fix: isolate visual tool failures from agent runs

agent-run-gateway.test.mjs

@@ -1246,6 +1246,103 @@ test('agent run replaces reasoning-poisoned Goose session and retries with visib
   assert.equal(replacedData?.reason, 'SESSION_REASONING_CONTENT_POISONED');
 });
 
+test('agent run degrades visual inspection after an earlier session-history recovery', async () => {
+  const pool = createFakePool();
+  const submitted = [];
+  const started = [];
+  const replacementIds = ['session-tool-clean', 'session-visual-fallback'];
+  const priorConversation = [
+    {
+      role: 'user',
+      content: [{ type: 'text', text: '帮我写一首诗，做成页面' }],
+    },
+    {
+      role: 'assistant',
+      content: [{ type: 'text', text: '页面已经生成。' }],
+    },
+  ];
+  const gateway = createAgentRunGateway({
+    pool,
+    userAuth: {},
+    tkmindProxy: {
+      async startSessionForUser(_userId, options) {
+        started.push(options ?? null);
+        return { id: replacementIds.shift() };
+      },
+      async fetchSessionConversationForUser() {
+        return priorConversation;
+      },
+      async submitSessionReplyAndAwaitFinishForUser(
+        _userId,
+        sessionId,
+        _requestId,
+        userMessage,
+        options,
+      ) {
+        submitted.push({ sessionId, userMessage, options });
+        if (sessionId === 'session-poisoned') {
+          const error = new Error('tool history requires fresh session');
+          error.code = 'SESSION_TOOL_HISTORY_FRESH_SESSION_REQUIRED';
+          error.repairedConversation = priorConversation;
+          throw error;
+        }
+        if (sessionId === 'session-tool-clean') {
+          const error = new Error(
+            'messages[8]: unknown variant `image_url`, expected `text`',
+          );
+          error.code = 'SESSION_VISUAL_CONTEXT_UNSUPPORTED';
+          error.retryable = false;
+          throw error;
+        }
+        return {
+          ok: true,
+          finishEvent: { type: 'Finish' },
+          toolEvidence: { calls: ['sandbox-fs__edit_file'] },
+        };
+      },
+    },
+    retryDelaysMs: [],
+  });
+
+  const run = await gateway.createRun('user-1', {
+    sessionId: 'session-poisoned',
+    requestId: 'req-visual-fallback',
+    userMessage: {
+      role: 'user',
+      content: [{ type: 'text', text: '页面再精美一点' }],
+    },
+  });
+
+  await waitFor(() => pool.runs.get(run.id)?.status === 'succeeded');
+  assert.deepEqual(submitted.map((item) => item.sessionId), [
+    'session-poisoned',
+    'session-tool-clean',
+    'session-visual-fallback',
+  ]);
+  assert.deepEqual(started, [
+    null,
+    { disableImageReading: true },
+  ]);
+  assert.equal(
+    submitted.at(-1).options.disableImageReading,
+    true,
+  );
+  assert.match(
+    submitted.at(-1).userMessage.content[0].text,
+    /视觉检查已降级/,
+  );
+  assert.match(
+    submitted.at(-1).userMessage.content[0].text,
+    /不要再次调用 read_image/,
+  );
+  assert.equal(
+    pool.events.filter(
+      (event) => event.eventType === 'poisoned_session_replaced',
+    ).length,
+    2,
+  );
+});
+
 test('Page Data run fails closed when Finish arrives without a generated page', async () => {
   const pool = createFakePool();
   const repairSubmits = [];
@@ -2756,6 +2853,50 @@ test('agent run retries transient failures and then becomes terminal', async ()
   assert.match(pool.runs.get(run.id).error_message, /upstream unavailable/);
 });
 
+test('agent run cancels an active upstream request before retrying', async () => {
+  const pool = createFakePool();
+  const calls = [];
+  let submissions = 0;
+  const gateway = createAgentRunGateway({
+    pool,
+    userAuth: {},
+    tkmindProxy: {
+      async startSessionForUser() {
+        return { id: 'session-retry-cancel' };
+      },
+      async submitSessionReplyForUser() {
+        submissions += 1;
+        calls.push(`submit-${submissions}`);
+        if (submissions === 1) throw new Error('upstream unavailable');
+      },
+    },
+    async cancelSessionOnRetry(input) {
+      calls.push('cancel');
+      assert.equal(input.sessionId, 'session-retry-cancel');
+      assert.equal(input.requestId, 'req-retry-cancel');
+      return { cancelled: true, skipped: false };
+    },
+    retryDelaysMs: [0, 0],
+  });
+
+  const run = await gateway.createRun('user-1', {
+    requestId: 'req-retry-cancel',
+    userMessage: {
+      role: 'user',
+      content: [{ type: 'text', text: '继续任务' }],
+    },
+  });
+
+  await waitFor(() => pool.runs.get(run.id)?.status === 'succeeded');
+  assert.deepEqual(calls, ['submit-1', 'cancel', 'submit-2']);
+  assert.equal(
+    pool.events.some(
+      (event) => event.eventType === 'session_retry_cancelled',
+    ),
+    true,
+  );
+});
+
 test('agent run queue limits concurrent execution', async () => {
   const pool = createFakePool();
   let active = 0;