From 18618ad8ea4c55352c3516caeca2f856f66e0fc3 Mon Sep 17 00:00:00 2001
From: ekko <152005280+EKKOLearnAI@users.noreply.github.com>
Date: Thu, 2 Jul 2026 09:54:04 +0800
Subject: [PATCH] fix bridge terminal error detection (#1890)

---
 ...07-02-issue-1881-bridge-terminal-errors.md | 11 ++++++
 .../hermes/run-chat/handle-bridge-run.ts      | 24 +++++++++----
 .../run-chat-bridge-terminal-error.test.ts    | 34 +++++++++++++++++++
 3 files changed, 62 insertions(+), 7 deletions(-)
 create mode 100644 docs/chat-chain-changes/2026-07-02-issue-1881-bridge-terminal-errors.md

diff --git a/docs/chat-chain-changes/2026-07-02-issue-1881-bridge-terminal-errors.md b/docs/chat-chain-changes/2026-07-02-issue-1881-bridge-terminal-errors.md
new file mode 100644
index 00000000..030ac53e
--- /dev/null
+++ b/docs/chat-chain-changes/2026-07-02-issue-1881-bridge-terminal-errors.md
@@ -0,0 +1,11 @@
+---
+date: 2026-07-02
+pr: 1890
+feature: Bridge terminal error detection
+impact: Completed bridge runs no longer fail just because successful assistant text mentions HTTP status handling or a generic result message.
+---
+
+Issue #1881 showed successful assistant output being rendered as a failed run
+when the response text mentioned HTTP status handling such as 401 redirects.
+Terminal error detection now keeps explicit bridge or agent failure flags as
+authoritative and only uses text matching for compact provider/API error shapes.
diff --git a/packages/server/src/services/hermes/run-chat/handle-bridge-run.ts b/packages/server/src/services/hermes/run-chat/handle-bridge-run.ts
index 8c0585e2..db9914a3 100644
--- a/packages/server/src/services/hermes/run-chat/handle-bridge-run.ts
+++ b/packages/server/src/services/hermes/run-chat/handle-bridge-run.ts
@@ -117,9 +117,18 @@ function shouldPollBridgeGeneratedTitle(sessionId: string): boolean {
 }
 
 function looksLikeAgentFailure(value: string): boolean {
-  return /\bAPI call failed after\b/i.test(value)
-    || /\bHTTP\s+(?:4\d\d|5\d\d)\b/i.test(value)
-    || /\b(?:401|403|429|500|502|503|504)\b/.test(value) && /\b(?:unauthorized|forbidden|rate limit|unavailable|failed|error)\b/i.test(value)
+  const text = value.replace(/\s+/g, ' ').trim()
+  if (!text) return false
+
+  return /\bAPI call failed after\b/i.test(text)
+    || /\bHTTP\s+(?:4\d\d|5\d\d)\b/i.test(text)
+    || /\b(?:401|403)\b.{0,100}\b(?:unauthorized|forbidden|authentication|auth|invalid api key|permission denied)\b/i.test(text)
+    || /\b(?:unauthorized|forbidden|authentication|auth|invalid api key|permission denied)\b.{0,100}\b(?:401|403)\b/i.test(text)
+    || /\b429\b.{0,100}\b(?:rate limit|too many requests|quota)\b/i.test(text)
+    || /\b(?:rate limit|too many requests|quota)\b.{0,100}\b429\b/i.test(text)
+    || /\b(?:500|502|503|504)\b.{0,100}\b(?:server error|bad gateway|service unavailable|gateway timeout|upstream|provider|request failed|api)\b/i.test(text)
+    || /\b(?:server error|bad gateway|service unavailable|gateway timeout|upstream|provider|request failed|api)\b.{0,100}\b(?:500|502|503|504)\b/i.test(text)
+    || /(?:无可用渠道|渠道不可用|认证失败|鉴权失败|额度不足|余额不足|请求失败|接口调用失败|限流)/i.test(text)
 }
 
 export function bridgeTerminalError(chunk: Pick<AgentBridgeOutput, 'status' | 'error' | 'result'>): string | null {
@@ -129,19 +138,20 @@ export function bridgeTerminalError(chunk: Pick<AgentBridgeOutput, 'status' | 'e
   const resultError = result
     ? stringValue(result.error)
       || stringValue(result.exception)
-      || stringValue(result.message)
     : ''
+  const resultMessage = result ? stringValue(result.message) : ''
   const finalResponse = result ? stringValue(result.final_response) : ''
 
   if (chunk.status === 'error') {
-    return stringValue(chunk.error) || resultError || finalResponse || 'Agent run failed'
+    return stringValue(chunk.error) || resultError || resultMessage || finalResponse || 'Agent run failed'
   }
 
   if (result?.failed === true || result?.completed === false) {
-    return resultError || finalResponse || 'Agent reported failure'
+    return resultError || resultMessage || finalResponse || 'Agent reported failure'
   }
 
-  if (resultError) return resultError
+  if (resultError && looksLikeAgentFailure(resultError)) return resultError
+  if (!finalResponse && resultMessage && looksLikeAgentFailure(resultMessage)) return resultMessage
   if (finalResponse && looksLikeAgentFailure(finalResponse)) return finalResponse
 
   return null
diff --git a/tests/server/run-chat-bridge-terminal-error.test.ts b/tests/server/run-chat-bridge-terminal-error.test.ts
index 465c74c4..48e464b8 100644
--- a/tests/server/run-chat-bridge-terminal-error.test.ts
+++ b/tests/server/run-chat-bridge-terminal-error.test.ts
@@ -68,4 +68,38 @@ describe('bridge terminal error detection', () => {
       },
     } as any)).toBeNull()
   })
+
+  it('does not flag implementation notes that mention HTTP status handling', () => {
+    expect(bridgeTerminalError({
+      status: 'complete',
+      result: {
+        completed: true,
+        final_response: [
+          '全部完成，构建通过。',
+          'src/api/request.js: Axios 实例 + 请求拦截器 + 响应拦截器(401 互斥跳转)。',
+          '登录按钮状态: normal / hover / active / loading / success / error / locked / disabled。',
+        ].join('\n\n'),
+      },
+    } as any)).toBeNull()
+  })
+
+  it('does not treat a successful result message as an error', () => {
+    expect(bridgeTerminalError({
+      status: 'complete',
+      result: {
+        completed: true,
+        message: '全部完成，构建通过。',
+        final_response: '全部完成，构建通过。',
+      },
+    } as any)).toBeNull()
+  })
+
+  it('still surfaces compact auth errors in result messages when no final response exists', () => {
+    expect(bridgeTerminalError({
+      status: 'complete',
+      result: {
+        message: 'HTTP 401: unauthorized',
+      },
+    } as any)).toBe('HTTP 401: unauthorized')
+  })
 })