n9router 0.4.49 → 0.4.51

package/app/open-sse/translator/request/claude-to-kiro.js

@@ -6,17 +6,10 @@
  * direct `claude:kiro` route in ../index.js uses; it is NOT reached through the
  * claude→openai→kiro pivot.
  *
- * It reproduces the two 400-guards that live in openai-to-kiro.js so that a
- * Claude client which omits the `tools` array on a follow-up turn (typical
- * after client-side compaction) does not trip Kiro's schema validator and get
- * "Improperly formed request" (HTTP 400):
- *
- *   1. flattenClaudeToolInteractions — when the client sent NO tools, collapse
- *      every tool_use / tool_result block to plain text so no structured tool
- *      reference survives to trigger the "tools required" rule.
- *   2. reconcileOrphanedToolResults — when tools ARE present, fold any
- *      tool_result whose tool_use_id has no matching tool_use back into the
- *      user text instead of leaving a dangling structured reference.
+ * After session replay it delegates to the shared Kiro conversation
+ * canonicalizer. That layer enforces adjacent one-to-one tool use/results,
+ * repairs partial parallel calls, and flattens compacted structured references
+ * that can no longer be represented safely.
  *
  * It also handles the 9router-synthetic `-agentic` / `-thinking` suffixes and
  * the `<thinking_mode>enabled</thinking_mode>` reasoning trigger, matching
@@ -24,92 +17,31 @@
  */
 import { register } from "../index.js";
 import { FORMATS } from "../formats.js";
-import { v4 as uuidv4 } from "uuid";
+import { applyKiroSessionReplay } from "../../utils/kiroSessionReplay.js";
+import { resolveContinuationId, resolveSessionIdentity } from "../../utils/sessionManager.js";
 import {
-  resolveKiroModel,
+  resolveKiroModelIntent,
+  applyKiroThinkingOverride,
   resolveKiroThinkingBudget,
   buildThinkingSystemPrefix,
   KIRO_AGENTIC_SYSTEM_PROMPT,
   resolveDefaultProfileArn,
+  buildKiroAdditionalModelRequestFieldsForModel,
+  usesKiroNativeGptEffort,
 } from "../../config/kiroConstants.js";
 import { DEFAULT_IMAGE_MIME } from "../schema/index.js";
 import { ROLE, CLAUDE_BLOCK } from "../schema/index.js";
-
-/** Stringify a tool_use input as a readable line. */
-function toolUseToText(name, input) {
-  let argStr;
-  try {
-    argStr = typeof input === "string" ? input : JSON.stringify(input ?? {});
-  } catch {
-    argStr = "{}";
-  }
-  return `[Tool call: ${name || "unknown"}(${argStr})]`;
-}
-
-/** Render a Claude tool_result block's content as a readable line. */
-function toolResultBlockToText(content) {
-  let text = "";
-  if (typeof content === "string") {
-    text = content;
-  } else if (Array.isArray(content)) {
-    text = content
-      .map((c) => (typeof c === "string" ? c : c?.text || ""))
-      .filter(Boolean)
-      .join("\n");
-  } else if (content) {
-    try {
-      text = JSON.stringify(content);
-    } catch {
-      text = "";
-    }
-  }
-  return `[Tool result: ${text}]`;
-}
-
-/**
- * When the client sent no tools, rewrite every tool_use (assistant) and
- * tool_result (user) content block into plain text. Keeps text + images.
- * Returns a new messages array; never mutates the input.
- */
-function flattenClaudeToolInteractions(messages) {
-  const out = [];
-  for (const msg of messages) {
-    if (!msg) continue;
-
-    if (msg.role === ROLE.ASSISTANT && Array.isArray(msg.content)) {
-      const parts = [];
-      for (const block of msg.content) {
-        if (block.type === CLAUDE_BLOCK.TEXT && block.text) {
-          parts.push(block.text);
-        } else if (block.type === CLAUDE_BLOCK.TOOL_USE) {
-          parts.push(toolUseToText(block.name, block.input));
-        }
-      }
-      out.push({ ...msg, content: parts.join("\n") });
-      continue;
-    }
-
-    if (msg.role === ROLE.USER && Array.isArray(msg.content)) {
-      const newContent = msg.content.map((block) =>
-        block.type === CLAUDE_BLOCK.TOOL_RESULT
-          ? { type: CLAUDE_BLOCK.TEXT, text: toolResultBlockToText(block.content) }
-          : block
-      );
-      out.push({ ...msg, content: newContent });
-      continue;
-    }
-
-    out.push(msg);
-  }
-  return out;
-}
+import {
+  canonicalizeKiroConversation,
+  normalizeKiroToolSpecs,
+} from "../concerns/kiroConversation.js";
 
 /**
  * Convert Claude messages to Kiro history + currentMessage.
  * Kiro requires alternating user/assistant turns; consecutive same-role
  * messages are merged.
  */
-function convertClaudeMessagesToKiro(messages, tools, model) {
+function convertClaudeMessagesToKiro(messages, model) {
   const history = [];
   let currentMessage = null;
 
@@ -118,27 +50,6 @@ function convertClaudeMessagesToKiro(messages, tools, model) {
   let pendingToolResults = [];
   let pendingImages = [];
   let currentRole = null;
-  let toolsInjected = false;
-
-  const clientProvidedTools = Array.isArray(tools) && tools.length > 0;
-
-  const buildToolSpecs = () =>
-    tools.map((t) => {
-      const name = t.name;
-      const description = t.description || `Tool: ${name}`;
-      const schema = t.input_schema || {};
-      const normalizedSchema =
-        Object.keys(schema).length === 0
-          ? { type: "object", properties: {}, required: [] }
-          : { ...schema, required: schema.required ?? [] };
-      return {
-        toolSpecification: {
-          name,
-          description,
-          inputSchema: { json: normalizedSchema },
-        },
-      };
-    });
 
   const flushPending = () => {
     if (currentRole === ROLE.USER) {
@@ -153,15 +64,6 @@ function convertClaudeMessagesToKiro(messages, tools, model) {
           toolResults: pendingToolResults,
         };
       }
-      // Attach tools to the first user turn only.
-      if (clientProvidedTools && !toolsInjected) {
-        if (!userMsg.userInputMessage.userInputMessageContext) {
-          userMsg.userInputMessage.userInputMessageContext = {};
-        }
-        userMsg.userInputMessage.userInputMessageContext.tools = buildToolSpecs();
-        toolsInjected = true;
-      }
-
       history.push(userMsg);
       currentMessage = userMsg;
       pendingUserContent = [];
@@ -205,7 +107,7 @@ function convertClaudeMessagesToKiro(messages, tools, model) {
             }
             pendingToolResults.push({
               toolUseId: block.tool_use_id,
-              status: "success",
+              status: block.is_error ? "error" : "success",
               content: [{ text: resultContent }],
             });
           }
@@ -252,14 +154,7 @@ function convertClaudeMessagesToKiro(messages, tools, model) {
     }
   }
 
-  // Grab tools from the first history user turn before cleanup strips them.
-  const firstHistoryTools =
-    history[0]?.userInputMessage?.userInputMessageContext?.tools;
-
   history.forEach((item) => {
-    if (item.userInputMessage?.userInputMessageContext?.tools) {
-      delete item.userInputMessage.userInputMessageContext.tools;
-    }
     if (
       item.userInputMessage?.userInputMessageContext &&
       Object.keys(item.userInputMessage.userInputMessageContext).length === 0
@@ -303,95 +198,40 @@ function convertClaudeMessagesToKiro(messages, tools, model) {
     currentMessage = { userInputMessage: { content: "", modelId: model } };
   }
 
-  // Inject tools into currentMessage after cleanup if not already present.
-  if (
-    firstHistoryTools?.length > 0 &&
-    !currentMessage.userInputMessage.userInputMessageContext?.tools
-  ) {
-    if (!currentMessage.userInputMessage.userInputMessageContext) {
-      currentMessage.userInputMessage.userInputMessageContext = {};
-    }
-    currentMessage.userInputMessage.userInputMessageContext.tools =
-      firstHistoryTools;
-  }
-
   return { history: mergedHistory, currentMessage };
 }
 
-/**
- * Fold orphaned toolResults (those whose toolUseId has no matching toolUse in
- * any assistant turn) back into the user text, removing the dangling
- * structured reference that makes Kiro 400.
- */
-function reconcileOrphanedToolResults(history, currentMessage) {
-  const validIds = new Set();
-  for (const h of history) {
-    const arm = h.assistantResponseMessage;
-    if (!arm) continue;
-    for (const tu of arm.toolUses || []) {
-      if (tu.toolUseId) validIds.add(tu.toolUseId);
-    }
-  }
-
-  const carriers = currentMessage ? [...history, currentMessage] : history;
-  for (const item of carriers) {
-    const uim = item.userInputMessage;
-    const ctx = uim?.userInputMessageContext;
-    if (!ctx?.toolResults?.length) continue;
-
-    const kept = [];
-    const salvaged = [];
-    for (const tr of ctx.toolResults) {
-      if (validIds.has(tr.toolUseId)) {
-        kept.push(tr);
-      } else {
-        const text = Array.isArray(tr.content)
-          ? tr.content.map((c) => c?.text || "").join("\n")
-          : "";
-        salvaged.push(`[Tool result: ${text}]`);
-      }
-    }
-
-    if (salvaged.length === 0) continue;
-
-    const extra = salvaged.join("\n");
-    uim.content = uim.content ? `${uim.content}\n\n${extra}` : extra;
-    ctx.toolResults = kept;
-    if (kept.length === 0 && !ctx.tools?.length) {
-      delete uim.userInputMessageContext;
-    }
+function extractClaudeSystemText(system) {
+  if (!system) return "";
+  if (typeof system === "string") return system;
+  if (Array.isArray(system)) {
+    return system.map((s) => {
+      if (typeof s === "string") return s;
+      return s?.text || "";
+    }).filter(Boolean).join("\n");
   }
+  return "";
 }
 
 /**
  * Build a Kiro payload directly from a Claude Messages API request body.
  */
 export function claudeToKiroRequest(model, body, stream, credentials) {
-  let messages = Array.isArray(body.messages) ? body.messages : [];
+  const messages = Array.isArray(body.messages) ? body.messages : [];
   const tools = Array.isArray(body.tools) ? body.tools : [];
-  const clientProvidedTools = tools.length > 0;
   const maxTokens = body.max_tokens || 32000;
   const temperature = body.temperature;
   const topP = body.top_p;
 
-  const { upstream: upstreamModel, agentic } = resolveKiroModel(model);
-  const thinkingBudget = resolveKiroThinkingBudget(body, credentials?.rawHeaders, model);
-
-  // Guard 1: no client tools → flatten all tool interactions to text.
-  if (!clientProvidedTools) {
-    messages = flattenClaudeToolInteractions(messages);
-  }
-
-  const { history, currentMessage } = convertClaudeMessagesToKiro(
-    messages,
-    tools,
-    upstreamModel
-  );
+  const modelIntent = resolveKiroModelIntent(model);
+  const { upstream: upstreamModel, agentic } = modelIntent;
+  const thinkingBody = applyKiroThinkingOverride(body, modelIntent.thinkingOverride);
+  const thinkingBudget = resolveKiroThinkingBudget(thinkingBody, credentials?.rawHeaders, modelIntent.model);
+  const additionalModelRequestFields = buildKiroAdditionalModelRequestFieldsForModel(thinkingBody, upstreamModel);
+  const usesNativeGptEffort = usesKiroNativeGptEffort(thinkingBody, upstreamModel);
 
-  // Guard 2: tools present → reconcile dangling tool_results.
-  if (clientProvidedTools) {
-    reconcileOrphanedToolResults(history, currentMessage);
-  }
+  const { specs: toolSpecs, nameMap } = normalizeKiroToolSpecs(tools);
+  const { history, currentMessage } = convertClaudeMessagesToKiro(messages, upstreamModel);
 
   // api_key / idc / external_idp must never use the shared default ARN (belongs
   // to another account → 403 "bearer token invalid"); OAuth/social fall back to it.
@@ -402,62 +242,83 @@ export function claudeToKiroRequest(model, body, stream, credentials) {
     ? (credentials?.providerSpecificData?.profileArn || "")
     : (credentials?.providerSpecificData?.profileArn || resolveDefaultProfileArn(authMethod));
 
-  let finalContent = currentMessage?.userInputMessage?.content || "";
-
-  // System prompt: pass via native systemInstruction field (Kiro/Q API supports it)
-  // and also prepend as <instructions> in user content as fallback for upstreams
-  // that don't support the native field.
-  let systemInstruction = undefined;
-  if (body.system) {
-    let systemText = "";
-    if (typeof body.system === "string") {
-      systemText = body.system;
-    } else if (Array.isArray(body.system)) {
-      systemText = body.system.map((s) => s.text || "").join("\n");
-    }
-    if (systemText) {
-      systemInstruction = systemText;
-      finalContent = `<instructions>\n${systemText}\n</instructions>\n\n${finalContent}`;
-    }
-  }
-
-  // Prefix order: thinking_mode tag, timestamp marker, then agentic prompt.
+  // Kiro CLI/KAS sends system prompt as top-level `systemPrompt`. Keep a
+  // content fallback too because the CodeWhisperer surface does not always
+  // enforce top-level systemPrompt for direct calls.
   const timestamp = new Date().toISOString();
-  const prefixParts = [];
-  if (thinkingBudget !== null) prefixParts.push(buildThinkingSystemPrefix(thinkingBudget));
-  prefixParts.push(`[Context: Current time is ${timestamp}]`);
-  if (agentic) prefixParts.push(KIRO_AGENTIC_SYSTEM_PROMPT);
-  finalContent = `${prefixParts.join("\n\n")}\n\n${finalContent}`;
-
+  const systemPromptParts = [];
+  if (thinkingBudget !== null && !usesNativeGptEffort) {
+    systemPromptParts.push(buildThinkingSystemPrefix(thinkingBudget));
+  }
+  if (agentic) systemPromptParts.push(KIRO_AGENTIC_SYSTEM_PROMPT);
+  const systemInstruction = extractClaudeSystemText(body.system);
+  if (systemInstruction) systemPromptParts.push(systemInstruction);
+  const systemPrompt = systemPromptParts.filter(Boolean).join("\n\n");
+  const currentTimeContext = `[Context: Current time is ${timestamp}]`;
+  const contentPrefix = [systemPrompt, currentTimeContext].filter(Boolean).join("\n\n");
+
+  const sessionIdentity = resolveSessionIdentity({
+    headers: credentials?.rawHeaders,
+    body,
+    connectionId: credentials?.connectionId,
+    scope: "kiro",
+  });
+  const conversationId = sessionIdentity.sessionId;
+  const continuationId = resolveContinuationId({
+    sessionId: conversationId,
+    connectionId: credentials?.connectionId,
+    scope: "kiro",
+    ephemeral: sessionIdentity.ephemeral,
+  });
+  const replay = applyKiroSessionReplay({
+    conversationId,
+    connectionId: credentials?.connectionId,
+    modelId: upstreamModel,
+    systemPrompt,
+    contentPrefix,
+    currentContentPrefix: currentTimeContext,
+    history,
+    currentMessage,
+  });
+  const canonical = canonicalizeKiroConversation({
+    history: replay.history,
+    currentMessage: replay.currentMessage,
+    modelId: upstreamModel,
+    toolSpecs,
+    nameMap,
+  });
+  const replayCurrent = canonical.currentMessage.userInputMessage;
   const userInputMessage = {
-    content: finalContent,
+    content: replayCurrent.content || "",
     modelId: upstreamModel,
     origin: "AI_EDITOR",
-    ...(currentMessage?.userInputMessage?.userInputMessageContext && {
-      userInputMessageContext:
-        currentMessage.userInputMessage.userInputMessageContext,
+    ...(replayCurrent.userInputMessageContext && {
+      userInputMessageContext: replayCurrent.userInputMessageContext,
     }),
-    ...(currentMessage?.userInputMessage?.images && {
-      images: currentMessage.userInputMessage.images,
+    ...(replayCurrent.images && {
+      images: replayCurrent.images,
     }),
   };
 
-  if (systemInstruction) {
-    userInputMessage.systemInstruction = systemInstruction;
-  }
-
   const payload = {
     conversationState: {
       chatTriggerType: "MANUAL",
-      conversationId: uuidv4(),
+      conversationId,
+      agentContinuationId: continuationId,
+      agentTaskType: "vibe",
       currentMessage: {
         userInputMessage,
       },
-      history,
+      history: canonical.history,
     },
+    agentMode: "vibe",
   };
 
   if (profileArn) payload.profileArn = profileArn;
+  if (systemPrompt) payload.systemPrompt = systemPrompt;
+  if (additionalModelRequestFields) {
+    payload.additionalModelRequestFields = additionalModelRequestFields;
+  }
 
   if (maxTokens || temperature !== undefined || topP !== undefined) {
     payload.inferenceConfig = {};

package/app/open-sse/translator/request/openai-responses.js

@@ -32,6 +32,8 @@ export function openaiResponsesToOpenAIRequest(model, body, stream, credentials)
   let pendingToolResults = [];
   let pendingReasoning = "";
   let pendingReasoningEncrypted = "";
+  const additionalTools = [];
+  const customToolNames = new Set();
 
   const inputItems = normalizeResponsesInput(body.input);
   if (!inputItems) return body;
@@ -96,7 +98,7 @@ export function openaiResponsesToOpenAIRequest(model, body, stream, credentials)
       }
       result.messages.push(msg);
     }
-    else if (itemType === RESPONSES_ITEM.FUNCTION_CALL) {
+    else if (itemType === RESPONSES_ITEM.FUNCTION_CALL || itemType === RESPONSES_ITEM.CUSTOM_TOOL_CALL) {
       // Start or append to assistant message with tool_calls
       if (!currentAssistantMsg) {
         currentAssistantMsg = {
@@ -108,16 +110,20 @@ export function openaiResponsesToOpenAIRequest(model, body, stream, credentials)
       }
       // Skip items with empty/missing name — Codex/OpenAI reject nameless tool calls (#444)
       if (!item.name || typeof item.name !== "string" || item.name.trim() === "") continue;
+      if (itemType === RESPONSES_ITEM.CUSTOM_TOOL_CALL) customToolNames.add(item.name);
+      const toolInput = itemType === RESPONSES_ITEM.CUSTOM_TOOL_CALL
+        ? { input: typeof item.input === "string" ? item.input : JSON.stringify(item.input ?? "") }
+        : item.arguments;
       currentAssistantMsg.tool_calls.push({
         id: item.call_id,
         type: OPENAI_BLOCK.FUNCTION,
         function: {
           name: item.name,
-          arguments: item.arguments
+          arguments: typeof toolInput === "string" ? toolInput : JSON.stringify(toolInput ?? {})
         }
       });
     }
-    else if (itemType === RESPONSES_ITEM.FUNCTION_CALL_OUTPUT) {
+    else if (itemType === RESPONSES_ITEM.FUNCTION_CALL_OUTPUT || itemType === RESPONSES_ITEM.CUSTOM_TOOL_CALL_OUTPUT) {
       // Flush assistant message first if exists
       if (currentAssistantMsg) {
         result.messages.push(currentAssistantMsg);
@@ -137,6 +143,9 @@ export function openaiResponsesToOpenAIRequest(model, body, stream, credentials)
         content: typeof item.output === "string" ? item.output : JSON.stringify(item.output)
       });
     }
+    else if (itemType === RESPONSES_ITEM.ADDITIONAL_TOOLS) {
+      if (Array.isArray(item.tools)) additionalTools.push(...item.tools);
+    }
     else if (itemType === RESPONSES_ITEM.REASONING) {
       // Buffer reasoning text; attached to next assistant message/function_call.
       // Also stash encrypted_content so a later openai→responses hop can restore
@@ -166,15 +175,45 @@ export function openaiResponsesToOpenAIRequest(model, body, stream, credentials)
   // explicit `name` field and cannot be represented as Chat Completions function declarations.
   // Filter them out to avoid sending nameless functionDeclarations to downstream providers
   // such as Gemini, which strictly validates function names.
-  if (body.tools && Array.isArray(body.tools)) {
-    result.tools = body.tools
+  const responseTools = [
+    ...(Array.isArray(body.tools) ? body.tools : []),
+    ...additionalTools,
+  ];
+  if (responseTools.length > 0) {
+    result.tools = responseTools
       .map(tool => {
         // Already in Chat Completions format: { type: "function", function: { name, ... } }
         if (tool.function) return tool;
-        // Responses API function tool: { type: "function", name, description, parameters }
-        // Only convert when a non-empty name is present; skip hosted tools without one.
+        // Responses API function/custom tool: { type, name, description, parameters|format }.
+        // Chat Completions has no freeform custom-tool declaration, so expose custom
+        // tools as functions with one raw `input` string while retaining their names
+        // in translator-only metadata for the response conversion.
         const name = tool.name;
         if (!name || typeof name !== "string" || name.trim() === "") return null;
+        if (tool.type === "custom") {
+          customToolNames.add(name);
+          const formatHint = [tool.format?.syntax, tool.format?.definition].filter(Boolean).join("\n");
+          return {
+            type: OPENAI_BLOCK.FUNCTION,
+            function: {
+              name,
+              description: [String(tool.description || ""), formatHint].filter(Boolean).join("\n\n"),
+              parameters: {
+                type: "object",
+                properties: {
+                  input: {
+                    type: "string",
+                    description: "Raw freeform input for this custom tool"
+                  }
+                },
+                required: ["input"],
+                additionalProperties: false
+              }
+            }
+          };
+        }
+        // Responses API function tool: { type: "function", name, description, parameters }
+        // Only convert when a non-empty name is present; skip hosted tools without one.
         return {
           type: OPENAI_BLOCK.FUNCTION,
           function: {
@@ -187,6 +226,7 @@ export function openaiResponsesToOpenAIRequest(model, body, stream, credentials)
       })
       .filter(Boolean);
   }
+  if (customToolNames.size > 0) result._customToolNames = [...customToolNames];
 
   // Cleanup Responses API specific fields
   // Map Responses-only max_output_tokens to Chat max_tokens (avoid leaking unknown field upstream)
@@ -200,6 +240,9 @@ export function openaiResponsesToOpenAIRequest(model, body, stream, credentials)
   delete result.include;
   delete result.prompt_cache_key;
   delete result.store;
+  if (typeof result.reasoning?.effort === "string") {
+    result.reasoning_effort = result.reasoning.effort;
+  }
   delete result.reasoning;
 
   return result;
@@ -376,6 +419,7 @@ export function openaiToOpenAIResponsesRequest(model, body, stream, credentials)
   if (body.top_p !== undefined) result.top_p = body.top_p;
   if (body.reasoning !== undefined) result.reasoning = body.reasoning;
   if (body.reasoning_effort !== undefined) result.reasoning = { effort: body.reasoning_effort, summary: "auto" };
+  if (body.service_tier !== undefined) result.service_tier = body.service_tier;
 
   return result;
 }