@quilltap/plugin-utils 2.4.0 → 2.5.0

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 All notable changes to @quilltap/plugin-utils will be documented in this file.
 
+## [2.5.0] - 2026-08-26
+
+### Added
+
+- `OpenAICompatibleProvider` gained a `protected buildRequestBody(params, stream)` — the one Chat Completions body build both `sendMessage` and `streamMessage` now call, replacing two byte-identical copies of the message mapping and body literal. Behavior-preserving: the streaming keys (`stream`, `stream_options`) keep their historical position between `stop` and `user`, so the serialized body is unchanged. Subclasses that keep the base send/stream loops get a single override point for reshaping the body.
+
 ## [2.4.0] - 2026-08-19
 
 ### Added

package/dist/index.d.mts

@@ -124,6 +124,6 @@ declare function getQuilltapUserAgent(): string;
  * Version of the plugin-utils package.
  * Can be used at runtime to check compatibility.
  */
-declare const PLUGIN_UTILS_VERSION = "2.4.0";
+declare const PLUGIN_UTILS_VERSION = "2.5.0";
 
 export { BUILTIN_TOOL_NAMES, PLUGIN_UTILS_VERSION, __clearQuilltapVersion, __injectQuilltapVersion, getBuiltinToolNames, getQuilltapUserAgent, getQuilltapVersion };

package/dist/index.d.ts

@@ -124,6 +124,6 @@ declare function getQuilltapUserAgent(): string;
  * Version of the plugin-utils package.
  * Can be used at runtime to check compatibility.
  */
-declare const PLUGIN_UTILS_VERSION = "2.4.0";
+declare const PLUGIN_UTILS_VERSION = "2.5.0";
 
 export { BUILTIN_TOOL_NAMES, PLUGIN_UTILS_VERSION, __clearQuilltapVersion, __injectQuilltapVersion, getBuiltinToolNames, getQuilltapUserAgent, getQuilltapVersion };

package/dist/index.js

@@ -639,16 +639,21 @@ var OpenAICompatibleProvider = class {
     return out;
   }
   /**
-   * Sends a message and returns the complete response.
+   * Build the Chat Completions request body — the ONE build both
+   * {@link sendMessage} and {@link streamMessage} call, so the streaming and
+   * non-streaming shapes cannot drift apart. Maps Quilltap messages to the
+   * OpenAI wire format (tool results and assistant tool calls included),
+   * applies {@link applySystemMessagePolicy}, assembles the body literal, and
+   * finishes with {@link applyProfileParams}.
    *
-   * @param params - LLM parameters including messages, model, and settings
-   * @param apiKey - API key for authentication
-   * @returns Complete LLM response with content and usage statistics
+   * Subclasses that keep the base class's send/stream loops can override this
+   * to reshape the body in one place.
+   *
+   * @param params - LLM parameters for the request
+   * @param stream - Whether the body is for a streaming request; adds
+   *   `stream: true` and `stream_options: { include_usage: true }`
    */
-  async sendMessage(params, apiKey) {
-    this.validateApiKeyRequirement(apiKey);
-    const attachmentResults = this.collectAttachmentFailures(params);
-    const client = this.createClient(apiKey);
+  buildRequestBody(params, stream) {
     const mappedMessages = params.messages.filter((m) => {
       if (m.role === "tool" && !m.toolCallId) return false;
       return true;
@@ -684,12 +689,30 @@ var OpenAICompatibleProvider = class {
       max_tokens: params.maxTokens ?? 4096,
       top_p: params.topP ?? 1,
       stop: params.stop,
+      // Spread here — not appended after the literal — so the streaming keys
+      // keep their historical position between `stop` and `user` and the
+      // serialized body stays byte-identical.
+      ...stream ? { stream: true, stream_options: { include_usage: true } } : {},
       ...params.cacheKey ? { user: params.cacheKey } : {},
       // Tools arrive only once the runtime gate (the profile's "Allow tool
       // use") has passed, so no separate guard is needed here.
       ...params.tools && params.tools.length > 0 ? { tools: params.tools, tool_choice: params.toolChoice ?? "auto" } : {}
     };
     this.applyProfileParams(body, params);
+    return body;
+  }
+  /**
+   * Sends a message and returns the complete response.
+   *
+   * @param params - LLM parameters including messages, model, and settings
+   * @param apiKey - API key for authentication
+   * @returns Complete LLM response with content and usage statistics
+   */
+  async sendMessage(params, apiKey) {
+    this.validateApiKeyRequirement(apiKey);
+    const attachmentResults = this.collectAttachmentFailures(params);
+    const client = this.createClient(apiKey);
+    const body = this.buildRequestBody(params, false);
     try {
       const response = await client.chat.completions.create(
         body,
@@ -729,47 +752,7 @@ var OpenAICompatibleProvider = class {
     this.validateApiKeyRequirement(apiKey);
     const attachmentResults = this.collectAttachmentFailures(params);
     const client = this.createClient(apiKey);
-    const mappedMessages = params.messages.filter((m) => {
-      if (m.role === "tool" && !m.toolCallId) return false;
-      return true;
-    }).map((m) => {
-      if (m.role === "tool" && m.toolCallId) {
-        return {
-          role: "tool",
-          tool_call_id: m.toolCallId,
-          content: m.content
-        };
-      }
-      if (m.role === "assistant" && m.toolCalls && m.toolCalls.length > 0) {
-        return {
-          role: "assistant",
-          content: m.content || null,
-          tool_calls: m.toolCalls.map((tc) => ({
-            id: tc.id,
-            type: tc.type,
-            function: tc.function
-          }))
-        };
-      }
-      return {
-        role: m.role,
-        content: m.content
-      };
-    });
-    const messages = this.applySystemMessagePolicy(mappedMessages);
-    const body = {
-      model: params.model,
-      messages,
-      temperature: params.temperature ?? 0.7,
-      max_tokens: params.maxTokens ?? 4096,
-      top_p: params.topP ?? 1,
-      stop: params.stop,
-      stream: true,
-      stream_options: { include_usage: true },
-      ...params.cacheKey ? { user: params.cacheKey } : {},
-      ...params.tools && params.tools.length > 0 ? { tools: params.tools, tool_choice: params.toolChoice ?? "auto" } : {}
-    };
-    this.applyProfileParams(body, params);
+    const body = this.buildRequestBody(params, true);
     try {
       const stream = await client.chat.completions.create(
         body,
@@ -1138,7 +1121,7 @@ function rewriteLocalhostUrl(url) {
 }
 
 // src/version.generated.ts
-var PLUGIN_UTILS_VERSION = "2.4.0";
+var PLUGIN_UTILS_VERSION = "2.5.0";
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   BUILTIN_TOOL_NAMES,