fix(responses): make reasoning replay restart-safe and observable (#950)

src/adapters/openai-chat.ts

@@ -8,7 +8,11 @@ import { isCyberPolicyCode } from "../lib/errors";
 import { redactSecretString } from "../lib/redact";
 import { contentPartsToText } from "./image";
 import { identifyRoutedModel } from "./identity";
-import { peekReasoningForCall } from "../responses/reasoning-replay-cache";
+import {
+  getReasoningReplayStats,
+  peekReasoningForCall,
+  recordBareToolCallSerialization,
+} from "../responses/reasoning-replay-cache";
 import { buildNonOpenAIToolCatalogNudgeForTools, shouldInjectNonOpenAIToolCatalogNudge } from "./tool-catalog-nudge";
 import { openRouterProviderPayload, resolveOpenRouterRouting } from "../providers/openrouter-routing";
 import {
@@ -193,6 +197,22 @@ function toolResultImageChatParts(content: string | OcxContentPart[]): unknown[]
   return parts;
 }
 
+// #950 diagnostics: a bare tool-call continuation for a preserveReasoningContentModels
+// provider is the exact 400 shape this fix eliminates. Count every occurrence
+// (privacy-safe) and surface a throttled counter line — never reasoning text.
+let lastBareToolCallWarnAt = 0;
+const BARE_TOOL_CALL_WARN_MIN_INTERVAL_MS = 60_000;
+function noteBareToolCallSerialization(modelId: string): void {
+  recordBareToolCallSerialization(modelId);
+  const at = Date.now();
+  if (at - lastBareToolCallWarnAt < BARE_TOOL_CALL_WARN_MIN_INTERVAL_MS) return;
+  lastBareToolCallWarnAt = at;
+  const stats = getReasoningReplayStats();
+  console.warn(
+    `[opencodex] reasoning replay miss: bare tool-call continuation for preserveReasoningContentModels model="${modelId}" (cache hits=${stats.hits}, misses=${stats.misses}); reasoning text is never logged`,
+  );
+}
+
 function messagesToChatFormat(parsed: OcxParsedRequest, provider: OcxProviderConfig): unknown[] {
   const out: unknown[] = [];
   const { context, options } = parsed;
@@ -346,6 +366,8 @@ function messagesToChatFormat(parsed: OcxParsedRequest, provider: OcxProviderCon
           // recorded under every call id — join unique texts only.
           if (cached.length > 0) {
             reasoningContent = [...new Set(cached)].join("\n");
+          } else {
+            noteBareToolCallSerialization(parsed.modelId);
           }
         }
         if (reasoningContent.length > 0 && modelInList(provider.preserveReasoningContentModels, parsed.modelId)) {
@@ -412,6 +434,13 @@ function messagesToChatFormat(parsed: OcxParsedRequest, provider: OcxProviderCon
             toolCallId && modelInList(provider.preserveReasoningContentModels, parsed.modelId)
               ? peekReasoningForCall(toolCallId, replayCacheScope)
               : undefined;
+          if (
+            !cachedReasoning
+            && toolCallId
+            && modelInList(provider.preserveReasoningContentModels, parsed.modelId)
+          ) {
+            noteBareToolCallSerialization(parsed.modelId);
+          }
           out.push({
             role: "assistant",
             content: emptyAssistantContent(provider),

src/bridge.ts

@@ -827,8 +827,10 @@ export function bridgeToResponsesSSE(
               if (currentReasoning) closeCurrentReasoning();
               if (currentRawReasoning) closeCurrentRawReasoning();
               flushHiddenRawReasoning();
-              // Reasoning consumed by a text turn, not a tool call: no cache target.
-              rawReasoningForNextToolCall = "";
+              // Reasoning consumed by a REAL text turn, not a tool call: no cache target.
+              // Empty text deltas must not wipe reasoning that precedes a tool call
+              // (chat-completions providers emit empty content deltas mid-tool-turn).
+              if (event.text.length > 0) rawReasoningForNextToolCall = "";
               if (currentToolCall) closeCurrentToolCall();
               // Only flush on an explicit phase change. A later delta that omits `phase` must
               // keep appending to the current message rather than wiping the earlier phase.
@@ -880,7 +882,7 @@ export function bridgeToResponsesSSE(
               if (currentMsg) closeCurrentMessage("commentary");
               if (currentRawReasoning) closeCurrentRawReasoning();
               flushHiddenRawReasoning();
-              rawReasoningForNextToolCall = "";
+              if (event.thinking.length > 0) rawReasoningForNextToolCall = "";
               if (currentToolCall) closeCurrentToolCall();
               if (!currentReasoning) {
                 const itemId = `rs_${uuid()}`;
@@ -1561,7 +1563,9 @@ function buildResponseJSONWithBudget(
         if (currentText && e.phase !== undefined && currentTextPhase !== e.phase) flushText("commentary");
         if (currentSummaryReasoning) flushSummaryReasoning();
         if (currentRawReasoning) flushRawReasoning();
-        rawReasoningForNextToolCall = "";
+        // Empty text deltas (batch chat responses always carry content, often "") must
+        // not wipe reasoning that precedes a tool call (#950 non-streaming path).
+        if (e.text.length > 0) rawReasoningForNextToolCall = "";
         if (currentToolCallId) flushToolCall();
         // Compaction turns keep the summary out of normal message output (replay dedup — see
         // bridgeToResponsesSSE); it ships only inside the synthetic compaction item below.
@@ -1580,7 +1584,7 @@ function buildResponseJSONWithBudget(
       case "thinking_delta":
         if (currentText) flushText("commentary");
         if (currentRawReasoning) flushRawReasoning();
-        rawReasoningForNextToolCall = "";
+        if (e.thinking.length > 0) rawReasoningForNextToolCall = "";
         if (currentToolCallId) flushToolCall();
         {
           ({ value: currentSummaryReasoning, bytes: currentSummaryReasoningBytes } = appendBatchString(

src/config.ts

@@ -5,6 +5,7 @@ import { homedir } from "node:os";
 import { dirname, join, resolve } from "node:path";
 import { Database } from "bun:sqlite";
 import * as z from "zod/v4";
+import { resolveOpenCodexConfigDir } from "./lib/config-dir";
 import {
   bumpConfigGenerationAtPath,
   bumpCurrentConfigGeneration,
@@ -549,7 +550,7 @@ let resolvedConfigDirCache: { raw: string | undefined; path: string } | null = n
 function resolveConfigDir(): string {
   const raw = process.env["OPENCODEX_HOME"]?.trim() || undefined;
   if (resolvedConfigDirCache && resolvedConfigDirCache.raw === raw) return resolvedConfigDirCache.path;
-  const path = raw ? resolve(expandUserPath(raw)) : join(homedir(), ".opencodex");
+  const path = resolveOpenCodexConfigDir();
   resolvedConfigDirCache = { raw, path };
   return path;
 }

src/lib/config-dir.ts

@@ -0,0 +1,18 @@
+/**
+ * Dependency-free config-directory resolution shared by config.ts and the
+ * reasoning replay spill (PR #1126): OPENCODEX_HOME (with `~` expansion) wins,
+ * otherwise <home>/.opencodex. Kept primitive on purpose so leaf modules can
+ * import it without pulling in the whole config surface.
+ */
+import { homedir } from "node:os";
+import { join, resolve } from "node:path";
+
+export function resolveOpenCodexConfigDir(
+  env: Record<string, string | undefined> = process.env,
+): string {
+  const raw = env["OPENCODEX_HOME"]?.trim() || undefined;
+  if (!raw) return join(homedir(), ".opencodex");
+  if (raw === "~") return homedir();
+  if (raw.startsWith("~/") || raw.startsWith("~\\")) return join(homedir(), raw.slice(2));
+  return resolve(raw);
+}