@cloudcli-ai/cloudcli 1.37.2 → 1.37.4

package/server/modules/providers/tests/claude-message-model.test.ts

@@ -0,0 +1,160 @@
+import assert from 'node:assert/strict';
+import { mkdtemp, rm, writeFile } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+import test from 'node:test';
+
+import { closeConnection, initializeDatabase, sessionsDb } from '@/modules/database/index.js';
+import { ClaudeSessionsProvider } from '@/modules/providers/list/claude/claude-sessions.provider.js';
+
+const SESSION_ID = 'claude-model-session';
+
+async function withIsolatedDatabase(runTest: () => void | Promise<void>): Promise<void> {
+  const previousDatabasePath = process.env.DATABASE_PATH;
+  const tempDirectory = await mkdtemp(path.join(os.tmpdir(), 'claude-model-db-'));
+
+  closeConnection();
+  process.env.DATABASE_PATH = path.join(tempDirectory, 'auth.db');
+  await initializeDatabase();
+
+  try {
+    await runTest();
+  } finally {
+    closeConnection();
+    if (previousDatabasePath === undefined) {
+      delete process.env.DATABASE_PATH;
+    } else {
+      process.env.DATABASE_PATH = previousDatabasePath;
+    }
+    await rm(tempDirectory, { recursive: true, force: true });
+  }
+}
+
+/**
+ * One assistant transcript row in the shape Claude Code writes it: the model
+ * that answered sits on `message.model`, beside the content parts.
+ */
+const assistantRow = (uuid: string, model: string | undefined, content: unknown) => ({
+  parentUuid: 'user-1',
+  isSidechain: false,
+  type: 'assistant',
+  uuid,
+  timestamp: '2026-07-10T00:00:01.000Z',
+  cwd: '/workspace/demo',
+  sessionId: SESSION_ID,
+  message: { role: 'assistant', ...(model === undefined ? {} : { model }), content },
+});
+
+test('an assistant reply carries the model that answered it', () => {
+  const messages = new ClaudeSessionsProvider().normalizeMessage(
+    assistantRow('asst-1', 'claude-opus-5', [
+      { type: 'thinking', thinking: 'weighing the options' },
+      { type: 'text', text: 'Here is the plan.' },
+      { type: 'tool_use', id: 'toolu_1', name: 'Read', input: { file_path: '/repo/a.ts' } },
+    ]),
+    SESSION_ID,
+  );
+
+  const prose = messages.find((message) => message.kind === 'text');
+  assert.equal(prose?.model, 'claude-opus-5');
+
+  // Only prose has a footer to show it in; a tool call and a reasoning block
+  // are drawn as cards, so stamping them would carry a model nothing renders.
+  assert.equal(messages.find((message) => message.kind === 'tool_use')?.model, undefined);
+  assert.equal(messages.find((message) => message.kind === 'thinking')?.model, undefined);
+});
+
+test('an assistant reply whose content is a bare string carries the model too', () => {
+  const [prose] = new ClaudeSessionsProvider().normalizeMessage(
+    assistantRow('asst-2', 'claude-haiku-4-5-20251001', 'Done.'),
+    SESSION_ID,
+  );
+
+  assert.equal(prose?.kind, 'text');
+  assert.equal(prose?.model, 'claude-haiku-4-5-20251001');
+});
+
+test('the reported model id is passed through verbatim, suffix and all', () => {
+  // Claude Code writes the base id on assistant rows today, even in 1M-context
+  // sessions. The normalizer must still never rewrite what the provider wrote:
+  // if a `[1m]` id ever appears, it has to reach the client as itself.
+  const [prose] = new ClaudeSessionsProvider().normalizeMessage(
+    assistantRow('asst-3', 'claude-opus-5[1m]', [{ type: 'text', text: 'Reading the whole repo.' }]),
+    SESSION_ID,
+  );
+
+  assert.equal(prose?.model, 'claude-opus-5[1m]');
+});
+
+test('a locally fabricated row reports no model at all', () => {
+  // Claude Code stamps rows it synthesized itself — the usage-limit notice, an
+  // API-error placeholder — with `<synthetic>`. No request ran, so there is no
+  // model to name, and a guessed one would be a lie about what was charged.
+  const [prose] = new ClaudeSessionsProvider().normalizeMessage(
+    assistantRow('asst-4', '<synthetic>', [{ type: 'text', text: 'Claude usage limit reached.' }]),
+    SESSION_ID,
+  );
+
+  assert.equal(prose?.kind, 'text');
+  assert.equal(prose?.model, undefined);
+});
+
+test('a user turn carries no model, because no transcript records one', () => {
+  const [turn] = new ClaudeSessionsProvider().normalizeMessage(
+    {
+      type: 'user',
+      uuid: 'user-1',
+      timestamp: '2026-07-10T00:00:00.000Z',
+      sessionId: SESSION_ID,
+      message: { role: 'user', content: 'Plan the refactor.' },
+    },
+    SESSION_ID,
+  );
+
+  assert.equal(turn?.role, 'user');
+  assert.equal(turn?.model, undefined);
+});
+
+test('Claude history replays the model of every assistant turn from the transcript', { concurrency: false }, async () => {
+  const tempRoot = await mkdtemp(path.join(os.tmpdir(), 'claude-model-history-'));
+
+  try {
+    const transcriptPath = path.join(tempRoot, `${SESSION_ID}.jsonl`);
+    await writeFile(
+      transcriptPath,
+      [
+        JSON.stringify({
+          parentUuid: null,
+          isSidechain: false,
+          type: 'user',
+          uuid: 'user-1',
+          timestamp: '2026-07-10T00:00:00.000Z',
+          cwd: '/workspace/demo',
+          sessionId: SESSION_ID,
+          message: { role: 'user', content: 'Plan the refactor.' },
+        }),
+        JSON.stringify(assistantRow('asst-1', 'claude-opus-5', [{ type: 'text', text: 'Here is the plan.' }])),
+        JSON.stringify(assistantRow('asst-2', '<synthetic>', [{ type: 'text', text: 'Claude usage limit reached.' }])),
+        '',
+      ].join('\n'),
+      'utf8',
+    );
+
+    await withIsolatedDatabase(async () => {
+      const now = new Date().toISOString();
+      sessionsDb.createSession(SESSION_ID, 'claude', tempRoot, 'Model session', now, now, transcriptPath);
+
+      const history = await new ClaudeSessionsProvider().fetchHistory(SESSION_ID, {
+        providerSessionId: SESSION_ID,
+      });
+
+      const replies = history.messages.filter((message) => message.kind === 'text' && message.role === 'assistant');
+      assert.equal(replies.length, 2);
+      assert.equal(replies[0]?.model, 'claude-opus-5');
+      assert.equal(replies[1]?.model, undefined);
+      assert.equal(history.messages.find((message) => message.role === 'user')?.model, undefined);
+    });
+  } finally {
+    await rm(tempRoot, { recursive: true, force: true });
+  }
+});

package/server/modules/providers/tests/claude-models.test.ts

@@ -0,0 +1,87 @@
+import assert from 'node:assert/strict';
+import test from 'node:test';
+
+import { extractClaudeEventModel } from '@/modules/providers/list/claude/claude-models.provider.js';
+
+const SESSION_ID = 'session-1';
+
+test('ignores the <synthetic> placeholder Claude Code stamps on synthesized rows', () => {
+  assert.equal(
+    extractClaudeEventModel(
+      { sessionId: SESSION_ID, message: { model: '<synthetic>' } },
+      SESSION_ID,
+    ),
+    null,
+  );
+  assert.equal(
+    extractClaudeEventModel({ sessionId: SESSION_ID, model: '<synthetic>' }, SESSION_ID),
+    null,
+  );
+});
+
+test('still surfaces real model ids from message and event fields', () => {
+  assert.equal(
+    extractClaudeEventModel(
+      { sessionId: SESSION_ID, message: { model: 'claude-sonnet-5' } },
+      SESSION_ID,
+    ),
+    'claude-sonnet-5',
+  );
+  assert.equal(
+    extractClaudeEventModel({ sessionId: SESSION_ID, model: 'opus' }, SESSION_ID),
+    'opus',
+  );
+});
+
+test('skips a placeholder content part so a later real model tag still wins', () => {
+  assert.equal(
+    extractClaudeEventModel(
+      {
+        sessionId: SESSION_ID,
+        message: {
+          content: [
+            { text: '<model><synthetic></model>' },
+            { text: '<model>claude-sonnet-5</model>' },
+          ],
+        },
+      },
+      SESSION_ID,
+    ),
+    'claude-sonnet-5',
+  );
+});
+
+test('a placeholder stdout hit does not shadow a real <model> tag in the same text', () => {
+  const text = '<local-command-stdout>Set model to <synthetic></local-command-stdout>'
+    + '<model>claude-sonnet-5</model>';
+  assert.equal(
+    extractClaudeEventModel(
+      { sessionId: SESSION_ID, message: { content: text } },
+      SESSION_ID,
+    ),
+    'claude-sonnet-5',
+  );
+  assert.equal(
+    extractClaudeEventModel(
+      { sessionId: SESSION_ID, message: { content: [{ text }] } },
+      SESSION_ID,
+    ),
+    'claude-sonnet-5',
+  );
+});
+
+test('falls back to the message model when every content hit is a placeholder', () => {
+  assert.equal(
+    extractClaudeEventModel(
+      {
+        sessionId: SESSION_ID,
+        message: {
+          content: '<model><synthetic></model>',
+          model: 'claude-sonnet-5',
+        },
+      },
+      SESSION_ID,
+    ),
+    'claude-sonnet-5',
+  );
+});

package/server/modules/providers/tests/claude-runtime-hold.test.ts

@@ -0,0 +1,309 @@
+import assert from 'node:assert/strict';
+import { mkdtemp, rm } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+import test from 'node:test';
+
+import { ClaudeSessionsProvider } from '@/modules/providers/list/claude/claude-sessions.provider.js';
+import { CLAUDE_PREDEFINED_MODELS } from '@/modules/providers/list/claude/claude-models.provider.js';
+import {
+  listClaudeSDKBackgroundWork,
+  queryClaudeSDK,
+  stopClaudeSDKTask,
+} from '@/modules/providers/list/claude/claude-runtime.provider.js';
+import type { NormalizedMessage, ProviderRuntimeContext } from '@/shared/types.js';
+
+/**
+ * The runtime keeps the CLI's stdin open after a turn's `result` while the
+ * turn's background work is outstanding, and lets go when that work has
+ * reported. These drive `queryClaudeSDK` with a scripted SDK stream — the
+ * seam is `context.createQuery` — and watch the held prompt stream: the CLI
+ * exits when it ends, so "released" is the whole outcome.
+ */
+
+const SESSION_ID = 'app-hold-session';
+const NATIVE_ID = 'native-hold-session';
+
+type Scripted = {
+  emit: (message: Record<string, unknown>) => void;
+  end: () => void;
+  released: () => boolean;
+  stopped: string[];
+};
+
+/** A stand-in for the SDK query: yields what the test emits, and reads the held prompt to notice its release. */
+function createScriptedQuery(): { createQuery: NonNullable<ProviderRuntimeContext['createQuery']>; script: Scripted } {
+  const queue: Array<Record<string, unknown> | null> = [];
+  let wake: (() => void) | null = null;
+  let released = false;
+  const stopped: string[] = [];
+
+  const script: Scripted = {
+    emit: (message) => { queue.push(message); wake?.(); },
+    end: () => { queue.push(null); wake?.(); },
+    released: () => released,
+    stopped,
+  };
+
+  const createQuery: NonNullable<ProviderRuntimeContext['createQuery']> = ({ prompt }) => {
+    void (async () => {
+      for await (const _message of prompt) { /* the CLI reads its stdin */ }
+      released = true;
+    })();
+
+    const iterator = (async function* () {
+      for (;;) {
+        if (queue.length === 0) {
+          await new Promise<void>((resolve) => { wake = resolve; });
+          wake = null;
+          continue;
+        }
+        const next = queue.shift();
+        if (next === null || next === undefined) {
+          return;
+        }
+        yield next;
+      }
+    })();
+
+    return Object.assign(iterator, {
+      interrupt: async () => {},
+      stopTask: async (taskId: string) => { stopped.push(taskId); },
+    });
+  };
+
+  return { createQuery, script };
+}
+
+async function withRun(
+  runTest: (context: { script: Scripted; sent: NormalizedMessage[]; done: Promise<unknown> }) => Promise<void>,
+): Promise<void> {
+  const cwd = await mkdtemp(path.join(os.tmpdir(), 'claude-runtime-hold-'));
+  const { createQuery, script } = createScriptedQuery();
+  const sent: NormalizedMessage[] = [];
+  const writer = { send: (message: NormalizedMessage) => { sent.push(message); }, userId: null };
+  const sessions = new ClaudeSessionsProvider({ getLiveRunStartTime: () => null });
+  const context: ProviderRuntimeContext = {
+    resolveProviderSessionId: () => null,
+    resolveResumeModel: async () => undefined,
+    getProviderModels: async () => CLAUDE_PREDEFINED_MODELS as never,
+    normalizeMessage: (raw, sessionId) => sessions.normalizeMessage(raw, sessionId),
+    isProviderInstalled: async () => true,
+    createQuery,
+  };
+
+  try {
+    const done = queryClaudeSDK('hello', { sessionId: SESSION_ID, cwd }, writer as never, context);
+    await runTest({ script, sent, done });
+    script.end();
+    await done;
+  } finally {
+    await rm(cwd, { recursive: true, force: true });
+  }
+}
+
+const settle = () => new Promise((resolve) => { setTimeout(resolve, 25); });
+
+// A fixed settle can be too short on a loaded machine; the supersede tests
+// wait for the run to have actually processed its turn instead.
+async function waitFor(condition: () => boolean): Promise<void> {
+  for (let attempt = 0; attempt < 200 && !condition(); attempt += 1) {
+    await settle();
+  }
+  assert.ok(condition(), 'condition never became true');
+}
+
+const init = () => ({ type: 'system', subtype: 'init', session_id: NATIVE_ID });
+const toolUse = (id: string, name: string, input: Record<string, unknown>) => ({
+  type: 'assistant', session_id: NATIVE_ID, parent_tool_use_id: null,
+  message: { role: 'assistant', content: [{ type: 'tool_use', id, name, input }] },
+});
+const ack = (id: string, text: string, toolUseResult: Record<string, unknown>) => ({
+  type: 'user', session_id: NATIVE_ID, parent_tool_use_id: null,
+  message: { role: 'user', content: [{ type: 'tool_result', tool_use_id: id, content: text }] },
+  tool_use_result: toolUseResult,
+});
+const taskStarted = (taskId: string, toolUseId: string, taskType: string) => ({
+  type: 'system', subtype: 'task_started', session_id: NATIVE_ID, task_id: taskId, tool_use_id: toolUseId, description: `Task ${taskId}`, task_type: taskType,
+});
+const taskNotification = (taskId: string, toolUseId: string, status: string) => ({
+  type: 'system', subtype: 'task_notification', session_id: NATIVE_ID, task_id: taskId, tool_use_id: toolUseId, status, summary: `Task ${taskId} ${status}`, output_file: '',
+});
+const result = () => ({ type: 'result', subtype: 'success', session_id: NATIVE_ID, result: 'launched', duration_ms: 1, num_turns: 1 });
+
+test('stopping the last outstanding task releases the held process', async () => {
+  await withRun(async ({ script, sent, done }) => {
+    script.emit(init());
+    script.emit(toolUse('toolu_wf', 'Workflow', { script: 'export const meta = {}' }));
+    script.emit(taskStarted('wf1', 'toolu_wf', 'local_workflow'));
+    script.emit(ack('toolu_wf', 'Workflow launched in background. Task ID: wf1', { status: 'async_launched', taskId: 'wf1', taskType: 'local_workflow' }));
+    script.emit(result());
+    await settle();
+
+    // The turn is over for the client, the process is held for the workflow.
+    assert.ok(sent.some((message) => message.kind === 'complete'));
+    assert.deepEqual(listClaudeSDKBackgroundWork().map((entry) => [entry.sessionId, entry.tasks.map((task) => task.taskId)]), [[SESSION_ID, ['wf1']]]);
+    assert.equal(script.released(), false, 'stdin stays open while the workflow runs');
+
+    // The user stops it. The CLI answers with a `stopped` notification and
+    // pushes no follow-up turn, so nothing else would ever end the hold.
+    assert.equal(await stopClaudeSDKTask(SESSION_ID, 'wf1'), true);
+    assert.deepEqual(script.stopped, ['wf1']);
+    script.emit(taskNotification('wf1', 'toolu_wf', 'stopped'));
+    await settle();
+
+    assert.deepEqual(listClaudeSDKBackgroundWork(), []);
+    assert.equal(script.released(), true, 'the process is let go once nothing is outstanding');
+    void done;
+  });
+});
+
+test('a task that reported completed keeps the hold for the turn that relays its result', async () => {
+  await withRun(async ({ script }) => {
+    script.emit(init());
+    script.emit(toolUse('toolu_wf', 'Workflow', { script: 'export const meta = {}' }));
+    script.emit(taskStarted('wf1', 'toolu_wf', 'local_workflow'));
+    script.emit(ack('toolu_wf', 'Workflow launched in background. Task ID: wf1', { status: 'async_launched', taskId: 'wf1', taskType: 'local_workflow' }));
+    script.emit(result());
+    await settle();
+
+    // Completed, unlike stopped, is followed by a turn the CLI pushes to relay
+    // the result; closing stdin at the notification would cut it short.
+    script.emit(taskNotification('wf1', 'toolu_wf', 'completed'));
+    await settle();
+    assert.equal(script.released(), false);
+
+    script.emit(result());
+    await settle();
+    assert.equal(script.released(), true, 'the follow-up turn\'s result ends the hold');
+  });
+});
+
+test('an agent that ran in the foreground and settled before the result does not hold the process', async () => {
+  await withRun(async ({ script }) => {
+    // An Agent call without `run_in_background` is scored as background by
+    // the static rule, but the CLI ran it in the foreground: its task started
+    // and settled before the turn ended. The task events know that.
+    script.emit(init());
+    script.emit(toolUse('toolu_agent', 'Agent', { prompt: 'Return the word FOUR', subagent_type: 'general-purpose' }));
+    script.emit(taskStarted('a1', 'toolu_agent', 'local_agent'));
+    script.emit(taskNotification('a1', 'toolu_agent', 'completed'));
+    script.emit(ack('toolu_agent', 'FOUR', { status: 'completed', agentId: 'a1' }));
+    script.emit(result());
+    await settle();
+
+    assert.deepEqual(listClaudeSDKBackgroundWork(), []);
+    assert.equal(script.released(), true, 'nothing is outstanding, so nothing to hold for');
+  });
+});
+
+test('a new turn stops the tasks the held process still has going before letting it go', async () => {
+  // Two runs on one session, each with its own scripted CLI: the first is
+  // held for a background agent, the second is the user's next message.
+  const cwd = await mkdtemp(path.join(os.tmpdir(), 'claude-runtime-hold-'));
+  const first = createScriptedQuery();
+  const second = createScriptedQuery();
+  const sent: NormalizedMessage[] = [];
+  const writer = { send: (message: NormalizedMessage) => { sent.push(message); }, userId: null };
+  const sessions = new ClaudeSessionsProvider({ getLiveRunStartTime: () => null });
+  const context: Omit<ProviderRuntimeContext, 'createQuery'> = {
+    resolveProviderSessionId: () => null,
+    resolveResumeModel: async () => undefined,
+    getProviderModels: async () => CLAUDE_PREDEFINED_MODELS as never,
+    normalizeMessage: (raw, sessionId) => sessions.normalizeMessage(raw, sessionId),
+    isProviderInstalled: async () => true,
+  };
+
+  try {
+    const firstRun = queryClaudeSDK('start an agent', { sessionId: SESSION_ID, cwd }, writer as never, { ...context, createQuery: first.createQuery });
+    await settle();
+    first.script.emit(init());
+    first.script.emit(toolUse('toolu_agent', 'Agent', { prompt: 'sleep 90', subagent_type: 'general-purpose', run_in_background: true }));
+    first.script.emit(taskStarted('a1', 'toolu_agent', 'local_agent'));
+    first.script.emit(ack('toolu_agent', 'Async agent launched successfully.', { status: 'async_launched', agentId: 'a1' }));
+    first.script.emit(result());
+    await waitFor(() => sent.some((message) => message.kind === 'complete'));
+    assert.equal(first.script.released(), false, 'the first process is held for its agent');
+
+    // Closing stdin alone would leave the CLI waiting for the agent, so the
+    // superseded process would live on next to the new one for as long as
+    // the agent runs. The stop has to reach it before stdin ends, because the
+    // SDK drops control requests written to an ended stdin.
+    const secondRun = queryClaudeSDK('Continue', { sessionId: SESSION_ID, cwd }, writer as never, { ...context, createQuery: second.createQuery });
+    await settle();
+    assert.deepEqual(first.script.stopped, ['a1'], 'the held process is told to stop its agent');
+    assert.equal(first.script.released(), true, 'and is then let go');
+    assert.deepEqual(second.script.stopped, [], 'the new process is untouched');
+
+    first.script.emit(taskNotification('a1', 'toolu_agent', 'stopped'));
+    first.script.end();
+    await firstRun;
+    second.script.emit(init());
+    second.script.emit(result());
+    second.script.end();
+    await secondRun;
+    assert.deepEqual(listClaudeSDKBackgroundWork(), []);
+  } finally {
+    await rm(cwd, { recursive: true, force: true });
+  }
+});
+
+test('a held process is still let go when stopping its tasks throws', async () => {
+  // Stopping is best effort; a stop that throws on the spot must not keep the
+  // held process alive or take the new turn down with it.
+  const cwd = await mkdtemp(path.join(os.tmpdir(), 'claude-runtime-hold-'));
+  const first = createScriptedQuery();
+  const second = createScriptedQuery();
+  const sent: NormalizedMessage[] = [];
+  const writer = { send: (message: NormalizedMessage) => { sent.push(message); }, userId: null };
+  const sessions = new ClaudeSessionsProvider({ getLiveRunStartTime: () => null });
+  const context: Omit<ProviderRuntimeContext, 'createQuery'> = {
+    resolveProviderSessionId: () => null,
+    resolveResumeModel: async () => undefined,
+    getProviderModels: async () => CLAUDE_PREDEFINED_MODELS as never,
+    normalizeMessage: (raw, sessionId) => sessions.normalizeMessage(raw, sessionId),
+    isProviderInstalled: async () => true,
+  };
+  const throwingStop: NonNullable<ProviderRuntimeContext['createQuery']> = (options) =>
+    Object.assign(first.createQuery(options), {
+      stopTask: () => { throw new Error('stop failed'); },
+    });
+
+  try {
+    const firstRun = queryClaudeSDK('start an agent', { sessionId: SESSION_ID, cwd }, writer as never, { ...context, createQuery: throwingStop });
+    await settle();
+    first.script.emit(init());
+    first.script.emit(toolUse('toolu_agent', 'Agent', { prompt: 'sleep 90', subagent_type: 'general-purpose', run_in_background: true }));
+    first.script.emit(taskStarted('a1', 'toolu_agent', 'local_agent'));
+    first.script.emit(ack('toolu_agent', 'Async agent launched successfully.', { status: 'async_launched', agentId: 'a1' }));
+    first.script.emit(result());
+    await waitFor(() => sent.some((message) => message.kind === 'complete'));
+    assert.equal(first.script.released(), false, 'the first process is held for its agent');
+
+    const secondRun = queryClaudeSDK('Continue', { sessionId: SESSION_ID, cwd }, writer as never, { ...context, createQuery: second.createQuery });
+    await settle();
+    assert.equal(first.script.released(), true, 'the held process is let go anyway');
+
+    first.script.end();
+    await firstRun;
+    second.script.emit(init());
+    second.script.emit(result());
+    second.script.end();
+    await secondRun;
+    assert.equal(second.script.released(), true, 'the new turn ran to completion');
+  } finally {
+    await rm(cwd, { recursive: true, force: true });
+  }
+});
+
+test('a turn whose tool emits no task events still holds on the static rule', async () => {
+  await withRun(async ({ script }) => {
+    script.emit(init());
+    script.emit(toolUse('toolu_monitor', 'Monitor', { command: 'tail -f x', description: 'watch', timeout_ms: 1000 }));
+    script.emit(ack('toolu_monitor', 'Monitor started', {}));
+    script.emit(result());
+    await settle();
+
+    assert.equal(script.released(), false, 'Monitor reports no task, so the launch rule decides');
+  });
+});