feat: improve codex app controls and recovery

scripts/mock-codex-app-server.js

@@ -25,6 +25,7 @@ if (args[0] !== 'app-server') {
 const threads = new Map();
 const pendingServerRequests = new Map();
 let nextServerRequestId = 1;
+let mcpReloadCount = 0;
 
 function send(message) {
   process.stdout.write(`${JSON.stringify(message)}\n`);
@@ -429,6 +430,20 @@ function startTurn(params) {
     },
   });
 
+  if (/runtime warning/i.test(text)) {
+    const message = 'Heads up: Long threads and multiple compactions can cause the model to be less accurate. Start a new thread when possible to keep threads small and targeted.';
+    for (let i = 0; i < 2; i += 1) {
+      send({
+        method: 'warning',
+        params: {
+          threadId: thread.id,
+          turnId,
+          message,
+        },
+      });
+    }
+  }
+
   if (/collaboration/i.test(text)) {
     completeTurn(thread, turnId, `collaboration mode: ${collaborationSummary(params)}`);
     return { turn: { id: turnId, status: 'running', items: [] } };
@@ -516,6 +531,11 @@ function handleRequest(message) {
       send({ id, result: { data: [{ mode: 'default' }, { mode: 'plan' }] } });
       return;
     }
+    if (method === 'config/mcpServer/reload') {
+      mcpReloadCount += 1;
+      send({ id, result: { reloaded: true, reloadCount: mcpReloadCount } });
+      return;
+    }
     if (method === 'thread/start') {
       const thread = ensureThread(null, params);
       send({ id, result: { thread: threadPayload(thread), model: params.model || 'gpt-5.5', cwd: thread.cwd, modelProvider: 'mock', approvalPolicy: params.approvalPolicy || 'never', approvalsReviewer: 'user', sandbox: params.sandbox || 'danger-full-access' } });

scripts/regression.js

@@ -168,6 +168,20 @@ async function fetchAuthedJson(port, token, pathname) {
   return payload;
 }
 
+async function postAuthedJson(port, token, pathname, body = {}) {
+  const response = await fetch(`http://127.0.0.1:${port}${pathname}`, {
+    method: 'POST',
+    headers: {
+      'Content-Type': 'application/json',
+      Authorization: `Bearer ${token}`,
+    },
+    body: JSON.stringify(body),
+  });
+  const payload = await response.json();
+  assert(response.ok && payload.ok, `POST failed for ${pathname}: ${payload.message || response.status}`);
+  return payload;
+}
+
 async function callInternalMcp(port, token, payload) {
   const response = await fetch(`http://127.0.0.1:${port}/api/internal/mcp`, {
     method: 'POST',
@@ -785,6 +799,22 @@ async function main() {
     assert(/"hasTopLevelEffort":false/.test(codexAppDefaultCollab.text || ''), 'Codex App collaboration turn should not duplicate effort at top level');
     await nextMessage(messages, ws, (msg) => msg.type === 'done' && msg.sessionId === codexAppSession.sessionId);
 
+    ws.send(JSON.stringify({ type: 'message', text: 'codexapp runtime warning prompt', sessionId: codexAppSession.sessionId, mode: 'yolo', agent: 'codexapp' }));
+    const codexAppRuntimeWarning = await nextMessage(messages, ws, (msg) => (
+      msg.type === 'system_message' &&
+      msg.sessionId === codexAppSession.sessionId &&
+      /Long threads and multiple compactions/.test(msg.message || '')
+    ));
+    assert(/Long threads and multiple compactions/.test(codexAppRuntimeWarning.message || ''), 'Codex App should surface the first runtime warning');
+    await nextMessage(messages, ws, (msg) => msg.type === 'done' && msg.sessionId === codexAppSession.sessionId);
+    await sleep(150);
+    const duplicateRuntimeWarnings = messages.filter((msg) => (
+      msg.type === 'system_message' &&
+      msg.sessionId === codexAppSession.sessionId &&
+      /Long threads and multiple compactions/.test(msg.message || '')
+    ));
+    assert(duplicateRuntimeWarnings.length === 0, 'Codex App should suppress duplicate runtime warning banners');
+
     ws.send(JSON.stringify({ type: 'message', text: 'codexapp empty reasoning prompt', sessionId: codexAppSession.sessionId, mode: 'yolo', agent: 'codexapp' }));
     await nextMessage(messages, ws, (msg) => msg.type === 'done' && msg.sessionId === codexAppSession.sessionId);
     const storedCodexAppAfterReasoning = JSON.parse(fs.readFileSync(path.join(sessionsDir, `${codexAppSession.sessionId}.json`), 'utf8'));
@@ -804,6 +834,10 @@ async function main() {
     assert(storedCodexApp.messages.some((message) => message.role === 'assistant' && /codexapp tool prompt/.test(String(message.content || ''))), 'Codex App assistant response should be persisted');
     assert((storedCodexApp.totalUsage?.inputTokens || 0) > 0, 'Codex App token usage should be persisted');
 
+    const reloadMcpResult = await postAuthedJson(port, token, `/api/sessions/${codexAppSession.sessionId}/reload-mcp`);
+    assert(reloadMcpResult.sessionId === codexAppSession.sessionId, 'Codex App MCP reload should return the target session id');
+    assert(reloadMcpResult.result?.reloaded === true, 'Codex App MCP reload should call app-server config/mcpServer/reload');
+
     ws.send(JSON.stringify({ type: 'message', text: 'codexapp dynamic prompt', sessionId: codexAppSession.sessionId, mode: 'yolo', agent: 'codexapp' }));
     const codexAppDynamicTool = await nextMessage(messages, ws, (msg) => msg.type === 'tool_end' && msg.sessionId === codexAppSession.sessionId && msg.toolUseId === 'mcp-ccweb-list');
     assert(codexAppDynamicTool.kind === 'mcp_tool_call', 'Codex App should surface ccweb MCP tool calls');
@@ -835,10 +869,17 @@ async function main() {
     assert(guidedRequest.questions?.[0]?.id === 'choice', 'Codex App should forward request_user_input questions');
     ws.send(JSON.stringify({
       type: 'codex_app_user_input_response',
+      action: 'submit',
       sessionId: codexAppSession.sessionId,
       requestId: guidedRequest.requestId,
       answers: { choice: { answers: ['A'] } },
     }));
+    const guidedSubmitted = await nextMessage(messages, ws, (msg) =>
+      msg.type === 'system_message' &&
+      msg.sessionId === codexAppSession.sessionId &&
+      /已提交.*引导输入/.test(msg.message || '')
+    );
+    assert(/已提交.*引导输入/.test(guidedSubmitted.message || ''), 'Codex App should show guided input submission hint');
     const guidedDelta = await nextMessage(messages, ws, (msg) => msg.type === 'text_delta' && msg.sessionId === codexAppSession.sessionId && /guided answer: A/.test(msg.text || ''));
     assert(/guided answer: A/.test(guidedDelta.text || ''), 'Codex App should continue after guided input response');
     await nextMessage(messages, ws, (msg) => msg.type === 'done' && msg.sessionId === codexAppSession.sessionId);