read_file"}, nil, promptcompat.DefaultToolChoicePolicy(), "")

 	body := rec.Body.String()
 	if !strings.Contains(body, "event: response.reasoning.delta") {
@@ -288,7 +288,7 @@ func TestHandleResponsesStreamPromotesHiddenThinkingDSMLToolCallsOnFinalize(t *t
 		Mode:    promptcompat.ToolChoiceRequired,
 		Allowed: map[string]struct{}{"read_file": {}},
 	}
-	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_hidden", "deepseek-v4-pro", "prompt", false, false, []string{"read_file"}, policy, "")
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_hidden", "deepseek-v4-pro", "prompt", false, false, []string{"read_file"}, nil, policy, "")

 	body := rec.Body.String()
 	if strings.Contains(body, "event: response.reasoning.delta") {
@@ -317,7 +317,7 @@ func TestHandleResponsesNonStreamRequiredToolChoiceViolation(t *testing.T) {
 		Allowed: map[string]struct{}{"read_file": {}},
 	}

-	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, []string{"read_file"}, policy, "")
+	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, []string{"read_file"}, nil, policy, "")
 	if rec.Code != http.StatusUnprocessableEntity {
 		t.Fatalf("expected 422 for required tool_choice violation, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -344,7 +344,7 @@ func TestHandleResponsesNonStreamRequiredToolChoiceIgnoresThinkingToolPayloadWhe
 		Allowed: map[string]struct{}{"read_file": {}},
 	}

-	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", true, false, []string{"read_file"}, policy, "")
+	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", true, false, []string{"read_file"}, nil, policy, "")
 	if rec.Code != http.StatusUnprocessableEntity {
 		t.Fatalf("expected 422 for required tool_choice violation, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -366,7 +366,7 @@ func TestHandleResponsesNonStreamReturns429WhenUpstreamOutputEmpty(t *testing.T)
 		)),
 	}

-	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, nil, promptcompat.DefaultToolChoicePolicy(), "")
 	if rec.Code != http.StatusTooManyRequests {
 		t.Fatalf("expected 429 for empty upstream output, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -388,7 +388,7 @@ func TestHandleResponsesNonStreamReturnsContentFilterErrorWhenUpstreamFilteredWi
 		)),
 	}

-	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-flash", "prompt", false, false, nil, nil, promptcompat.DefaultToolChoicePolicy(), "")
 	if rec.Code != http.StatusBadRequest {
 		t.Fatalf("expected 400 for filtered empty upstream output, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -410,7 +410,7 @@ func TestHandleResponsesNonStreamReturns429WhenUpstreamHasOnlyThinking(t *testin
 		)),
 	}

-	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, nil, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, nil, nil, promptcompat.DefaultToolChoicePolicy(), "")
 	if rec.Code != http.StatusTooManyRequests {
 		t.Fatalf("expected 429 for thinking-only upstream output, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -432,7 +432,7 @@ func TestHandleResponsesNonStreamPromotesThinkingToolCallsWhenTextEmpty(t *testi
 		)),
 	}

-	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, []string{"read_file"}, promptcompat.DefaultToolChoicePolicy(), "")
+	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_test", "deepseek-v4-pro", "prompt", true, false, []string{"read_file"}, nil, promptcompat.DefaultToolChoicePolicy(), "")
 	if rec.Code != http.StatusOK {
 		t.Fatalf("expected 200 for thinking tool calls, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -462,7 +462,7 @@ func TestHandleResponsesNonStreamPromotesHiddenThinkingDSMLToolCallsWhenTextEmpt
 		Mode:    promptcompat.ToolChoiceRequired,
 		Allowed: map[string]struct{}{"read_file": {}},
 	}
-	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_hidden", "deepseek-v4-pro", "prompt", false, false, []string{"read_file"}, policy, "")
+	h.handleResponsesNonStream(rec, resp, "owner-a", "resp_hidden", "deepseek-v4-pro", "prompt", false, false, []string{"read_file"}, nil, policy, "")
 	if rec.Code != http.StatusOK {
 		t.Fatalf("expected 200 for hidden thinking tool calls, got %d body=%s", rec.Code, rec.Body.String())
 	}
@@ -480,6 +480,53 @@ func TestHandleResponsesNonStreamPromotesHiddenThinkingDSMLToolCallsWhenTextEmpt
 	}
 }

+func TestHandleResponsesStreamCoercesSchemaDeclaredStringArguments(t *testing.T) {
+	h := &Handler{}
+	req := httptest.NewRequest(http.MethodPost, "/v1/responses", nil)
+	rec := httptest.NewRecorder()
+	toolsRaw := []any{
+		map[string]any{
+			"type": "function",
+			"function": map[string]any{
+				"name": "Write",
+				"parameters": map[string]any{
+					"type": "object",
+					"properties": map[string]any{
+						"content": map[string]any{"type": "string"},
+						"taskId":  map[string]any{"type": "string"},
+					},
+				},
+			},
+		},
+	}
+	sseLine := func(v string) string {
+		b, _ := json.Marshal(map[string]any{"p": "response/content", "v": v})
+		return "data: " + string(b) + "\n"
+	}
+	streamBody := sseLine(`<tool_calls><invoke name="Write">{"input":{"content":{"message":"hi"},"taskId":1}}</invoke></tool_calls>`) + "data: [DONE]\n"
+	resp := &http.Response{
+		StatusCode: http.StatusOK,
+		Body:       io.NopCloser(strings.NewReader(streamBody)),
+	}
+
+	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_string_protect", "deepseek-v4-flash", "prompt", false, false, []string{"Write"}, toolsRaw, promptcompat.DefaultToolChoicePolicy(), "")
+
+	payload, ok := extractSSEEventPayload(rec.Body.String(), "response.function_call_arguments.done")
+	if !ok {
+		t.Fatalf("expected response.function_call_arguments.done payload, body=%s", rec.Body.String())
+	}
+	args := map[string]any{}
+	if err := json.Unmarshal([]byte(asString(payload["arguments"])), &args); err != nil {
+		t.Fatalf("decode streamed response arguments failed: %v", err)
+	}
+	if args["content"] != `{"message":"hi"}` {
+		t.Fatalf("expected response content stringified by schema, got %#v", args["content"])
+	}
+	if args["taskId"] != "1" {
+		t.Fatalf("expected response taskId stringified by schema, got %#v", args["taskId"])
+	}
+}
+
 func extractSSEEventPayload(body, targetEvent string) (map[string]any, bool) {
 	scanner := bufio.NewScanner(strings.NewReader(body))
 	matched := false
--- a/internal/httpapi/openai/shared/citation_links.go
+++ b/internal/httpapi/openai/shared/citation_links.go
@@ -7,22 +7,27 @@ import (
 	"strings"
 )

-var citationMarkerPattern = regexp.MustCompile(`(?i)\[citation:\s*(\d+)\]`)
+var citationMarkerPattern = regexp.MustCompile(`(?i)\[(citation|reference):\s*(\d+)\]`)

 func ReplaceCitationMarkersWithLinks(text string, links map[int]string) string {
 	if strings.TrimSpace(text) == "" || len(links) == 0 {
 		return text
 	}
+	zeroBased := strings.Contains(strings.ToLower(text), "[reference:0]")
 	return citationMarkerPattern.ReplaceAllStringFunc(text, func(match string) string {
 		sub := citationMarkerPattern.FindStringSubmatch(match)
-		if len(sub) < 2 {
+		if len(sub) < 3 {
 			return match
 		}
-		idx, err := strconv.Atoi(strings.TrimSpace(sub[1]))
-		if err != nil || idx <= 0 {
+		idx, err := strconv.Atoi(strings.TrimSpace(sub[2]))
+		if err != nil || idx < 0 {
 			return match
 		}
-		url := strings.TrimSpace(links[idx])
+		lookupIdx := idx
+		if zeroBased {
+			lookupIdx = idx + 1
+		}
+		url := strings.TrimSpace(links[lookupIdx])
 		if url == "" {
 			return match
 		}
--- a/internal/httpapi/openai/shared/handler_toolcall_format.go
+++ b/internal/httpapi/openai/shared/handler_toolcall_format.go
@@ -70,12 +70,13 @@ func FilterIncrementalToolCallDeltasByAllowed(deltas []toolstream.ToolCallDelta,
 	return out
 }

-func FormatFinalStreamToolCallsWithStableIDs(calls []toolcall.ParsedToolCall, ids map[int]string) []map[string]any {
+func FormatFinalStreamToolCallsWithStableIDs(calls []toolcall.ParsedToolCall, ids map[int]string, toolsRaw any) []map[string]any {
 	if len(calls) == 0 {
 		return nil
 	}
+	normalizedCalls := toolcall.NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
 	out := make([]map[string]any, 0, len(calls))
-	for i, c := range calls {
+	for i, c := range normalizedCalls {
 		callID := ""
 		if ids != nil {
 			callID = strings.TrimSpace(ids[i])
--- a/internal/js/chat-stream/vercel_stream_impl.js
+++ b/internal/js/chat-stream/vercel_stream_impl.js
@@ -205,14 +205,14 @@ async function handleVercelStream(req, res, rawBody, payload) {
      if (detected.length > 0 && !toolCallsDoneEmitted) {
        toolCallsEmitted = true;
        toolCallsDoneEmitted = true;
-        sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(detected, streamToolCallIDs) });
+        sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(detected, streamToolCallIDs, payload.tools) });
      } else if (toolSieveEnabled) {
        const tailEvents = flushToolSieve(toolSieveState, toolNames);
        for (const evt of tailEvents) {
          if (evt.type === 'tool_calls' && Array.isArray(evt.calls) && evt.calls.length > 0) {
            toolCallsEmitted = true;
            toolCallsDoneEmitted = true;
-            sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(evt.calls, streamToolCallIDs) });
+            sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(evt.calls, streamToolCallIDs, payload.tools) });
            resetStreamToolCallState(streamToolCallIDs, streamToolNames);
            continue;
          }
@@ -352,14 +352,14 @@ async function handleVercelStream(req, res, rawBody, payload) {
                      const formatted = formatIncrementalToolCallDeltas(filtered, streamToolCallIDs);
                      if (formatted.length > 0) {
                        toolCallsEmitted = true;
-                        sendDeltaFrame({ tool_calls: formatted });
+                      sendDeltaFrame({ tool_calls: formatted });
                      }
                      continue;
                    }
                    if (evt.type === 'tool_calls') {
                      toolCallsEmitted = true;
                      toolCallsDoneEmitted = true;
-                      sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(evt.calls, streamToolCallIDs) });
+                      sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(evt.calls, streamToolCallIDs, payload.tools) });
                      resetStreamToolCallState(streamToolCallIDs, streamToolNames);
                      continue;
                    }
--- a/internal/js/helpers/stream-tool-sieve/format.js
+++ b/internal/js/helpers/stream-tool-sieve/format.js
@@ -2,11 +2,12 @@

 const crypto = require('crypto');

-function formatOpenAIStreamToolCalls(calls, idStore) {
+function formatOpenAIStreamToolCalls(calls, idStore, toolsRaw) {
  if (!Array.isArray(calls) || calls.length === 0) {
    return [];
  }
-  return calls.map((c, idx) => ({
+  const normalized = normalizeParsedToolCallsForSchemas(calls, toolsRaw);
+  return normalized.map((c, idx) => ({
    index: idx,
    id: ensureStreamToolCallID(idStore, idx),
    type: 'function',
@@ -17,6 +18,194 @@ function formatOpenAIStreamToolCalls(calls, idStore) {
  }));
 }

+function normalizeParsedToolCallsForSchemas(calls, toolsRaw) {
+  if (!Array.isArray(calls) || calls.length === 0) {
+    return calls;
+  }
+  const schemas = buildToolSchemaIndex(toolsRaw);
+  if (!schemas) {
+    return calls;
+  }
+  let changedAny = false;
+  const out = calls.map((call) => {
+    const name = String(call && call.name || '').trim().toLowerCase();
+    const schema = schemas[name];
+    if (!schema || !call || !call.input || typeof call.input !== 'object' || Array.isArray(call.input)) {
+      return call;
+    }
+    const [normalized, changed] = normalizeToolValueWithSchema(call.input, schema);
+    if (!changed || !normalized || typeof normalized !== 'object' || Array.isArray(normalized)) {
+      return call;
+    }
+    changedAny = true;
+    return { ...call, input: normalized };
+  });
+  return changedAny ? out : calls;
+}
+
+function buildToolSchemaIndex(toolsRaw) {
+  if (!Array.isArray(toolsRaw) || toolsRaw.length === 0) {
+    return null;
+  }
+  const out = {};
+  for (const item of toolsRaw) {
+    if (!item || typeof item !== 'object' || Array.isArray(item)) {
+      continue;
+    }
+    const [name, schema] = extractToolNameAndSchema(item);
+    if (!name || !schema || typeof schema !== 'object' || Array.isArray(schema)) {
+      continue;
+    }
+    out[name.toLowerCase()] = schema;
+  }
+  return Object.keys(out).length > 0 ? out : null;
+}
+
+function extractToolNameAndSchema(tool) {
+  const fn = tool && typeof tool.function === 'object' && !Array.isArray(tool.function) ? tool.function : null;
+  const name = firstNonEmptyString(tool.name, fn && fn.name);
+  const schema = firstNonNil(
+    tool.parameters,
+    tool.input_schema,
+    tool.inputSchema,
+    tool.schema,
+    fn && fn.parameters,
+    fn && fn.input_schema,
+    fn && fn.inputSchema,
+    fn && fn.schema,
+  );
+  return [name, schema];
+}
+
+function normalizeToolValueWithSchema(value, schema) {
+  if (value == null || !schema || typeof schema !== 'object' || Array.isArray(schema)) {
+    return [value, false];
+  }
+  if (shouldCoerceSchemaToString(schema)) {
+    return stringifySchemaValue(value);
+  }
+  if (looksLikeObjectSchema(schema)) {
+    if (!value || typeof value !== 'object' || Array.isArray(value)) {
+      return [value, false];
+    }
+    const properties = schema.properties && typeof schema.properties === 'object' && !Array.isArray(schema.properties) ? schema.properties : null;
+    const additional = schema.additionalProperties;
+    let changed = false;
+    const out = {};
+    for (const [key, current] of Object.entries(value)) {
+      let next = current;
+      let fieldChanged = false;
+      if (properties && Object.prototype.hasOwnProperty.call(properties, key)) {
+        [next, fieldChanged] = normalizeToolValueWithSchema(current, properties[key]);
+      } else if (additional != null) {
+        [next, fieldChanged] = normalizeToolValueWithSchema(current, additional);
+      }
+      out[key] = next;
+      changed = changed || fieldChanged;
+    }
+    return changed ? [out, true] : [value, false];
+  }
+  if (looksLikeArraySchema(schema)) {
+    if (!Array.isArray(value) || value.length === 0 || schema.items == null) {
+      return [value, false];
+    }
+    let changed = false;
+    const out = value.map((item, idx) => {
+      const itemSchema = Array.isArray(schema.items) ? schema.items[idx] : schema.items;
+      if (itemSchema == null) {
+        return item;
+      }
+      const [next, itemChanged] = normalizeToolValueWithSchema(item, itemSchema);
+      changed = changed || itemChanged;
+      return next;
+    });
+    return changed ? [out, true] : [value, false];
+  }
+  return [value, false];
+}
+
+function shouldCoerceSchemaToString(schema) {
+  if (!schema || typeof schema !== 'object' || Array.isArray(schema)) {
+    return false;
+  }
+  if (typeof schema.const === 'string') {
+    return true;
+  }
+  if (Array.isArray(schema.enum) && schema.enum.length > 0 && schema.enum.every((item) => typeof item === 'string')) {
+    return true;
+  }
+  if (typeof schema.type === 'string') {
+    return schema.type.trim().toLowerCase() === 'string';
+  }
+  if (Array.isArray(schema.type) && schema.type.length > 0) {
+    let hasString = false;
+    for (const item of schema.type) {
+      if (typeof item !== 'string') {
+        return false;
+      }
+      const typ = item.trim().toLowerCase();
+      if (typ === 'string') {
+        hasString = true;
+      } else if (typ !== 'null') {
+        return false;
+      }
+    }
+    return hasString;
+  }
+  return false;
+}
+
+function looksLikeObjectSchema(schema) {
+  return !!schema && typeof schema === 'object' && !Array.isArray(schema) && (
+    (typeof schema.type === 'string' && schema.type.trim().toLowerCase() === 'object') ||
+    (schema.properties && typeof schema.properties === 'object' && !Array.isArray(schema.properties)) ||
+    schema.additionalProperties != null
+  );
+}
+
+function looksLikeArraySchema(schema) {
+  return !!schema && typeof schema === 'object' && !Array.isArray(schema) && (
+    (typeof schema.type === 'string' && schema.type.trim().toLowerCase() === 'array') ||
+    schema.items != null
+  );
+}
+
+function stringifySchemaValue(value) {
+  if (value == null) {
+    return [value, false];
+  }
+  if (typeof value === 'string') {
+    return [value, false];
+  }
+  try {
+    return [JSON.stringify(value), true];
+  } catch {
+    return [value, false];
+  }
+}
+
+function firstNonNil(...values) {
+  for (const value of values) {
+    if (value != null) {
+      return value;
+    }
+  }
+  return null;
+}
+
+function firstNonEmptyString(...values) {
+  for (const value of values) {
+    if (typeof value !== 'string') {
+      continue;
+    }
+    const trimmed = value.trim();
+    if (trimmed) {
+      return trimmed;
+    }
+  }
+  return '';
+}
+
 function ensureStreamToolCallID(idStore, index) {
  if (!(idStore instanceof Map)) {
    return `call_${newCallID()}`;
--- a/internal/js/helpers/stream-tool-sieve/parse_payload.js
+++ b/internal/js/helpers/stream-tool-sieve/parse_payload.js
@@ -530,6 +530,7 @@ function findPartialToolMarkupStart(text) {
    '<|tool_calls', '<|invoke', '<|parameter',
    '<｜tool_calls', '<｜invoke', '<｜parameter',
    '<|dsml|tool_calls', '<|dsml|invoke', '<|dsml|parameter',
+    '<｜dsml|tool_calls', '<｜dsml|invoke', '<｜dsml|parameter',
    '<dsmltool_calls', '<dsmlinvoke', '<dsmlparameter',
    '<dsml tool_calls', '<dsml invoke', '<dsml parameter',
    '<dsml|tool_calls', '<dsml|invoke', '<dsml|parameter',
@@ -812,6 +813,9 @@ function parseStructuredCDATAParameterValue(paramName, raw) {
  if (!normalized.includes('<') || !normalized.includes('>')) {
    return { ok: false, value: null };
  }
+  if (!cdataFragmentLooksExplicitlyStructured(normalized)) {
+    return { ok: false, value: null };
+  }
  const parsed = parseMarkupInput(normalized);
  if (Array.isArray(parsed)) {
    return { ok: true, value: parsed };
@@ -826,6 +830,21 @@ function normalizeCDATAForStructuredParse(raw) {
  return unescapeHtml(toStringSafe(raw).replace(/<br\s*\/?>/gi, '\n').trim());
 }

+function cdataFragmentLooksExplicitlyStructured(raw) {
+  const blocks = findGenericXmlElementBlocks(raw);
+  if (blocks.length === 0) {
+    return false;
+  }
+  if (blocks.length > 1) {
+    return true;
+  }
+  const block = blocks[0];
+  if (toStringSafe(block.localName).trim().toLowerCase() === 'item') {
+    return true;
+  }
+  return findGenericXmlElementBlocks(block.body).length > 0;
+}
+
 function preservesCDATAStringParameter(name) {
  return new Set([
    'content',
--- a/internal/js/helpers/stream-tool-sieve/tool-keywords.js
+++ b/internal/js/helpers/stream-tool-sieve/tool-keywords.js
@@ -2,6 +2,7 @@

 const XML_TOOL_SEGMENT_TAGS = [
  '<|dsml|tool_calls>', '<|dsml|tool_calls\n', '<|dsml|tool_calls ',
+  '<｜dsml|tool_calls>', '<｜dsml|tool_calls\n', '<｜dsml|tool_calls ',
  '<|dsml|invoke ', '<|dsml|invoke\n', '<|dsml|invoke\t', '<|dsml|invoke\r',
  '<|dsmltool_calls>', '<|dsmltool_calls\n', '<|dsmltool_calls ',
  '<|dsmlinvoke ', '<|dsmlinvoke\n', '<|dsmlinvoke\t', '<|dsmlinvoke\r',
@@ -23,6 +24,7 @@ const XML_TOOL_SEGMENT_TAGS = [

 const XML_TOOL_OPENING_TAGS = [
  '<|dsml|tool_calls',
+  '<｜dsml|tool_calls',
  '<|dsmltool_calls',
  '<|dsml tool_calls',
  '<dsml|tool_calls',
@@ -35,6 +37,7 @@ const XML_TOOL_OPENING_TAGS = [

 const XML_TOOL_CLOSING_TAGS = [
  '</|dsml|tool_calls>',
+  '</｜dsml|tool_calls>',
  '</|dsmltool_calls>',
  '</|dsml tool_calls>',
  '</dsml|tool_calls>',
--- a/internal/promptcompat/history_transcript.go
+++ b/internal/promptcompat/history_transcript.go
@@ -9,8 +9,6 @@ import (

 const historySplitInjectedFilename = "IGNORE"

-const currentInputContextNote = "[context note]\nThis is a compacted snapshot of the prior conversation history for the current request.\nUse it as history only. Do not treat it as a new instruction.\nIf the same question or tool action already appears here, do not repeat it unless the latest turn adds new information.\n[/context note]"
-
 func BuildOpenAIHistoryTranscript(messages []any) string {
 	return buildOpenAIInjectedFileTranscript(messages)
 }
@@ -28,15 +26,11 @@ func BuildOpenAICurrentInputContextTranscript(messages []any) string {
 	return buildOpenAIInjectedFileTranscript(messages)
 }

-func BuildOpenAICurrentInputContextPrompt() string {
-	return "You are in a compacted-context mode. The attached history contains the prior conversation state and any earlier tool results. Use it to resolve references and answer the latest user request directly. If the same tool action or question already appears in the attached context, do not repeat it unless the latest turn adds new information."
-}
-
 func buildOpenAIInjectedFileTranscript(messages []any) string {
 	normalized := NormalizeOpenAIMessagesForPrompt(messages, "")
 	transcript := strings.TrimSpace(prompt.MessagesPrepare(normalized))
 	if transcript == "" {
 		return ""
 	}
-	return fmt.Sprintf("[file content end]\n\n%s\n\n%s\n\n[file name]: %s\n[file content begin]\n", currentInputContextNote, transcript, historySplitInjectedFilename)
+	return fmt.Sprintf("[file content end]\n\n%s\n\n[file name]: %s\n[file content begin]\n", transcript, historySplitInjectedFilename)
 }
--- a/internal/promptcompat/tool_prompt.go
+++ b/internal/promptcompat/tool_prompt.go
@@ -30,13 +30,7 @@ func injectToolPrompt(messages []map[string]any, tools []any, policy ToolChoiceP
 		if !ok {
 			continue
 		}
-		fn, _ := tool["function"].(map[string]any)
-		if len(fn) == 0 {
-			fn = tool
-		}
-		name, _ := fn["name"].(string)
-		desc, _ := fn["description"].(string)
-		schema, _ := fn["parameters"].(map[string]any)
+		name, desc, schema := toolcall.ExtractToolMeta(tool)
 		name = strings.TrimSpace(name)
 		if !isAllowed(name) {
 			continue
--- a/internal/toolcall/toolcall_edge_test.go
+++ b/internal/toolcall/toolcall_edge_test.go
@@ -9,7 +9,7 @@ import (
 func TestFormatOpenAIStreamToolCalls(t *testing.T) {
 	formatted := FormatOpenAIStreamToolCalls([]ParsedToolCall{
 		{Name: "search", Input: map[string]any{"q": "test"}},
-	})
+	}, nil)
 	if len(formatted) != 1 {
 		t.Fatalf("expected 1, got %d", len(formatted))
 	}
--- a/internal/toolcall/toolcalls_format.go
+++ b/internal/toolcall/toolcalls_format.go
@@ -7,9 +7,10 @@ import (
 	"github.com/google/uuid"
 )

-func FormatOpenAIToolCalls(calls []ParsedToolCall) []map[string]any {
+func FormatOpenAIToolCalls(calls []ParsedToolCall, toolsRaw any) []map[string]any {
+	normalized := NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
 	out := make([]map[string]any, 0, len(calls))
-	for _, c := range calls {
+	for _, c := range normalized {
 		args, _ := json.Marshal(c.Input)
 		out = append(out, map[string]any{
 			"id":   "call_" + strings.ReplaceAll(uuid.NewString(), "-", ""),
@@ -23,9 +24,10 @@ func FormatOpenAIToolCalls(calls []ParsedToolCall) []map[string]any {
 	return out
 }

-func FormatOpenAIStreamToolCalls(calls []ParsedToolCall) []map[string]any {
+func FormatOpenAIStreamToolCalls(calls []ParsedToolCall, toolsRaw any) []map[string]any {
+	normalized := NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
 	out := make([]map[string]any, 0, len(calls))
-	for i, c := range calls {
+	for i, c := range normalized {
 		args, _ := json.Marshal(c.Input)
 		out = append(out, map[string]any{
 			"index": i,
--- a/internal/toolcall/toolcalls_parse_markup.go
+++ b/internal/toolcall/toolcalls_parse_markup.go
@@ -2,6 +2,7 @@ package toolcall

 import (
 	"encoding/json"
+	"encoding/xml"
 	"html"
 	"regexp"
 	"strings"
@@ -350,6 +351,9 @@ func parseStructuredCDATAParameterValue(paramName, raw string) (any, bool) {
 	if !strings.Contains(normalized, "<") || !strings.Contains(normalized, ">") {
 		return nil, false
 	}
+	if !cdataFragmentLooksExplicitlyStructured(normalized) {
+		return nil, false
+	}
 	parsed, ok := parseXMLFragmentValue(normalized)
 	if !ok {
 		return nil, false
@@ -375,6 +379,65 @@ func normalizeCDATAForStructuredParse(raw string) string {
 	return html.UnescapeString(strings.TrimSpace(normalized))
 }

+// Preserve flat CDATA fragments as strings. Only recover structure when the
+// fragment clearly encodes a data shape: multiple sibling elements, nested
+// child elements, or an explicit item list.
+func cdataFragmentLooksExplicitlyStructured(raw string) bool {
+	trimmed := strings.TrimSpace(raw)
+	if trimmed == "" {
+		return false
+	}
+
+	dec := xml.NewDecoder(strings.NewReader("<root>" + trimmed + "</root>"))
+	tok, err := dec.Token()
+	if err != nil {
+		return false
+	}
+	start, ok := tok.(xml.StartElement)
+	if !ok || !strings.EqualFold(start.Name.Local, "root") {
+		return false
+	}
+
+	depth := 0
+	directChildren := 0
+	firstChildName := ""
+	firstChildHasNested := false
+
+	for {
+		tok, err := dec.Token()
+		if err != nil {
+			return false
+		}
+		switch t := tok.(type) {
+		case xml.StartElement:
+			if depth == 0 {
+				directChildren++
+				if directChildren == 1 {
+					firstChildName = strings.ToLower(strings.TrimSpace(t.Name.Local))
+				} else {
+					return true
+				}
+			} else if directChildren == 1 && depth == 1 {
+				firstChildHasNested = true
+			}
+			depth++
+		case xml.EndElement:
+			if strings.EqualFold(t.Name.Local, "root") {
+				if directChildren != 1 {
+					return false
+				}
+				if firstChildName == "item" {
+					return true
+				}
+				return firstChildHasNested
+			}
+			if depth > 0 {
+				depth--
+			}
+		}
+	}
+}
+
 func preservesCDATAStringParameter(name string) bool {
 	switch strings.ToLower(strings.TrimSpace(name)) {
 	case "content", "file_content", "text", "prompt", "query", "command", "cmd", "script", "code", "old_string", "new_string", "pattern", "path", "file_path":
--- a/internal/toolcall/toolcalls_schema_normalize.go
+++ b/internal/toolcall/toolcalls_schema_normalize.go
@@ -0,0 +1,282 @@
+package toolcall
+
+import (
+	"encoding/json"
+	"strings"
+)
+
+func NormalizeParsedToolCallsForSchemas(calls []ParsedToolCall, toolsRaw any) []ParsedToolCall {
+	if len(calls) == 0 {
+		return calls
+	}
+	schemas := buildToolSchemaIndex(toolsRaw)
+	if len(schemas) == 0 {
+		return calls
+	}
+
+	var changedAny bool
+	out := make([]ParsedToolCall, len(calls))
+	for i, call := range calls {
+		out[i] = call
+		schema, ok := schemas[strings.ToLower(strings.TrimSpace(call.Name))]
+		if !ok || call.Input == nil {
+			continue
+		}
+		normalized, changed := normalizeToolValueWithSchema(call.Input, schema)
+		if !changed {
+			continue
+		}
+		changedAny = true
+		if input, ok := normalized.(map[string]any); ok {
+			out[i].Input = input
+		}
+	}
+	if !changedAny {
+		return calls
+	}
+	return out
+}
+
+func buildToolSchemaIndex(toolsRaw any) map[string]any {
+	tools, ok := toolsRaw.([]any)
+	if !ok || len(tools) == 0 {
+		return nil
+	}
+	out := make(map[string]any, len(tools))
+	for _, item := range tools {
+		tool, ok := item.(map[string]any)
+		if !ok {
+			continue
+		}
+		name, _, schema := ExtractToolMeta(tool)
+		if name == "" || schema == nil {
+			continue
+		}
+		out[strings.ToLower(name)] = schema
+	}
+	if len(out) == 0 {
+		return nil
+	}
+	return out
+}
+
+func ExtractToolMeta(tool map[string]any) (string, string, any) {
+	name := strings.TrimSpace(asStringValue(tool["name"]))
+	desc := strings.TrimSpace(asStringValue(tool["description"]))
+	schema := firstNonNil(
+		tool["parameters"],
+		tool["input_schema"],
+		tool["inputSchema"],
+		tool["schema"],
+	)
+	if fn, ok := tool["function"].(map[string]any); ok {
+		if name == "" {
+			name = strings.TrimSpace(asStringValue(fn["name"]))
+		}
+		if desc == "" {
+			desc = strings.TrimSpace(asStringValue(fn["description"]))
+		}
+		schema = firstNonNil(
+			schema,
+			fn["parameters"],
+			fn["input_schema"],
+			fn["inputSchema"],
+			fn["schema"],
+		)
+	}
+	return name, desc, schema
+}
+
+func normalizeToolValueWithSchema(value any, schema any) (any, bool) {
+	if value == nil || schema == nil {
+		return value, false
+	}
+	schemaMap, ok := schema.(map[string]any)
+	if !ok || len(schemaMap) == 0 {
+		return value, false
+	}
+	if shouldCoerceSchemaToString(schemaMap) {
+		return stringifySchemaValue(value)
+	}
+	if looksLikeObjectSchema(schemaMap) {
+		obj, ok := value.(map[string]any)
+		if !ok || len(obj) == 0 {
+			return value, false
+		}
+		properties, _ := schemaMap["properties"].(map[string]any)
+		additional := schemaMap["additionalProperties"]
+		changed := false
+		out := make(map[string]any, len(obj))
+		for key, current := range obj {
+			next := current
+			var fieldChanged bool
+			if propSchema, ok := properties[key]; ok {
+				next, fieldChanged = normalizeToolValueWithSchema(current, propSchema)
+			} else if additional != nil {
+				next, fieldChanged = normalizeToolValueWithSchema(current, additional)
+			}
+			out[key] = next
+			changed = changed || fieldChanged
+		}
+		if !changed {
+			return value, false
+		}
+		return out, true
+	}
+	if looksLikeArraySchema(schemaMap) {
+		arr, ok := value.([]any)
+		if !ok || len(arr) == 0 {
+			return value, false
+		}
+		itemsSchema := schemaMap["items"]
+		if itemsSchema == nil {
+			return value, false
+		}
+		changed := false
+		out := make([]any, len(arr))
+		switch itemSchemas := itemsSchema.(type) {
+		case []any:
+			for i, item := range arr {
+				if i >= len(itemSchemas) {
+					out[i] = item
+					continue
+				}
+				next, itemChanged := normalizeToolValueWithSchema(item, itemSchemas[i])
+				out[i] = next
+				changed = changed || itemChanged
+			}
+		default:
+			for i, item := range arr {
+				next, itemChanged := normalizeToolValueWithSchema(item, itemsSchema)
+				out[i] = next
+				changed = changed || itemChanged
+			}
+		}
+		if !changed {
+			return value, false
+		}
+		return out, true
+	}
+	return value, false
+}
+
+func shouldCoerceSchemaToString(schema map[string]any) bool {
+	if schema == nil {
+		return false
+	}
+	if isStringConst(schema["const"]) {
+		return true
+	}
+	if isStringEnum(schema["enum"]) {
+		return true
+	}
+	switch v := schema["type"].(type) {
+	case string:
+		return strings.EqualFold(strings.TrimSpace(v), "string")
+	case []any:
+		return isOnlyStringLikeTypes(v)
+	case []string:
+		items := make([]any, 0, len(v))
+		for _, item := range v {
+			items = append(items, item)
+		}
+		return isOnlyStringLikeTypes(items)
+	default:
+		return false
+	}
+}
+
+func looksLikeObjectSchema(schema map[string]any) bool {
+	if schema == nil {
+		return false
+	}
+	if typ, ok := schema["type"].(string); ok && strings.EqualFold(strings.TrimSpace(typ), "object") {
+		return true
+	}
+	if _, ok := schema["properties"].(map[string]any); ok {
+		return true
+	}
+	_, hasAdditional := schema["additionalProperties"]
+	return hasAdditional
+}
+
+func looksLikeArraySchema(schema map[string]any) bool {
+	if schema == nil {
+		return false
+	}
+	if typ, ok := schema["type"].(string); ok && strings.EqualFold(strings.TrimSpace(typ), "array") {
+		return true
+	}
+	_, hasItems := schema["items"]
+	return hasItems
+}
+
+func isOnlyStringLikeTypes(values []any) bool {
+	if len(values) == 0 {
+		return false
+	}
+	hasString := false
+	for _, item := range values {
+		typ, ok := item.(string)
+		if !ok {
+			return false
+		}
+		switch strings.ToLower(strings.TrimSpace(typ)) {
+		case "string":
+			hasString = true
+		case "null":
+			continue
+		default:
+			return false
+		}
+	}
+	return hasString
+}
+
+func isStringConst(v any) bool {
+	_, ok := v.(string)
+	return ok
+}
+
+func isStringEnum(v any) bool {
+	values, ok := v.([]any)
+	if !ok || len(values) == 0 {
+		return false
+	}
+	for _, item := range values {
+		if _, ok := item.(string); !ok {
+			return false
+		}
+	}
+	return true
+}
+
+func stringifySchemaValue(value any) (any, bool) {
+	if value == nil {
+		return value, false
+	}
+	if s, ok := value.(string); ok {
+		return s, false
+	}
+	b, err := json.Marshal(value)
+	if err != nil {
+		return value, false
+	}
+	return string(b), true
+}
+
+func asStringValue(v any) string {
+	if s, ok := v.(string); ok {
+		return s
+	}
+	return ""
+}
+
+func firstNonNil(values ...any) any {
+	for _, value := range values {
+		if value != nil {
+			return value
+		}
+	}
+	return nil
+}
--- a/internal/toolcall/toolcalls_schema_normalize_test.go
+++ b/internal/toolcall/toolcalls_schema_normalize_test.go
@@ -0,0 +1,161 @@
+package toolcall
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestNormalizeParsedToolCallsForSchemasCoercesDeclaredStringFieldsRecursively(t *testing.T) {
+	toolsRaw := []any{
+		map[string]any{
+			"type": "function",
+			"function": map[string]any{
+				"name": "TaskUpdate",
+				"parameters": map[string]any{
+					"type": "object",
+					"properties": map[string]any{
+						"taskId": map[string]any{"type": "string"},
+						"payload": map[string]any{
+							"type": "object",
+							"properties": map[string]any{
+								"content": map[string]any{"type": "string"},
+								"tags": map[string]any{
+									"type":  "array",
+									"items": map[string]any{"type": "string"},
+								},
+								"count": map[string]any{"type": "number"},
+							},
+						},
+					},
+				},
+			},
+		},
+	}
+	calls := []ParsedToolCall{{
+		Name: "TaskUpdate",
+		Input: map[string]any{
+			"taskId": 1,
+			"payload": map[string]any{
+				"content": map[string]any{"text": "hello"},
+				"tags":    []any{1, true, map[string]any{"k": "v"}},
+				"count":   2,
+			},
+		},
+	}}
+
+	got := NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
+	if len(got) != 1 {
+		t.Fatalf("expected one normalized call, got %#v", got)
+	}
+	if got[0].Input["taskId"] != "1" {
+		t.Fatalf("expected taskId coerced to string, got %#v", got[0].Input["taskId"])
+	}
+	payload, ok := got[0].Input["payload"].(map[string]any)
+	if !ok {
+		t.Fatalf("expected payload object, got %#v", got[0].Input["payload"])
+	}
+	if payload["content"] != `{"text":"hello"}` {
+		t.Fatalf("expected nested content coerced to json string, got %#v", payload["content"])
+	}
+	if payload["count"] != 2 {
+		t.Fatalf("expected non-string count unchanged, got %#v", payload["count"])
+	}
+	tags, ok := payload["tags"].([]any)
+	if !ok {
+		t.Fatalf("expected tags slice, got %#v", payload["tags"])
+	}
+	wantTags := []any{"1", "true", `{"k":"v"}`}
+	if !reflect.DeepEqual(tags, wantTags) {
+		t.Fatalf("unexpected normalized tags: got %#v want %#v", tags, wantTags)
+	}
+}
+
+func TestNormalizeParsedToolCallsForSchemasSupportsDirectToolSchemaShape(t *testing.T) {
+	toolsRaw := []any{
+		map[string]any{
+			"name": "Write",
+			"input_schema": map[string]any{
+				"type": "object",
+				"properties": map[string]any{
+					"content": map[string]any{"type": "string"},
+				},
+			},
+		},
+	}
+	calls := []ParsedToolCall{{Name: "Write", Input: map[string]any{"content": []any{"a", 1}}}}
+	got := NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
+	if got[0].Input["content"] != `["a",1]` {
+		t.Fatalf("expected direct-schema content coerced to string, got %#v", got[0].Input["content"])
+	}
+}
+
+func TestNormalizeParsedToolCallsForSchemasLeavesAmbiguousUnionUnchanged(t *testing.T) {
+	toolsRaw := []any{
+		map[string]any{
+			"type": "function",
+			"function": map[string]any{
+				"name": "TaskUpdate",
+				"parameters": map[string]any{
+					"type": "object",
+					"properties": map[string]any{
+						"taskId": map[string]any{"type": []any{"string", "integer"}},
+					},
+				},
+			},
+		},
+	}
+	calls := []ParsedToolCall{{Name: "TaskUpdate", Input: map[string]any{"taskId": 1}}}
+	got := NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
+	if got[0].Input["taskId"] != 1 {
+		t.Fatalf("expected ambiguous union to stay unchanged, got %#v", got[0].Input["taskId"])
+	}
+}
+
+func TestNormalizeParsedToolCallsForSchemasSupportsCamelCaseInputSchema(t *testing.T) {
+	toolsRaw := []any{
+		map[string]any{
+			"name": "Write",
+			"inputSchema": map[string]any{
+				"type": "object",
+				"properties": map[string]any{
+					"content": map[string]any{"type": "string"},
+				},
+			},
+		},
+	}
+	calls := []ParsedToolCall{{Name: "Write", Input: map[string]any{"content": map[string]any{"message": "hi"}}}}
+	got := NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
+	if got[0].Input["content"] != `{"message":"hi"}` {
+		t.Fatalf("expected camelCase inputSchema content coercion, got %#v", got[0].Input["content"])
+	}
+}
+
+func TestNormalizeParsedToolCallsForSchemasPreservesArrayWhenSchemaSaysArray(t *testing.T) {
+	toolsRaw := []any{
+		map[string]any{
+			"name": "todowrite",
+			"inputSchema": map[string]any{
+				"type": "object",
+				"properties": map[string]any{
+					"todos": map[string]any{
+						"type": "array",
+						"items": map[string]any{
+							"type": "object",
+							"properties": map[string]any{
+								"content":  map[string]any{"type": "string"},
+								"status":   map[string]any{"type": "string"},
+								"priority": map[string]any{"type": "string"},
+							},
+						},
+					},
+				},
+			},
+		},
+	}
+	todos := []any{map[string]any{"content": "x", "status": "pending", "priority": "high"}}
+	calls := []ParsedToolCall{{Name: "todowrite", Input: map[string]any{"todos": todos}}}
+	got := NormalizeParsedToolCallsForSchemas(calls, toolsRaw)
+	if !reflect.DeepEqual(got[0].Input["todos"], todos) {
+		t.Fatalf("expected todos array preserved, got %#v want %#v", got[0].Input["todos"], todos)
+	}
+}
--- a/internal/toolcall/toolcalls_test.go
+++ b/internal/toolcall/toolcalls_test.go
@@ -6,7 +6,7 @@ import (
 )

 func TestFormatOpenAIToolCalls(t *testing.T) {
-	formatted := FormatOpenAIToolCalls([]ParsedToolCall{{Name: "search", Input: map[string]any{"q": "x"}}})
+	formatted := FormatOpenAIToolCalls([]ParsedToolCall{{Name: "search", Input: map[string]any{"q": "x"}}}, nil)
 	if len(formatted) != 1 {
 		t.Fatalf("expected 1, got %d", len(formatted))
 	}
@@ -53,6 +53,21 @@ func TestParseToolCallsSupportsDSMLShellWithCanonicalExampleInCDATA(t *testing.T
 	}
 }

+func TestParseToolCallsPreservesSimpleCDATAInlineMarkupAsText(t *testing.T) {
+	text := `<tool_calls><invoke name="Write"><parameter name="description"><![CDATA[<b>urgent</b>]]></parameter></invoke></tool_calls>`
+	calls := ParseToolCalls(text, []string{"Write"})
+	if len(calls) != 1 {
+		t.Fatalf("expected 1 call, got %#v", calls)
+	}
+	got, ok := calls[0].Input["description"].(string)
+	if !ok {
+		t.Fatalf("expected description to remain a string, got %#v", calls[0].Input["description"])
+	}
+	if got != "<b>urgent</b>" {
+		t.Fatalf("expected inline markup CDATA to stay raw, got %q", got)
+	}
+}
+
 func TestParseToolCallsTreatsUnclosedCDATAAsText(t *testing.T) {
 	text := `<tool_calls><invoke name="Write"><parameter name="content"><![CDATA[hello world</parameter></invoke></tool_calls>`
 	res := ParseToolCallsDetailed(text, []string{"Write"})
@@ -218,6 +233,21 @@ func TestParseToolCallsTreatsCDATAItemOnlyBodyAsArray(t *testing.T) {
 	}
 }

+func TestParseToolCallsTreatsSingleItemCDATAAsArray(t *testing.T) {
+	text := `<tool_calls><invoke name="TodoWrite"><parameter name="todos"><![CDATA[<item>one</item>]]></parameter></invoke></tool_calls>`
+	calls := ParseToolCalls(text, []string{"TodoWrite"})
+	if len(calls) != 1 {
+		t.Fatalf("expected one TodoWrite call, got %#v", calls)
+	}
+	items, ok := calls[0].Input["todos"].([]any)
+	if !ok || len(items) != 1 {
+		t.Fatalf("expected single-item CDATA body to parse as array, got %#v", calls[0].Input["todos"])
+	}
+	if got, ok := items[0].(string); !ok || got != "one" {
+		t.Fatalf("expected single item value to stay intact, got %#v", items[0])
+	}
+}
+
 func TestParseToolCallsTreatsCDATAObjectFragmentAsObject(t *testing.T) {
 	payload := `<question><![CDATA[Pick one]]></question><options><item><label><![CDATA[A]]></label></item><item><label><![CDATA[B]]></label></item></options>`
 	text := `<tool_calls><invoke name="AskUserQuestion"><parameter name="questions"><![CDATA[` + payload + `]]></parameter></invoke></tool_calls>`
--- a/internal/toolstream/tool_sieve_xml.go
+++ b/internal/toolstream/tool_sieve_xml.go
@@ -154,6 +154,7 @@ func findPartialXMLToolTagStart(s string) int {
 		"<|tool_calls", "<|invoke", "<|parameter",
 		"<｜tool_calls", "<｜invoke", "<｜parameter",
 		"<|dsml|tool_calls", "<|dsml|invoke", "<|dsml|parameter",
+		"<｜dsml|tool_calls", "<｜dsml|invoke", "<｜dsml|parameter",
 		"<dsmltool_calls", "<dsmlinvoke", "<dsmlparameter",
 		"<dsml tool_calls", "<dsml invoke", "<dsml parameter",
 		"<dsml|tool_calls", "<dsml|invoke", "<dsml|parameter",
--- a/internal/toolstream/tool_sieve_xml_tags.go
+++ b/internal/toolstream/tool_sieve_xml_tags.go
@@ -15,6 +15,7 @@ var xmlToolCallBlockPattern = regexp.MustCompile(`(?is)((?:<tool_calls\b|<\|dsml
 // xmlToolTagsToDetect is the set of XML tag prefixes used by findToolSegmentStart.
 var xmlToolTagsToDetect = []string{
 	"<|dsml|tool_calls>", "<|dsml|tool_calls\n", "<|dsml|tool_calls ",
+	"<｜dsml|tool_calls>", "<｜dsml|tool_calls\n", "<｜dsml|tool_calls ",
 	"<|dsml|invoke ", "<|dsml|invoke\n", "<|dsml|invoke\t", "<|dsml|invoke\r",
 	"<|dsmltool_calls>", "<|dsmltool_calls\n", "<|dsmltool_calls ",
 	"<|dsmlinvoke ", "<|dsmlinvoke\n", "<|dsmlinvoke\t", "<|dsmlinvoke\r",
--- a/internal/toolstream/tool_sieve_xml_test.go
+++ b/internal/toolstream/tool_sieve_xml_test.go
@@ -745,6 +745,51 @@ func TestProcessToolSieveFullwidthPipeVariantDoesNotLeak(t *testing.T) {
 	}
 }

+// Test <｜DSML|tool_calls> with DSML invoke/parameter tags should buffer the
+// wrapper instead of leaking it before the block is complete.
+func TestProcessToolSieveFullwidthDSMLPrefixVariantDoesNotLeak(t *testing.T) {
+	var state State
+	chunks := []string{
+		"<｜DSML|tool",
+		"_calls>\n",
+		"<|DSML|invoke name=\"Bash\">\n",
+		"<|DSML|parameter name=\"command\"><![CDATA[ls -la /Users/aq/Desktop/myproject/ds2api/]]></|DSML|parameter>\n",
+		"<|DSML|parameter name=\"description\"><![CDATA[List project root contents]]></|DSML|parameter>\n",
+		"</|DSML|invoke>\n",
+		"<|DSML|invoke name=\"Bash\">\n",
+		"<|DSML|parameter name=\"command\"><![CDATA[cat /Users/aq/Desktop/myproject/ds2api/package.json 2>/dev/null || echo \"No package.json found\"]]></|DSML|parameter>\n",
+		"<|DSML|parameter name=\"description\"><![CDATA[Check for existing package.json]]></|DSML|parameter>\n",
+		"</|DSML|invoke>\n",
+		"</|DSML|tool_calls>",
+	}
+	var events []Event
+	for _, c := range chunks {
+		events = append(events, ProcessChunk(&state, c, []string{"Bash"})...)
+	}
+	events = append(events, Flush(&state, []string{"Bash"})...)
+
+	var textContent strings.Builder
+	var toolCalls int
+	var names []string
+	for _, evt := range events {
+		textContent.WriteString(evt.Content)
+		for _, call := range evt.ToolCalls {
+			toolCalls++
+			names = append(names, call.Name)
+		}
+	}
+
+	if toolCalls != 2 {
+		t.Fatalf("expected two tool calls from fullwidth DSML prefix variant, got %d events=%#v", toolCalls, events)
+	}
+	if len(names) != 2 || names[0] != "Bash" || names[1] != "Bash" {
+		t.Fatalf("expected two Bash tool calls, got %v", names)
+	}
+	if textContent.Len() != 0 {
+		t.Fatalf("expected fullwidth DSML prefix variant not to leak text, got %q", textContent.String())
+	}
+}
+
 // Test <DSML|tool_calls> with <|DSML|invoke> (DSML prefix without leading pipe on wrapper).
 func TestProcessToolSieveDSMLPrefixVariantDoesNotLeak(t *testing.T) {
 	var state State
--- a/internal/util/render.go
+++ b/internal/util/render.go
@@ -20,7 +20,7 @@ func BuildOpenAIChatCompletion(completionID, model, finalPrompt, finalThinking,
 	}
 	if len(detected) > 0 {
 		finishReason = "tool_calls"
-		messageObj["tool_calls"] = toolcall.FormatOpenAIToolCalls(detected)
+		messageObj["tool_calls"] = toolcall.FormatOpenAIToolCalls(detected, nil)
 		messageObj["content"] = nil
 	}
 	promptTokens := EstimateTokens(finalPrompt)
--- a/start.mjs
+++ b/start.mjs
@@ -126,9 +126,12 @@ function binaryExists() {

 // 查找占用端口的进程 PID
 function findPidByPort(port) {
+  const numericPort = parseInt(port, 10);
+  if (isNaN(numericPort)) return [];
+  
  try {
    if (isWindows) {
-      const output = execSync(`netstat -ano | findstr :${port} | findstr LISTENING`, {
+      const output = execSync(`netstat -ano | findstr :${numericPort} | findstr LISTENING`, {
        encoding: 'utf-8',
        shell: true,
        stdio: ['pipe', 'pipe', 'ignore'],
@@ -141,7 +144,7 @@ function findPidByPort(port) {
      }
      return [...pids];
    } else {
-      const output = execSync(`lsof -ti :${port}`, {
+      const output = execSync(`lsof -ti :${numericPort}`, {
        encoding: 'utf-8',
        stdio: ['pipe', 'pipe', 'ignore'],
      });
@@ -217,7 +220,7 @@ async function installFrontendDeps() {
    const proc = spawn('npm', ['ci', '--registry', MIRRORS.npm], {
      cwd: CONFIG.webuiDir,
      stdio: 'inherit',
-      shell: true,
+      shell: isWindows,
    });
    proc.on('close', code => code === 0 ? resolve() : reject(new Error('前端依赖安装失败')));
  });
@@ -239,7 +242,7 @@ async function buildBackend() {
    const proc = spawn('go', ['build', '-o', BINARY, './cmd/ds2api'], {
      cwd: __dirname,
      stdio: 'inherit',
-      shell: true,
+      shell: isWindows,
      env: { ...process.env, GOPROXY: MIRRORS.goproxy },
    });
    proc.on('close', code => code === 0 ? resolve() : reject(new Error('后端编译失败')));
@@ -257,22 +260,21 @@ async function buildWebui() {
  return new Promise((resolve, reject) => {
    const proc = spawn(
      'npm', ['run', 'build', '--', '--outDir', CONFIG.staticAdminDir, '--emptyOutDir'],
-      { cwd: CONFIG.webuiDir, stdio: 'inherit', shell: true }
+      { cwd: CONFIG.webuiDir, stdio: 'inherit', shell: isWindows }
    );
    proc.on('close', code => code === 0 ? resolve() : reject(new Error('前端构建失败')));
  });
 }

 // 启动后端（开发模式：go run，无需预编译）
-async function startBackendDev() {
-  if (!checkGo()) throw new Error('未找到 Go，请先安装 Go (https://go.dev/dl/)');
-  log.info(`启动后端（go run）... 本地 http://127.0.0.1:${CONFIG.port}  绑定 0.0.0.0:${CONFIG.port}`);
-  const proc = spawn('go', ['run', './cmd/ds2api'], {
+async function startBackendDev() {
+  if (!checkGo()) throw new Error('未找到 Go，请先安装 Go (https://go.dev/dl/)');
+  log.info(`启动后端（go run）... 本地 http://127.0.0.1:${CONFIG.port}  绑定 0.0.0.0:${CONFIG.port}`);
+  const proc = spawn('go', ['run', './cmd/ds2api'], {
    cwd: __dirname,
    stdio: 'inherit',
-    shell: true,
-    env: {
-      ...process.env,
+    shell: isWindows,
+    env: { ...process.env,
      PORT: CONFIG.port,
      LOG_LEVEL: CONFIG.logLevel,
      DS2API_ADMIN_KEY: CONFIG.adminKey,
@@ -284,13 +286,13 @@ async function startBackendDev() {
 }

 // 启动后端（生产模式：运行编译好的二进制）
-async function startBackendProd() {
-  if (!binaryExists()) {
-    log.warn('未找到编译产物，正在编译...');
-    await buildBackend();
-  }
-  log.info(`启动后端（二进制）... 本地 http://127.0.0.1:${CONFIG.port}  绑定 0.0.0.0:${CONFIG.port}`);
-  const proc = spawn(BINARY, [], {
+async function startBackendProd() {
+  if (!binaryExists()) {
+    log.warn('未找到编译产物，正在编译...');
+    await buildBackend();
+  }
+  log.info(`启动后端（二进制）... 本地 http://127.0.0.1:${CONFIG.port}  绑定 0.0.0.0:${CONFIG.port}`);
+  const proc = spawn(BINARY, [], {
    cwd: __dirname,
    stdio: 'inherit',
    shell: false,
@@ -323,14 +325,14 @@ async function startFrontend() {
 }

 // 显示状态信息
-function showStatus() {
-  console.log('\n' + '─'.repeat(50));
-  log.success(`后端 API:  http://127.0.0.1:${CONFIG.port}`);
-  log.success(`管理界面: http://127.0.0.1:${CONFIG.port}/admin`);
-  log.info(`后端绑定:  0.0.0.0:${CONFIG.port} (可通过局域网 IP 访问)`);
-  if (existsSync(CONFIG.webuiDir)) {
-    log.success(`前端 Dev:  http://localhost:${CONFIG.frontendPort}`);
-  }
+function showStatus() {
+  console.log('\n' + '─'.repeat(50));
+  log.success(`后端 API:  http://127.0.0.1:${CONFIG.port}`);
+  log.success(`管理界面: http://127.0.0.1:${CONFIG.port}/admin`);
+  log.info(`后端绑定:  0.0.0.0:${CONFIG.port} (可通过局域网 IP 访问)`);
+  if (existsSync(CONFIG.webuiDir)) {
+    log.success(`前端 Dev:  http://localhost:${CONFIG.frontendPort}`);
+  }
  console.log('─'.repeat(50));
  log.info('按 Ctrl+C 停止所有服务\n');
 }
--- a/tests/node/stream-tool-sieve.test.js
+++ b/tests/node/stream-tool-sieve.test.js
@@ -104,6 +104,13 @@ test('parseToolCalls keeps canonical XML examples inside DSML CDATA', () => {
  assert.deepEqual(calls[0].input, { path: 'notes.md', content });
 });

+test('parseToolCalls preserves simple inline markup inside CDATA as text', () => {
+  const payload = '<tool_calls><invoke name="Write"><parameter name="description"><![CDATA[<b>urgent</b>]]></parameter></invoke></tool_calls>';
+  const calls = parseToolCalls(payload, ['Write']);
+  assert.equal(calls.length, 1);
+  assert.equal(calls[0].input.description, '<b>urgent</b>');
+});
+
 test('parseToolCalls recovers when CDATA never closes inside a valid wrapper', () => {
  const payload = '<tool_calls><invoke name="Write"><parameter name="content"><![CDATA[hello world</parameter></invoke></tool_calls>';
  const calls = parseToolCalls(payload, ['Write']);
@@ -174,6 +181,37 @@ test('parseToolCalls treats CDATA item-only body as array', () => {
  ]);
 });

+test('parseToolCalls treats single-item CDATA body as array', () => {
+  const payload = '<tool_calls><invoke name="TodoWrite"><parameter name="todos"><![CDATA[<item>one</item>]]></parameter></invoke></tool_calls>';
+  const calls = parseToolCalls(payload, ['TodoWrite']);
+  assert.equal(calls.length, 1);
+  assert.deepEqual(calls[0].input.todos, ['one']);
+});
+
+test('formatOpenAIStreamToolCalls normalizes camelCase inputSchema string fields', () => {
+  const formatted = formatOpenAIStreamToolCalls([
+    { name: 'Write', input: { content: { message: 'hi' }, taskId: 1 } },
+  ], new Map(), [
+    { name: 'Write', inputSchema: { type: 'object', properties: { content: { type: 'string' }, taskId: { type: 'string' } } } },
+  ]);
+  assert.equal(formatted.length, 1);
+  const args = JSON.parse(formatted[0].function.arguments);
+  assert.equal(args.content, '{"message":"hi"}');
+  assert.equal(args.taskId, '1');
+});
+
+test('formatOpenAIStreamToolCalls preserves arrays when schema says array', () => {
+  const todos = [{ content: 'x', status: 'pending', priority: 'high' }];
+  const formatted = formatOpenAIStreamToolCalls([
+    { name: 'todowrite', input: { todos } },
+  ], new Map(), [
+    { name: 'todowrite', inputSchema: { type: 'object', properties: { todos: { type: 'array', items: { type: 'object' } } } } },
+  ]);
+  assert.equal(formatted.length, 1);
+  const args = JSON.parse(formatted[0].function.arguments);
+  assert.deepEqual(args.todos, todos);
+});
+
 test('parseToolCalls treats CDATA object fragment as object', () => {
  const fragment = '<question><![CDATA[Pick one]]></question><options><item><label><![CDATA[A]]></label></item><item><label><![CDATA[B]]></label></item></options>';
  const payload = `<tool_calls><invoke name="AskUserQuestion"><parameter name="questions"><![CDATA[${fragment}]]></parameter></invoke></tool_calls>`;
@@ -400,6 +438,31 @@ test('sieve emits tool_calls when DSML tag spans multiple chunks', () => {
  assert.equal(finalCalls[0].name, 'read_file');
 });

+test('sieve emits tool_calls when fullwidth DSML prefix variant spans multiple chunks', () => {
+  const events = runSieve(
+    [
+      '<｜DSML|tool',
+      '_calls>\n',
+      '<|DSML|invoke name="Bash">\n',
+      '<|DSML|parameter name="command"><![CDATA[ls -la /Users/aq/Desktop/myproject/ds2api/]]></|DSML|parameter>\n',
+      '<|DSML|parameter name="description"><![CDATA[List project root contents]]></|DSML|parameter>\n',
+      '</|DSML|invoke>\n',
+      '<|DSML|invoke name="Bash">\n',
+      '<|DSML|parameter name="command"><![CDATA[cat /Users/aq/Desktop/myproject/ds2api/package.json 2>/dev/null || echo "No package.json found"]]></|DSML|parameter>\n',
+      '<|DSML|parameter name="description"><![CDATA[Check for existing package.json]]></|DSML|parameter>\n',
+      '</|DSML|invoke>\n',
+      '</|DSML|tool_calls>',
+    ],
+    ['Bash'],
+  );
+  const leakedText = collectText(events);
+  const finalCalls = events.filter((evt) => evt.type === 'tool_calls').flatMap((evt) => evt.calls || []);
+  assert.equal(leakedText, '');
+  assert.equal(finalCalls.length, 2);
+  assert.equal(finalCalls[0].name, 'Bash');
+  assert.equal(finalCalls[1].name, 'Bash');
+});
+
 test('sieve keeps long XML tool calls buffered until the closing tag arrives', () => {
  const longContent = 'x'.repeat(4096);
  const splitAt = longContent.length / 2;
--- a/webui/src/features/chatHistory/ChatHistoryContainer.jsx
+++ b/webui/src/features/chatHistory/ChatHistoryContainer.jsx
@@ -1,4 +1,4 @@
-import { ArrowDown, ArrowUp, Bot, ChevronDown, Clock3, Loader2, MessageSquareText, RefreshCcw, Sparkles, Trash2, UserRound, X } from 'lucide-react'
+import { ArrowDown, ArrowUp, Bot, ChevronDown, Clock3, Copy, Download, Loader2, MessageSquareText, RefreshCcw, Sparkles, Trash2, UserRound, X } from 'lucide-react'
 import { useEffect, useRef, useState } from 'react'
 import clsx from 'clsx'

@@ -9,9 +9,14 @@ const DISABLED_LIMIT = 0
 const MESSAGE_COLLAPSE_AT = 700
 const VIEW_MODE_KEY = 'ds2api_chat_history_view_mode'
 const BEGIN_SENTENCE_MARKER = '<｜begin▁of▁sentence｜>'
+const SYSTEM_MARKER = '<｜System｜>'
 const USER_MARKER = '<｜User｜>'
 const ASSISTANT_MARKER = '<｜Assistant｜>'
+const TOOL_MARKER = '<｜Tool｜>'
+const END_INSTRUCTIONS_MARKER = '<｜end▁of▁instructions｜>'
 const END_SENTENCE_MARKER = '<｜end▁of▁sentence｜>'
+const END_TOOL_RESULTS_MARKER = '<｜end▁of▁toolresults｜>'
+const CURRENT_INPUT_FILE_PROMPT = 'The current request and prior conversation context have already been provided. Answer the latest user request directly.'

 function formatDateTime(value, lang) {
    if (!value) return '-'
@@ -109,6 +114,54 @@ function MergeModeIcon() {
    )
 }

+function downloadTextFile(filename, text) {
+    const blob = new Blob([text], { type: 'text/plain;charset=utf-8' })
+    const url = URL.createObjectURL(blob)
+    const link = document.createElement('a')
+    link.href = url
+    link.download = filename
+    document.body.appendChild(link)
+    link.click()
+    document.body.removeChild(link)
+    URL.revokeObjectURL(url)
+}
+
+function fallbackCopyText(text) {
+    const textArea = document.createElement('textarea')
+    textArea.value = text
+    textArea.setAttribute('readonly', '')
+    textArea.style.position = 'fixed'
+    textArea.style.top = '-9999px'
+    textArea.style.left = '-9999px'
+
+    document.body.appendChild(textArea)
+    textArea.focus()
+    textArea.select()
+
+    let copied = false
+    try {
+        copied = document.execCommand('copy')
+    } finally {
+        document.body.removeChild(textArea)
+    }
+
+    if (!copied) {
+        throw new Error('copy failed')
+    }
+}
+
+async function copyTextWithFallback(text) {
+    try {
+        if (navigator.clipboard?.writeText) {
+            await navigator.clipboard.writeText(text)
+            return
+        }
+    } catch {
+        // Fall through to execCommand fallback.
+    }
+    fallbackCopyText(text)
+}
+
 function skipWhitespace(text, start) {
    let cursor = start
    while (cursor < text.length && /\s/.test(text[cursor])) {
@@ -131,7 +184,9 @@ function parseStrictHistoryMessages(historyText) {

    while (cursor < transcript.length) {
        if (expectedRole === null) {
-            if (transcript.startsWith(USER_MARKER, cursor)) {
+            if (transcript.startsWith(SYSTEM_MARKER, cursor)) {
+                expectedRole = 'system'
+            } else if (transcript.startsWith(USER_MARKER, cursor)) {
                expectedRole = 'user'
            } else if (transcript.startsWith(ASSISTANT_MARKER, cursor)) {
                expectedRole = 'assistant'
@@ -142,13 +197,32 @@ function parseStrictHistoryMessages(historyText) {
            }
        }

+        if (transcript.startsWith(SYSTEM_MARKER, cursor)) {
+            if (expectedRole !== 'system') return null
+            cursor += SYSTEM_MARKER.length
+            const nextInstructionsEnd = transcript.indexOf(END_INSTRUCTIONS_MARKER, cursor)
+            if (nextInstructionsEnd < 0) return null
+            parsed.push({
+                role: 'system',
+                content: transcript.slice(cursor, nextInstructionsEnd),
+            })
+            cursor = nextInstructionsEnd + END_INSTRUCTIONS_MARKER.length
+            expectedRole = 'user'
+            continue
+        }
+
        if (transcript.startsWith(USER_MARKER, cursor)) {
-            if (expectedRole !== 'user') return null
+            if (expectedRole !== 'user' && expectedRole !== 'user_or_tool' && expectedRole !== 'assistant_or_user') return null
            cursor += USER_MARKER.length
            const nextAssistant = transcript.indexOf(ASSISTANT_MARKER, cursor)
+            const nextTool = transcript.indexOf(TOOL_MARKER, cursor)
            const nextSentenceEnd = transcript.indexOf(END_SENTENCE_MARKER, cursor)
-            if (nextAssistant < 0) return null
-            if (nextSentenceEnd >= 0 && nextSentenceEnd < nextAssistant) {
+            let nextRoleIndex = nextAssistant
+            if (nextRoleIndex < 0 || (nextTool >= 0 && nextTool < nextRoleIndex)) {
+                nextRoleIndex = nextTool
+            }
+            if (nextRoleIndex < 0) return null
+            if (nextSentenceEnd >= 0 && nextSentenceEnd < nextRoleIndex) {
                const assistantStart = skipWhitespace(transcript, nextSentenceEnd + END_SENTENCE_MARKER.length)
                if (!transcript.startsWith(ASSISTANT_MARKER, assistantStart)) return null
                parsed.push({
@@ -161,21 +235,26 @@ function parseStrictHistoryMessages(historyText) {
            }
            parsed.push({
                role: 'user',
-                content: transcript.slice(cursor, nextAssistant),
+                content: transcript.slice(cursor, nextRoleIndex),
            })
-            const assistantStart = nextAssistant + ASSISTANT_MARKER.length
+            if (transcript.startsWith(TOOL_MARKER, nextRoleIndex)) {
+                cursor = nextRoleIndex
+                expectedRole = 'tool'
+                continue
+            }
+            const assistantStart = nextRoleIndex + ASSISTANT_MARKER.length
            if (transcript.indexOf(END_SENTENCE_MARKER, assistantStart) < 0) {
                trailingAssistantPromptOnly = true
                cursor = assistantStart
                break
            }
-            cursor = nextAssistant
+            cursor = nextRoleIndex
            expectedRole = 'assistant'
            continue
        }

        if (transcript.startsWith(ASSISTANT_MARKER, cursor)) {
-            if (expectedRole !== 'assistant') return null
+            if (expectedRole !== 'assistant' && expectedRole !== 'assistant_or_user') return null
            cursor += ASSISTANT_MARKER.length
            const nextSentenceEnd = transcript.indexOf(END_SENTENCE_MARKER, cursor)
            if (nextSentenceEnd < 0) return null
@@ -184,11 +263,28 @@ function parseStrictHistoryMessages(historyText) {
                content: transcript.slice(cursor, nextSentenceEnd),
            })
            cursor = nextSentenceEnd + END_SENTENCE_MARKER.length
-            expectedRole = 'user'
+            expectedRole = 'user_or_tool'
            continue
        }

-        if (parsed.length && expectedRole === 'user') break
+        if (transcript.startsWith(TOOL_MARKER, cursor)) {
+            if (expectedRole !== 'tool' && expectedRole !== 'user' && expectedRole !== 'user_or_tool') return null
+            cursor += TOOL_MARKER.length
+            const nextToolResultsEnd = transcript.indexOf(END_TOOL_RESULTS_MARKER, cursor)
+            if (nextToolResultsEnd < 0) return null
+            parsed.push({
+                role: 'tool',
+                content: transcript.slice(cursor, nextToolResultsEnd),
+            })
+            cursor = nextToolResultsEnd + END_TOOL_RESULTS_MARKER.length
+            expectedRole = 'assistant_or_user'
+            continue
+        }
+
+        if (
+            parsed.length
+            && (expectedRole === 'user' || expectedRole === 'user_or_tool' || expectedRole === 'assistant_or_user')
+        ) break
        if (transcript.slice(cursor).trim() === '') break
        return null
    }
@@ -214,6 +310,14 @@ function buildListModeMessages(item, t) {
        return { messages: liveMessages, historyMerged: false }
    }

+    const placeholderOnly = liveMessages.length === 1
+        && String(liveMessages[0]?.role || '').trim().toLowerCase() === 'user'
+        && String(liveMessages[0]?.content || '').trim() === CURRENT_INPUT_FILE_PROMPT
+
+    if (placeholderOnly) {
+        return { messages: historyMessages, historyMerged: true }
+    }
+
    const insertAt = liveMessages.findIndex(message => {
        const role = String(message?.role || '').trim().toLowerCase()
        return role !== 'system' && role !== 'developer'
@@ -275,8 +379,28 @@ function RequestMessages({ item, t, messages }) {
    )
 }

-function MergedPromptView({ item, t }) {
+function MergedPromptView({ item, t, onMessage }) {
    const merged = item?.final_prompt || ''
+    const mergedFilename = `Merged_${item?.id || 'prompt'}.txt`
+
+    const handleCopy = async () => {
+        try {
+            await copyTextWithFallback(merged)
+            onMessage?.('success', t('chatHistory.copySuccess'))
+        } catch {
+            onMessage?.('error', t('chatHistory.copyFailed'))
+        }
+    }
+
+    const handleDownload = () => {
+        try {
+            downloadTextFile(mergedFilename, merged)
+            onMessage?.('success', t('chatHistory.downloadSuccess'))
+        } catch {
+            onMessage?.('error', t('chatHistory.downloadFailed'))
+        }
+    }
+
    return (
        <div
            className="max-w-4xl mx-auto rounded-2xl border px-5 py-4"
@@ -285,8 +409,28 @@ function MergedPromptView({ item, t }) {
                borderColor: 'rgba(231, 176, 8, 0.45)',
            }}
        >
-            <div className="text-[11px] uppercase tracking-[0.12em] text-[#5b4300] mb-3">
-                {t('chatHistory.mergedInput')}
+            <div className="mb-3 flex items-center justify-between gap-3">
+                <div className="text-[11px] uppercase tracking-[0.12em] text-[#5b4300]">
+                    {t('chatHistory.mergedInput')}
+                </div>
+                <div className="flex items-center gap-2">
+                    <button
+                        type="button"
+                        onClick={handleCopy}
+                        className="h-8 w-8 rounded-lg text-[#5b4300] hover:text-black hover:bg-[#fff8db]/45 flex items-center justify-center transition-colors"
+                        title={t('chatHistory.copyMerged')}
+                    >
+                        <Copy className="w-4 h-4" />
+                    </button>
+                    <button
+                        type="button"
+                        onClick={handleDownload}
+                        className="h-8 w-8 rounded-lg text-[#5b4300] hover:text-black hover:bg-[#fff8db]/45 flex items-center justify-center transition-colors"
+                        title={t('chatHistory.downloadMerged')}
+                    >
+                        <Download className="w-4 h-4" />
+                    </button>
+                </div>
            </div>
            <div className="text-sm leading-7 text-[#2f2200] whitespace-pre-wrap break-words font-mono">
                <ExpandableText
@@ -300,14 +444,53 @@ function MergedPromptView({ item, t }) {
    )
 }

-function HistoryTextView({ item, t }) {
+function HistoryTextView({ item, t, onMessage }) {
    const historyText = (item?.history_text || '').trim()
    if (!historyText) return null
+    const historyFilename = `History_${item?.id || 'history'}.txt`
+
+    const handleCopy = async () => {
+        try {
+            await copyTextWithFallback(historyText)
+            onMessage?.('success', t('chatHistory.copySuccess'))
+        } catch {
+            onMessage?.('error', t('chatHistory.copyFailed'))
+        }
+    }
+
+    const handleDownload = () => {
+        try {
+            downloadTextFile(historyFilename, historyText)
+            onMessage?.('success', t('chatHistory.downloadSuccess'))
+        } catch {
+            onMessage?.('error', t('chatHistory.downloadFailed'))
+        }
+    }

    return (
        <div className="max-w-4xl mx-auto rounded-2xl border border-border bg-background px-5 py-4">
-            <div className="text-[11px] uppercase tracking-[0.12em] text-muted-foreground mb-3 text-left">
-                HISTORY
+            <div className="mb-3 flex items-center justify-between gap-3">
+                <div className="text-[11px] uppercase tracking-[0.12em] text-muted-foreground text-left">
+                    HISTORY
+                </div>
+                <div className="flex items-center gap-2">
+                    <button
+                        type="button"
+                        onClick={handleCopy}
+                        className="h-8 w-8 rounded-lg border border-border bg-background text-muted-foreground hover:text-foreground hover:bg-secondary/70 flex items-center justify-center"
+                        title={t('chatHistory.copyHistory')}
+                    >
+                        <Copy className="w-4 h-4" />
+                    </button>
+                    <button
+                        type="button"
+                        onClick={handleDownload}
+                        className="h-8 w-8 rounded-lg border border-border bg-background text-muted-foreground hover:text-foreground hover:bg-secondary/70 flex items-center justify-center"
+                        title={t('chatHistory.downloadHistory')}
+                    >
+                        <Download className="w-4 h-4" />
+                    </button>
+                </div>
            </div>
            <div className="text-sm leading-7 text-foreground whitespace-pre-wrap break-words font-mono">
                <ExpandableText
@@ -322,18 +505,18 @@ function HistoryTextView({ item, t }) {
    )
 }

-function DetailConversation({ selectedItem, t, viewMode, detailScrollRef, assistantStartRef, bottomButtonClassName }) {
+function DetailConversation({ selectedItem, t, viewMode, detailScrollRef, assistantStartRef, bottomButtonClassName, onMessage }) {
    if (!selectedItem) return null
    const listModeState = viewMode === 'list' ? buildListModeMessages(selectedItem, t) : null
    const showHistoryAtTop = viewMode !== 'list' || !listModeState?.historyMerged

    return (
        <>
-            {showHistoryAtTop && <HistoryTextView item={selectedItem} t={t} />}
+            {showHistoryAtTop && <HistoryTextView item={selectedItem} t={t} onMessage={onMessage} />}

            {viewMode === 'list'
                ? <RequestMessages item={selectedItem} t={t} messages={listModeState?.messages} />
-                : <MergedPromptView item={selectedItem} t={t} />}
+                : <MergedPromptView item={selectedItem} t={t} onMessage={onMessage} />}

            <div ref={assistantStartRef} className="flex gap-4 max-w-4xl mx-auto">
                <div className={clsx(
@@ -908,6 +1091,7 @@ export default function ChatHistoryContainer({ authFetch, onMessage }) {
                                detailScrollRef={detailScrollRef}
                                assistantStartRef={assistantStartRef}
                                bottomButtonClassName="absolute right-5 bottom-5"
+                                onMessage={onMessage}
                            />
                        )}
                    </div>
--- a/webui/src/locales/en.json
+++ b/webui/src/locales/en.json
@@ -284,6 +284,14 @@
        "selectPrompt": "Select a record on the left to view details.",
        "mergedInput": "Final message sent to DeepSeek",
        "emptyMergedPrompt": "No merged prompt is available.",
+        "copyHistory": "Copy HISTORY",
+        "downloadHistory": "Download HISTORY",
+        "copyMerged": "Copy merged prompt",
+        "downloadMerged": "Download merged prompt",
+        "copySuccess": "Copied successfully.",
+        "copyFailed": "Copy failed.",
+        "downloadSuccess": "Downloaded successfully.",
+        "downloadFailed": "Download failed.",
        "expand": "Expand",
        "collapse": "Collapse",
        "reasoningTrace": "Reasoning Trace",
--- a/webui/src/locales/zh.json
+++ b/webui/src/locales/zh.json
@@ -284,6 +284,14 @@
        "selectPrompt": "从左侧选择一条记录查看详情。",
        "mergedInput": "最终发送给 DeepSeek 的完整消息",
        "emptyMergedPrompt": "没有可展示的完整消息。",
+        "copyHistory": "复制 HISTORY",
+        "downloadHistory": "下载 HISTORY",
+        "copyMerged": "复制完整消息",
+        "downloadMerged": "下载完整消息",
+        "copySuccess": "复制成功",
+        "copyFailed": "复制失败",
+        "downloadSuccess": "下载成功",
+        "downloadFailed": "下载失败",
        "expand": "展开全部",
        "collapse": "收起",
        "reasoningTrace": "思维链过程",
Author	SHA1	Message	Date
CJACK.	d7125ea106	Bump version from 4.1.2 to 4.1.3	2026-04-29 07:55:48 +08:00
CJACK.	929d9a8ef7	Merge pull request #352 from shern-point/fix/tool-string-schema-protection Fix/tool type schema protection	2026-04-29 07:51:21 +08:00
CJACK.	c03f733b83	Merge pull request #353 from Gingiris/docs/add-toc docs: add Table of Contents to README.MD and README.en.md	2026-04-29 07:50:54 +08:00
Gingiris	047fc9bee2	docs: add Table of Contents to README.MD and README.en.md Both READMEs are 400+ lines with 14 top-level sections and multiple subsections but have no navigation aid. Add a Table of Contents at the top of each file to help readers quickly find relevant sections. Changes: - README.MD: add 目录 section with links to all h2/h3 headings - README.en.md: add Table of Contents with matching structure	2026-04-28 12:18:37 -07:00
shern-point	52558838ef	docs: document request-scoped tool schema authority	2026-04-29 02:00:20 +08:00
shern-point	f1926a6ced	fix: normalize Vercel stream tool arguments by schema	2026-04-29 02:00:01 +08:00
shern-point	6e21714e23	test: cover Claude schema-aware tool normalization	2026-04-29 01:59:42 +08:00
shern-point	48c4f0df9f	fix: preserve runtime tool schemas in Claude tool output	2026-04-29 01:59:24 +08:00
shern-point	a550de30af	fix: expand shared tool schema extraction	2026-04-29 01:59:05 +08:00
CJACK.	23422e4a8e	Merge pull request #350 from ouqiting/fix_chat_histroy feat: parse split context files in list view	2026-04-29 01:34:10 +08:00
CJACK.	9c33bed403	Merge pull request #349 from RinZ27/fix-docker-non-root build: improve Docker robustness and fix potential security issues	2026-04-29 01:34:00 +08:00
ouqiting	c81294f1b7	fix(chat-history): support tool turns in parsed HISTORY list view	2026-04-29 01:27:14 +08:00
ouqiting	28d2b0410f	feat: parse split context files in list view	2026-04-29 01:15:29 +08:00
RinZ27	0c782407f5	build: improve Docker robustness and fix potential security issues	2026-04-28 23:49:54 +07:00
CJACK.	27eb73d48b	Merge pull request #346 from CJackHwang/dev Normalize string tool inputs and enhance schema protection	2026-04-28 22:06:41 +08:00
CJACK.	685b5011e4	Merge pull request #343 from livesRan/fix-429Resend-pr 支持 reference 引用标签转链接，并兼容 0 基序号映射	2026-04-28 21:47:15 +08:00
songguoliang	15e9eb3639	支持 reference 引用标签转链接，并兼容 0 基序号映射	2026-04-28 16:42:37 +08:00
CJACK.	f18e6b9b11	Bump version from 4.1.1 to 4.1.2	2026-04-28 16:39:12 +08:00
CJACK.	40ebc8e942	Merge pull request #342 from shern-point/fix/tool-string-schema-protection Fix/tool string schema protection	2026-04-28 16:37:44 +08:00
shern-point	fa3e6d040d	docs: document schema-based string tool coercion Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:48:04 +08:00
shern-point	458e4469e5	test: cover openai formatter string protection Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:47:24 +08:00
shern-point	72c8e7e9f9	test: cover responses string-protected tool arguments Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:46:43 +08:00
shern-point	b9c8e90d98	refactor: thread tool schemas through responses tool outputs Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:46:06 +08:00
shern-point	36fcba1280	test: cover chat string-protected tool arguments Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:45:35 +08:00
shern-point	801b5abce3	refactor: thread tool schemas through chat tool outputs Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:38:57 +08:00
shern-point	206c3d5479	fix: apply string protection in shared tool formatters Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:27:41 +08:00
shern-point	b2903c35ed	fix: normalize schema-declared string tool inputs Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>	2026-04-28 13:23:58 +08:00
CJACK.	b26dc8b7de	Merge pull request #338 from CJackHwang/dev refactor: update tool call parsing and stream tool sieve logic	2026-04-28 01:48:10 +08:00
CJACK	63271aea8c	refactor: update tool call parsing and stream tool sieve logic Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>	2026-04-28 01:39:32 +08:00
CJACK.	516da04bcd	Merge pull request #337 from CJackHwang/codex/revert-current-input-file-prompt [codex] revert current_input_file prompt refactor	2026-04-28 00:35:36 +08:00
CJACK	9f7b671e5e	Revert "refactor: consolidate current_input_file prompt into BuildOpenAICurrentInputContextPrompt" This reverts commit `d40888496e`.	2026-04-28 00:31:12 +08:00
@@ -1 +1 @@
 .1.1
 .1.3