fix(chat): preserve durable tool replay at hosted boundary

deno.json

@@ -1,6 +1,6 @@
 {
   "name": "veryfront",
-  "version": "0.1.1133",
+  "version": "0.1.1141",
   "license": "Apache-2.0",
   "nodeModulesDir": "auto",
   "minimumDependencyAge": {

docs/api-reference/veryfront/chat.md

@@ -531,44 +531,44 @@ import { compactForStep, compactHistoricalUiMessageToolInputs, compactOldToolInp
 
 | Name | Description | Source |
 |------|-------------|--------|
-| `DEFAULT_MESSAGE_PREP_LIMITS` | Default limits for chat history preparation. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L35) |
+| `DEFAULT_MESSAGE_PREP_LIMITS` | Default limits for chat history preparation. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L36) |
 
 #### Functions
 
 | Name | Description | Source |
 |------|-------------|--------|
-| `compactForStep` | Compact for step. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1769) |
-| `compactHistoricalUiMessageToolInputs` | Compact large historical UI-message tool inputs after matching results are available. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1491) |
-| `compactOldToolInputs` | Compact large historical tool-call inputs after matching results are available. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1404) |
-| `compressTurn` | Compress turn. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L227) |
-| `dedupeToolHistory` | Dedupe tool history. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1793) |
-| `enforceTokenBudget` | Enforce token budget. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1860) |
-| `enforceTokenBudgetWithTurnCompression` | Enforce token budget with turn compression. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L302) |
-| `ensureToolCallInputs` | Ensure tool call inputs helper. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1738) |
-| `estimateMessageTokenBreakdown` | Estimate token categories for provider, UI, or runtime messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L180) |
-| `estimateOverhead` | Estimate overhead. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1732) |
-| `estimateTokens` | Estimate tokens. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L98) |
-| `isModelSupportedFileMediaType` | Check whether the model supports the file media type. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L658) |
-| `maskOldToolOutputs` | Mask old tool outputs. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1306) |
-| `normalizeMessageFilePartMediaTypes` | Normalizes message file part media types. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L664) |
-| `prepareProviderModelMessagesFromUiMessages` | Prepare provider model messages from UI messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1005) |
-| `repairToolPairs` | Repair tool pairs. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1589) |
-| `rewriteUnsupportedFilePartsAsAnnotations` | Rewrite unsupported file parts as annotations. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L691) |
-| `sanitizeProviderModelMessages` | Sanitize provider model messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L896) |
-| `stripPendingToolParts` | Strip pending tool parts. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L792) |
+| `compactForStep` | Compact for step. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1796) |
+| `compactHistoricalUiMessageToolInputs` | Compact large historical UI-message tool inputs after matching results are available. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1518) |
+| `compactOldToolInputs` | Compact large historical tool-call inputs after matching results are available. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1431) |
+| `compressTurn` | Compress turn. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L228) |
+| `dedupeToolHistory` | Dedupe tool history. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1820) |
+| `enforceTokenBudget` | Enforce token budget. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1887) |
+| `enforceTokenBudgetWithTurnCompression` | Enforce token budget with turn compression. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L303) |
+| `ensureToolCallInputs` | Ensure tool call inputs helper. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1765) |
+| `estimateMessageTokenBreakdown` | Estimate token categories for provider, UI, or runtime messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L181) |
+| `estimateOverhead` | Estimate overhead. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1759) |
+| `estimateTokens` | Estimate tokens. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L99) |
+| `isModelSupportedFileMediaType` | Check whether the model supports the file media type. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L659) |
+| `maskOldToolOutputs` | Mask old tool outputs. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1333) |
+| `normalizeMessageFilePartMediaTypes` | Normalizes message file part media types. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L665) |
+| `prepareProviderModelMessagesFromUiMessages` | Prepare provider model messages from UI messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1032) |
+| `repairToolPairs` | Repair tool pairs. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L1616) |
+| `rewriteUnsupportedFilePartsAsAnnotations` | Rewrite unsupported file parts as annotations. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L692) |
+| `sanitizeProviderModelMessages` | Sanitize provider model messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L923) |
+| `stripPendingToolParts` | Strip pending tool parts. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L796) |
 
 #### Types
 
 | Name | Description | Source |
 |------|-------------|--------|
-| `HistoricalToolInputCompactionDiagnostic` | Diagnostic emitted when a completed historical tool input is compacted. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L72) |
-| `HistoricalToolInputRetainedField` | Field selector retained in a historical tool-input summary. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L47) |
-| `HistoricalToolInputRetentionOptions` | Options for historical tool-input compaction. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L85) |
-| `HistoricalToolInputRetentionPolicy` | Policy for compacting a completed historical tool-call input. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L59) |
-| `HistoricalToolInputRetentionPolicyResolver` | Resolves the retention policy for a completed historical tool input. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L66) |
-| `MessagePrepLimits` | Tunable limits used while preparing chat history for model context. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L25) |
-| `MessageTokenBreakdown` | Approximate token categories for context diagnostics. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L103) |
-| `PrepareProviderModelMessagesFromUiMessagesOptions` | Options accepted by prepare provider model messages from UI messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L92) |
+| `HistoricalToolInputCompactionDiagnostic` | Diagnostic emitted when a completed historical tool input is compacted. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L73) |
+| `HistoricalToolInputRetainedField` | Field selector retained in a historical tool-input summary. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L48) |
+| `HistoricalToolInputRetentionOptions` | Options for historical tool-input compaction. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L86) |
+| `HistoricalToolInputRetentionPolicy` | Policy for compacting a completed historical tool-call input. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L60) |
+| `HistoricalToolInputRetentionPolicyResolver` | Resolves the retention policy for a completed historical tool input. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L67) |
+| `MessagePrepLimits` | Tunable limits used while preparing chat history for model context. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L26) |
+| `MessageTokenBreakdown` | Approximate token categories for context diagnostics. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L104) |
+| `PrepareProviderModelMessagesFromUiMessagesOptions` | Options accepted by prepare provider model messages from UI messages. | [source](https://github.com/veryfront/veryfront-code/blob/main/src/chat/message-prep.ts#L93) |
 
 ### `veryfront/chat/protocol`
 

docs/guides/middleware.md

@@ -90,7 +90,7 @@ const pipeline = new MiddlewarePipeline()
 
 ### Run the pipeline
 
-Use `handle()` to run the middleware chain and then your route handler. If a middleware short-circuits (returns a `Response` - e.g. a rate-limit rejection), `handle()` returns that response and your handler never runs; otherwise your handler runs as the terminal step:
+Use `handle()` to run the middleware chain and then your route handler. If a middleware short-circuits (returns a `Response`, e.g. a rate-limit rejection), `handle()` returns that response and your handler never runs; otherwise your handler runs as the terminal step:
 
 ```ts
 // app/api/users/route.ts
@@ -122,16 +122,16 @@ curl -i http://localhost:3000/api/users
 
 The response includes any headers added by the middleware that matched the request.
 
-> **`handle()` vs `execute()`.** `execute()` is a lower-level variant with **no terminal handler**: it returns the short-circuiting middleware's `Response`, or a synthesized **404 Not Found** when the chain passes through. It always resolves to a `Response` (never `undefined`), so `if (await pipeline.execute(req))` is always truthy - use `execute()` only when a middleware is always expected to produce the response. For the common "middleware, then my route handler" case, prefer `handle()`.
+> **`handle()` vs `execute()`.** `execute()` is a lower-level variant with **no terminal handler**: it returns the short-circuiting middleware's `Response`, or a synthesized **404 Not Found** when the chain passes through. It always resolves to a `Response` (never `undefined`), so `if (await pipeline.execute(req))` is always truthy. Use `execute()` only when a middleware is always expected to produce the response. For the common "middleware, then my route handler" case, prefer `handle()`.
 
 ### In-memory state across requests
 
-Middleware and route handlers created at module scope - for example a `rateLimit()` store, or a module-level counter - behave differently by environment:
+Middleware and route handlers created at module scope, for example a `rateLimit()` store or a module-level counter, behave differently by environment:
 
 - **In development**, the dev server re-evaluates each route module on every request so edits hot-reload. A fresh module scope means module-level variables and default in-memory stores are re-created per request: a counter always reads back its initial value, and the default in-memory rate-limit store never accumulates across requests. To exercise threshold behavior in dev, drive the pipeline multiple times within a single request, or use an external store.
 - **In production**, the compiled route module is cached per release, so module-scoped state persists across requests **within one server process and one release**. It is still **not** shared across multiple instances, and it resets on every redeploy (and under memory-pressure eviction).
 
-For anything that must be correct across requests, instances, and deploys - rate limiting, counters, sessions - use an external store (see the `RateLimitStore` interface and the Redis example in the rate-limit reference) rather than module-scoped memory.
+For anything that must be correct across requests, instances, and deploys, such as rate limiting, counters, and sessions, use an external store (see the `RateLimitStore` interface and the Redis example in the rate-limit reference) rather than module-scoped memory.
 
 ### Cleanup callbacks
 

scripts/lint/test-typecheck-baseline.json

@@ -33,8 +33,6 @@
   "src/build/production-build/manifest.test.ts",
   "src/build/renderer/services/css-bundler.test.ts",
   "src/cache/registry.test.ts",
-  "src/chat/conversation.test.ts",
-  "src/chat/message-prep.test.ts",
   "src/chat/upload-handler.test.ts",
   "src/config/env.test.ts",
   "src/discovery/agent-scoped-capabilities.test.ts",

src/agent/hosted/chat-request-parser.ts

@@ -1,9 +1,13 @@
 import type {
   ChatRequestContext,
   ChatRuntimeOverrides,
+  ChatToolPartState,
   ChatUiMessage,
+  ChatUiMessagePart,
+  ChatUiMessageRole,
   DurableRootRunDescriptor,
 } from "#veryfront/chat/types.ts";
+import { stringifyUnknown } from "#veryfront/chat/conversation.ts";
 import {
   buildHostedChatRequestInputFromRuntimeAgentInvocation,
   type HostedChatRequest,
@@ -20,6 +24,7 @@ import {
   type HostedRuntimeSourceIdentity,
   verifyHostedRuntimeSourceBinding,
 } from "./runtime-source-binding.ts";
+import { getHostedChatUiToolIdentity } from "./chat-request-tool-part.ts";
 
 /** Public API contract for hosted chat request principal. */
 export type HostedChatRequestPrincipal = {
@@ -123,6 +128,115 @@ function normalizeProjectSlug(projectSlug: string | undefined): string | undefin
   return normalized || undefined;
 }
 
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return value !== null && typeof value === "object" && !Array.isArray(value);
+}
+
+function mapRawToolCallState(state: string): ChatToolPartState {
+  switch (state) {
+    case "streaming":
+      return "input-streaming";
+    case "pending":
+      return "pending";
+    case "error":
+      return "completed";
+    default:
+      return "completed";
+  }
+}
+
+function normalizeHostedChatRequestMessages(
+  messages: HostedChatRequest["messages"],
+): ChatUiMessage[] {
+  const knownToolNames = new Map<string, string>();
+  const knownToolInputs = new Map<string, Record<string, unknown>>();
+
+  return messages.map((message) => {
+    const parts: ChatUiMessagePart[] = [];
+    const partIndexByToolCallId = new Map<string, number>();
+
+    for (const part of message.parts) {
+      const uiToolIdentity = getHostedChatUiToolIdentity(part);
+      if (uiToolIdentity) {
+        knownToolNames.set(uiToolIdentity.toolCallId, uiToolIdentity.toolName);
+        const input = isRecord(part as unknown)
+          ? (part as Record<string, unknown>)["input"]
+          : undefined;
+        if (isRecord(input)) {
+          knownToolInputs.set(uiToolIdentity.toolCallId, input);
+        }
+        partIndexByToolCallId.set(uiToolIdentity.toolCallId, parts.length);
+      }
+
+      if (isRecord(part) && part.type === "tool_call" && "id" in part && "name" in part) {
+        const rawPart: Record<string, unknown> = part;
+        const rawId = rawPart["id"];
+        const rawName = rawPart["name"];
+        const toolCallId = typeof rawId === "string" ? rawId : "";
+        const toolName = typeof rawName === "string" ? rawName : "";
+        const state = typeof part.state === "string" ? part.state : "completed";
+        if (!toolCallId || !toolName) continue;
+
+        const input = isRecord(part.input) ? part.input : {};
+        knownToolNames.set(toolCallId, toolName);
+        knownToolInputs.set(toolCallId, input);
+        partIndexByToolCallId.set(toolCallId, parts.length);
+        parts.push({
+          type: "tool_call",
+          toolCallId,
+          toolName,
+          input,
+          state: mapRawToolCallState(state),
+        });
+        continue;
+      }
+
+      if (isRecord(part) && part.type === "tool_result") {
+        const toolCallId = typeof part.tool_call_id === "string" ? part.tool_call_id : "";
+        const explicitToolName = typeof part.tool_name === "string" ? part.tool_name : "";
+        const toolName = explicitToolName || knownToolNames.get(toolCallId);
+        if (!toolCallId || !toolName) continue;
+
+        const resultPart: ChatUiMessagePart = {
+          type: "tool_call",
+          toolCallId,
+          toolName,
+          input: knownToolInputs.get(toolCallId) ?? {},
+          state: part.is_error === true ? "output-error" : "output-available",
+          output: part.output,
+          ...(part.is_error === true
+            ? { errorText: stringifyUnknown(part.output ?? "Tool error") }
+            : {}),
+        };
+
+        const existingIndex = partIndexByToolCallId.get(toolCallId);
+        if (existingIndex !== undefined) {
+          const existingPart = parts[existingIndex];
+          if (existingPart && isRecord(existingPart)) {
+            parts[existingIndex] = {
+              ...existingPart,
+              state: resultPart.state,
+              output: resultPart.output,
+              ...(resultPart.errorText ? { errorText: resultPart.errorText } : {}),
+            } as ChatUiMessagePart;
+          }
+        } else {
+          parts.push(resultPart);
+        }
+        continue;
+      }
+
+      parts.push(part as ChatUiMessagePart);
+    }
+
+    return {
+      ...message,
+      role: message.role as ChatUiMessageRole,
+      parts,
+    };
+  });
+}
+
 function isBlankProjectSlug(projectSlug: string | undefined): boolean {
   return projectSlug !== undefined && !normalizeProjectSlug(projectSlug);
 }
@@ -228,7 +342,7 @@ export async function buildParsedHostedChatRequest(input: {
     agentId: input.agentId,
     userId: input.userId,
     authToken: input.authToken,
-    messages: messages as ChatUiMessage[],
+    messages: normalizeHostedChatRequestMessages(messages),
     validatedContext,
     projectId,
     projectSlug: verifiedProjectSlug,