Merge pull request #149 from CJackHwang/codex/fix-tool-miscall-during-complex-json-test Ignore tool_call payloads inside fenced code blocks and chat envelopes; stream-aware code-fence tracking

VERSION

@@ -1 +1 @@
-2.3.8
+2.4.0

internal/adapter/claude/handler_stream_test.go

@@ -358,7 +358,7 @@ func TestHandleClaudeStreamRealtimeToolSafetyAcrossStructuredFormats(t *testing.
 	}
 }
 
-func TestHandleClaudeStreamRealtimePromotesUnclosedFencedToolExample(t *testing.T) {
+func TestHandleClaudeStreamRealtimeIgnoresUnclosedFencedToolExample(t *testing.T) {
 	h := &Handler{}
 	resp := makeClaudeSSEHTTPResponse(
 		"data: {\"p\":\"response/content\",\"v\":\"Here is an example:\\n```json\\n{\\\"tool_calls\\\":[{\\\"name\\\":\\\"Bash\\\",\\\"input\\\":{\\\"command\\\":\\\"pwd\\\"}}]}\"}",
@@ -379,8 +379,8 @@ func TestHandleClaudeStreamRealtimePromotesUnclosedFencedToolExample(t *testing.
 			break
 		}
 	}
-	if !foundToolUse {
-		t.Fatalf("expected tool_use for fenced example, body=%s", rec.Body.String())
+	if foundToolUse {
+		t.Fatalf("expected no tool_use for fenced example, body=%s", rec.Body.String())
 	}
 
 	foundToolStop := false
@@ -391,7 +391,12 @@ func TestHandleClaudeStreamRealtimePromotesUnclosedFencedToolExample(t *testing.
 			break
 		}
 	}
-	if !foundToolStop {
-		t.Fatalf("expected stop_reason=tool_use, body=%s", rec.Body.String())
+	if foundToolStop {
+		t.Fatalf("expected stop_reason to remain content-only, body=%s", rec.Body.String())
 	}
 }
+
+// Backward-compatible alias for historical test name used in CI logs.
+func TestHandleClaudeStreamRealtimePromotesUnclosedFencedToolExample(t *testing.T) {
+	TestHandleClaudeStreamRealtimeIgnoresUnclosedFencedToolExample(t)
+}

internal/adapter/openai/handler_toolcall_test.go

@@ -243,7 +243,7 @@ func TestHandleNonStreamEmbeddedToolCallExamplePromotesToolCall(t *testing.T) {
 	}
 }
 
-func TestHandleNonStreamFencedToolCallExamplePromotesToolCall(t *testing.T) {
+func TestHandleNonStreamFencedToolCallExampleDoesNotPromoteToolCall(t *testing.T) {
 	h := &Handler{}
 	resp := makeSSEHTTPResponse(
 		"data: {\"p\":\"response/content\",\"v\":\"```json\\n{\\\"tool_calls\\\":[{\\\"name\\\":\\\"search\\\",\\\"input\\\":{\\\"q\\\":\\\"go\\\"}}]}\\n```\"}",
@@ -259,20 +259,25 @@ func TestHandleNonStreamFencedToolCallExamplePromotesToolCall(t *testing.T) {
 	out := decodeJSONBody(t, rec.Body.String())
 	choices, _ := out["choices"].([]any)
 	choice, _ := choices[0].(map[string]any)
-	if choice["finish_reason"] != "tool_calls" {
-		t.Fatalf("expected finish_reason=tool_calls, got %#v", choice["finish_reason"])
+	if choice["finish_reason"] == "tool_calls" {
+		t.Fatalf("expected fenced example to remain content-only, got finish_reason=%#v", choice["finish_reason"])
 	}
 	msg, _ := choice["message"].(map[string]any)
 	toolCalls, _ := msg["tool_calls"].([]any)
-	if len(toolCalls) != 1 {
-		t.Fatalf("expected one tool_call field for fenced example: %#v", msg["tool_calls"])
+	if len(toolCalls) != 0 {
+		t.Fatalf("expected no tool_call field for fenced example: %#v", msg["tool_calls"])
 	}
 	content, _ := msg["content"].(string)
-	if strings.Contains(content, `"tool_calls"`) {
-		t.Fatalf("expected raw tool_calls json stripped from content, got %q", content)
+	if !strings.Contains(content, `"tool_calls"`) {
+		t.Fatalf("expected fenced example content preserved, got %q", content)
 	}
 }
 
+// Backward-compatible alias for historical test name used in CI logs.
+func TestHandleNonStreamFencedToolCallExamplePromotesToolCall(t *testing.T) {
+	TestHandleNonStreamFencedToolCallExampleDoesNotPromoteToolCall(t)
+}
+
 func TestHandleStreamToolCallInterceptsWithoutRawContentLeak(t *testing.T) {
 	h := &Handler{}
 	resp := makeSSEHTTPResponse(

internal/format/openai/render_test.go

@@ -2,6 +2,7 @@ package openai
 
 import (
 	"encoding/json"
+	"strings"
 	"testing"
 )
 
@@ -69,7 +70,7 @@ func TestBuildResponseObjectPromotesMixedProseToolPayloadToFunctionCall(t *testi
 	}
 }
 
-func TestBuildResponseObjectPromotesFencedToolPayloadToFunctionCall(t *testing.T) {
+func TestBuildResponseObjectKeepsFencedToolPayloadAsText(t *testing.T) {
 	obj := BuildResponseObject(
 		"resp_test",
 		"gpt-4o",
@@ -80,19 +81,24 @@ func TestBuildResponseObjectPromotesFencedToolPayloadToFunctionCall(t *testing.T
 	)
 
 	outputText, _ := obj["output_text"].(string)
-	if outputText != "" {
-		t.Fatalf("expected output_text hidden for fenced tool payload, got %q", outputText)
+	if !strings.Contains(outputText, "\"tool_calls\"") {
+		t.Fatalf("expected output_text to preserve fenced tool payload, got %q", outputText)
 	}
 	output, _ := obj["output"].([]any)
 	if len(output) != 1 {
-		t.Fatalf("expected one function_call output item, got %#v", obj["output"])
+		t.Fatalf("expected one message output item, got %#v", obj["output"])
 	}
 	first, _ := output[0].(map[string]any)
-	if first["type"] != "function_call" {
-		t.Fatalf("expected function_call output type, got %#v", first["type"])
+	if first["type"] != "message" {
+		t.Fatalf("expected message output type, got %#v", first["type"])
 	}
 }
 
+// Backward-compatible alias for historical test name used in CI logs.
+func TestBuildResponseObjectPromotesFencedToolPayloadToFunctionCall(t *testing.T) {
+	TestBuildResponseObjectKeepsFencedToolPayloadAsText(t)
+}
+
 func TestBuildResponseObjectReasoningOnlyFallsBackToOutputText(t *testing.T) {
 	obj := BuildResponseObject(
 		"resp_test",

internal/js/helpers/stream-tool-sieve/parse.js

@@ -8,9 +8,12 @@ const {
   parseToolCallsPayload,
   parseMarkupToolCalls,
   parseTextKVToolCalls,
+  stripFencedCodeBlocks,
 } = require('./parse_payload');
+const { TOOL_SEGMENT_KEYWORDS } = require('./tool-keywords');
 
 const TOOL_NAME_LOOSE_PATTERN = /[^a-z0-9]+/g;
+const TOOL_MARKUP_PREFIXES = ['<tool_call', '<function_call', '<invoke'];
 
 function extractToolNames(tools) {
   if (!Array.isArray(tools) || tools.length === 0) {
@@ -44,6 +47,9 @@ function parseToolCallsDetailed(text, toolNames) {
     return result;
   }
   result.sawToolCallSyntax = looksLikeToolCallSyntax(normalized);
+  if (shouldSkipToolCallParsingForCodeFenceExample(normalized)) {
+    return result;
+  }
 
   const candidates = buildToolCallCandidates(normalized);
   let parsed = [];
@@ -89,6 +95,9 @@ function parseStandaloneToolCallsDetailed(text, toolNames) {
     return result;
   }
   result.sawToolCallSyntax = looksLikeToolCallSyntax(trimmed);
+  if (shouldSkipToolCallParsingForCodeFenceExample(trimmed)) {
+    return result;
+  }
   const candidates = buildToolCallCandidates(trimmed);
   let parsed = [];
   for (const c of candidates) {
@@ -223,11 +232,16 @@ function resolveAllowedToolName(name, allowed, allowedCanonical) {
 
 function looksLikeToolCallSyntax(text) {
   const lower = toStringSafe(text).toLowerCase();
-  return lower.includes('tool_calls')
-    || lower.includes('<tool_call')
-    || lower.includes('<function_call')
-    || lower.includes('<invoke')
-    || lower.includes('function.name:');
+  return TOOL_SEGMENT_KEYWORDS.some((kw) => lower.includes(kw))
+    || TOOL_MARKUP_PREFIXES.some((prefix) => lower.includes(prefix));
+}
+
+function shouldSkipToolCallParsingForCodeFenceExample(text) {
+  if (!looksLikeToolCallSyntax(text)) {
+    return false;
+  }
+  const stripped = stripFencedCodeBlocks(text);
+  return !looksLikeToolCallSyntax(stripped);
 }
 
 module.exports = {

internal/js/helpers/stream-tool-sieve/parse_payload.js

@@ -114,13 +114,31 @@ function parseToolCallsPayload(payload) {
     return [];
   }
   if (decoded.tool_calls) {
+    if (isLikelyChatMessageEnvelope(decoded)) {
+      return [];
+    }
     return parseToolCallList(decoded.tool_calls);
   }
 
   const one = parseToolCallItem(decoded);
   return one ? [one] : [];
 }
 
+function isLikelyChatMessageEnvelope(value) {
+  if (!value || typeof value !== 'object' || Array.isArray(value)) {
+    return false;
+  }
+  if (!Object.prototype.hasOwnProperty.call(value, 'tool_calls')) {
+    return false;
+  }
+  const role = toStringSafe(value.role).trim().toLowerCase();
+  if (role === 'assistant' || role === 'tool' || role === 'user' || role === 'system') {
+    return true;
+  }
+  return Object.prototype.hasOwnProperty.call(value, 'tool_call_id')
+    || Object.prototype.hasOwnProperty.call(value, 'content');
+}
+
 function parseMarkupToolCalls(text) {
   const raw = toStringSafe(text).trim();
   if (!raw) {

internal/js/helpers/stream-tool-sieve/sieve.js

@@ -1,8 +1,12 @@
 'use strict';
-const { resetIncrementalToolState, noteText, insideCodeFence } = require('./state');
+const {
+  resetIncrementalToolState,
+  noteText,
+  insideCodeFenceWithState,
+} = require('./state');
 const { parseStandaloneToolCallsDetailed } = require('./parse');
 const { extractJSONObjectFrom } = require('./jsonscan');
-
+const { TOOL_SEGMENT_KEYWORDS, earliestKeywordIndex } = require('./tool-keywords');
 function processToolSieveChunk(state, chunk, toolNames) {
   if (!state) {
     return [];
@@ -53,7 +57,7 @@ function processToolSieveChunk(state, chunk, toolNames) {
     if (!pending) {
       break;
     }
-    const start = findToolSegmentStart(pending);
+    const start = findToolSegmentStart(state, pending);
     if (start >= 0) {
       const prefix = pending.slice(0, start);
       if (prefix) {
@@ -143,32 +147,21 @@ function findSuspiciousPrefixStart(s) {
   return start;
 }
 
-function findToolSegmentStart(s) {
+function findToolSegmentStart(state, s) {
   if (!s) {
     return -1;
   }
   const lower = s.toLowerCase();
-  const keywords = ['tool_calls', 'function.name:', '[tool_call_history]', '[tool_result_history]'];
   let offset = 0;
   while (true) {
-    let bestKeyIdx = -1;
-    let matchedKeyword = '';
-    for (const kw of keywords) {
-      const idx = lower.indexOf(kw, offset);
-      if (idx >= 0) {
-        if (bestKeyIdx < 0 || idx < bestKeyIdx) {
-          bestKeyIdx = idx;
-          matchedKeyword = kw;
-        }
-      }
-    }
+    const { index: bestKeyIdx, keyword: matchedKeyword } = earliestKeywordIndex(lower, TOOL_SEGMENT_KEYWORDS, offset);
     if (bestKeyIdx < 0) {
       return -1;
     }
     const keyIdx = bestKeyIdx;
     const start = s.slice(0, keyIdx).lastIndexOf('{');
     const candidateStart = start >= 0 ? start : keyIdx;
-    if (!insideCodeFence(s.slice(0, candidateStart))) {
+    if (!insideCodeFenceWithState(state, s.slice(0, candidateStart))) {
       return candidateStart;
     }
     offset = keyIdx + matchedKeyword.length;
@@ -181,14 +174,7 @@ function consumeToolCapture(state, toolNames) {
     return { ready: false, prefix: '', calls: [], suffix: '' };
   }
   const lower = captured.toLowerCase();
-  let keyIdx = -1;
-  const keywords = ['tool_calls', 'function.name:', '[tool_call_history]', '[tool_result_history]'];
-  for (const kw of keywords) {
-    const idx = lower.indexOf(kw);
-    if (idx >= 0 && (keyIdx < 0 || idx < keyIdx)) {
-      keyIdx = idx;
-    }
-  }
+  const { index: keyIdx } = earliestKeywordIndex(lower, TOOL_SEGMENT_KEYWORDS);
   if (keyIdx < 0) {
     return { ready: false, prefix: '', calls: [], suffix: '' };
   }
@@ -211,7 +197,7 @@ function consumeToolCapture(state, toolNames) {
   }
   const prefixPart = captured.slice(0, actualStart);
   const suffixPart = captured.slice(obj.end);
-  if (insideCodeFence((state.recentTextTail || '') + prefixPart)) {
+  if (insideCodeFenceWithState(state, prefixPart)) {
     return {
       ready: true,
       prefix: captured,
@@ -281,7 +267,6 @@ function trimWrappingJSONFence(prefix, suffix) {
   if (header && header !== 'json') {
     return { prefix, suffix };
   }
-
   const leftTrimmedSuffix = (suffix || '').replace(/^[ \t\r\n]+/g, '');
   if (!leftTrimmedSuffix.startsWith('```')) {
     return { prefix, suffix };
@@ -292,7 +277,6 @@ function trimWrappingJSONFence(prefix, suffix) {
     suffix: (suffix || '').slice(consumed + 3),
   };
 }
-
 module.exports = {
   processToolSieveChunk,
   flushToolSieve,