@geminilight/mindos 0.6.61 → 0.6.65

package/_standalone/__tests__/core/context.test.ts

@@ -1,428 +1,50 @@
-import { describe, it, expect, vi, beforeEach } from 'vitest';
-import {
-  truncateToolOutputs,
-  hardPrune,
-  estimateTokens,
-  estimateStringTokens,
-  getContextLimit,
-  needsCompact,
-  compactMessages,
-} from '../../lib/agent/context';
-import type { AgentMessage } from '@mariozechner/pi-agent-core';
-import type { Model } from '@mariozechner/pi-ai';
-
-// Mock pi-ai's complete() — intercept the summarizer call
-vi.mock('@mariozechner/pi-ai', async (importOriginal) => {
-  const actual = await importOriginal<typeof import('@mariozechner/pi-ai')>();
-  return {
-    ...actual,
-    complete: vi.fn(),
-  };
-});
-
-import { complete } from '@mariozechner/pi-ai';
-const mockComplete = vi.mocked(complete);
-
-// ---------------------------------------------------------------------------
-// Helpers — build AgentMessage objects matching pi-ai types
-// ---------------------------------------------------------------------------
-
-const now = Date.now();
-
-/** Create a UserMessage */
-function userMsg(text: string): AgentMessage {
-  return { role: 'user', content: text, timestamp: now } as AgentMessage;
-}
-
-/** Create an AssistantMessage */
-function assistantMsg(text: string): AgentMessage {
-  return {
-    role: 'assistant',
-    content: [{ type: 'text', text }],
-    api: 'anthropic-messages',
-    provider: 'anthropic',
-    model: 'claude-sonnet-4-6',
-    usage: { inputTokens: 0, outputTokens: 0 },
-  } as AgentMessage;
-}
-
-/** Create a ToolResultMessage */
-function toolResultMsg(toolName: string, text: string): AgentMessage {
-  return {
-    role: 'toolResult',
-    toolCallId: `call_${toolName}`,
-    toolName,
-    content: [{ type: 'text', text }],
-  } as AgentMessage;
-}
-
-/** Create an assistant message with a tool call */
-function assistantWithToolCall(toolName: string): AgentMessage {
-  return {
-    role: 'assistant',
-    content: [{ type: 'toolCall', toolCallId: `call_${toolName}`, toolName, args: {} }],
-    api: 'anthropic-messages',
-    provider: 'anthropic',
-    model: 'claude-sonnet-4-6',
-    usage: { inputTokens: 0, outputTokens: 0 },
-  } as AgentMessage;
-}
-
-/** Dummy Model (only used as pass-through; complete is mocked) */
-const fakeModel = {} as Model<any>;
-
-// ---------------------------------------------------------------------------
-// Token estimation
-// ---------------------------------------------------------------------------
-
-describe('context: token estimation', () => {
-  it('estimates string tokens as length/4', () => {
-    expect(estimateStringTokens('abcd')).toBe(1);
-    expect(estimateStringTokens('abcde')).toBe(2);
-    expect(estimateStringTokens('')).toBe(0);
-  });
-
-  it('estimates message tokens for string content', () => {
-    const msgs = [userMsg('a'.repeat(100))];
-    expect(estimateTokens(msgs)).toBe(25);
-  });
-
-  it('returns 0 for messages with no content', () => {
-    const msgs = [{ role: 'user' as const, content: undefined }] as unknown as AgentMessage[];
-    expect(estimateTokens(msgs)).toBe(0);
-  });
-
-  it('estimates tokens for array content (assistant with text)', () => {
-    const msgs = [assistantMsg('a'.repeat(100))];
-    expect(estimateTokens(msgs)).toBe(25);
-  });
-
-  it('estimates tokens for tool call args', () => {
-    const msgs = [assistantWithToolCall('read_file')];
-    // args = {} → JSON.stringify = "{}" = 2 chars → ceil(2/4) = 1
-    expect(estimateTokens(msgs)).toBeGreaterThanOrEqual(1);
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Context limits
-// ---------------------------------------------------------------------------
-
-describe('context: model limits', () => {
-  it('returns correct limit for known models', () => {
-    expect(getContextLimit('claude-3.5-sonnet')).toBe(200_000);
-    expect(getContextLimit('gpt-4o-mini')).toBe(128_000);
-    expect(getContextLimit('gpt-3.5-turbo')).toBe(16_000);
-  });
-
-  it('returns default for unknown models', () => {
-    expect(getContextLimit('llama-3-70b')).toBe(100_000);
-  });
-
-  it('matches gpt-4o before gpt-4 (prefix sort)', () => {
-    expect(getContextLimit('gpt-4o-mini')).toBe(128_000);
-    expect(getContextLimit('gpt-4o-2024-05-13')).toBe(128_000);
-    expect(getContextLimit('gpt-4-turbo')).toBe(128_000);
-  });
-
-  it('matches gpt-5 models correctly', () => {
-    expect(getContextLimit('gpt-5')).toBe(200_000);
-    expect(getContextLimit('gpt-5.4')).toBe(200_000);
-  });
-
-  it('is case-insensitive', () => {
-    expect(getContextLimit('Claude-3.5-Sonnet')).toBe(200_000);
-    expect(getContextLimit('GPT-4O-MINI')).toBe(128_000);
-  });
-
-  it('needsCompact detects threshold breach', () => {
-    const bigMsg = userMsg('x'.repeat(560_001));
-    expect(needsCompact([bigMsg], '', 'claude-3.5-sonnet')).toBe(true);
-
-    const smallMsg = userMsg('hello');
-    expect(needsCompact([smallMsg], '', 'claude-3.5-sonnet')).toBe(false);
-  });
-});
-
-// ---------------------------------------------------------------------------
-// truncateToolOutputs
-// ---------------------------------------------------------------------------
-
-describe('context: truncateToolOutputs', () => {
-  it('truncates long tool outputs in non-last tool messages', () => {
-    const msgs: AgentMessage[] = [
-      userMsg('hello'),
-      toolResultMsg('read_file', 'x'.repeat(5000)),
-      assistantMsg('done'),
-      toolResultMsg('read_file', 'y'.repeat(5000)),  // last tool — kept intact
-    ];
-    const result = truncateToolOutputs(msgs);
-    // First tool msg (idx 1) should be truncated (read_file limit = 2000)
-    const firstTool = result[1] as any;
-    expect(firstTool.content[0].text.length).toBeLessThan(5000);
-    expect(firstTool.content[0].text).toContain('[...truncated');
-    // Last tool msg (idx 3) should be untouched
-    const lastTool = result[3] as any;
-    expect(lastTool.content[0].text).toBe('y'.repeat(5000));
-  });
-
-  it('handles empty messages array', () => {
-    expect(truncateToolOutputs([])).toEqual([]);
-  });
-
-  it('handles messages with no tool messages', () => {
-    const msgs = [userMsg('hi'), assistantMsg('hello')];
-    const result = truncateToolOutputs(msgs);
-    expect(result).toEqual(msgs);
-  });
-
-  it('does not truncate short tool outputs', () => {
-    const msgs: AgentMessage[] = [
-      toolResultMsg('write_file', 'File written: test.md'),
-      toolResultMsg('read_file', 'short content'),
-    ];
-    const result = truncateToolOutputs(msgs);
-    // Last tool kept intact, first tool is short enough
-    expect((result[0] as any).content[0].text).toBe('File written: test.md');
-  });
-});
-
-// ---------------------------------------------------------------------------
-// hardPrune
-// ---------------------------------------------------------------------------
-
-describe('context: hardPrune', () => {
-  it('returns messages unchanged when under threshold', () => {
-    const msgs = [userMsg('hello'), assistantMsg('hi')];
-    const result = hardPrune(msgs, '', 'claude');
-    expect(result).toBe(msgs); // same reference — no pruning
-  });
-
-  it('prunes earliest messages when over 90% context', () => {
-    const msgs: AgentMessage[] = [
-      userMsg('a'.repeat(400_000)),
-      assistantMsg('b'.repeat(400_000)),
-      userMsg('c'.repeat(100)),
-      assistantMsg('d'.repeat(100)),
-    ];
-    const result = hardPrune(msgs, '', 'claude');
-    expect(result.length).toBeLessThan(msgs.length);
-  });
-
-  it('ensures first message is user role after pruning', () => {
-    const msgs: AgentMessage[] = [
-      userMsg('a'.repeat(100_000)),
-      assistantMsg('b'.repeat(100_000)),
-      assistantMsg('c'.repeat(1000)),
-      userMsg('keep this'),
-      assistantMsg('and this'),
-    ];
-    const result = hardPrune(msgs, '', 'gpt-3.5');
-    expect(result.length).toBeGreaterThan(0);
-    expect((result[0] as any).role).toBe('user');
-  });
-
-  it('skips toolResult messages that are orphaned after cut', () => {
-    const msgs: AgentMessage[] = [
-      userMsg('a'.repeat(100_000)),
-      assistantMsg('b'.repeat(100_000)),
-      toolResultMsg('read_file', 'x'.repeat(1000)),
-      userMsg('keep'),
-      assistantMsg('response'),
-    ];
-    const result = hardPrune(msgs, '', 'gpt-3.5');
-    expect((result[0] as any).role).toBe('user');
-    expect((result[0] as any).role).not.toBe('toolResult');
-  });
-
-  it('injects synthetic user message when no user message remains', () => {
-    const msgs: AgentMessage[] = [
-      userMsg('a'.repeat(200_000)),
-      assistantMsg('b'.repeat(1000)),
-      assistantMsg('c'.repeat(1000)),
-    ];
-    const result = hardPrune(msgs, '', 'gpt-3.5');
-    expect((result[0] as any).role).toBe('user');
-    const content = (result[0] as any).content;
-    expect(typeof content === 'string' && content.includes('truncated')).toBe(true);
-  });
-
-  it('preserves messages when exactly at threshold', () => {
-    const msgs = [userMsg('small'), assistantMsg('also small')];
-    const result = hardPrune(msgs, '', 'claude');
-    expect(result).toBe(msgs);
-  });
-});
-
-// ---------------------------------------------------------------------------
-// compactMessages
-// ---------------------------------------------------------------------------
-
-describe('context: compactMessages', () => {
-  beforeEach(() => {
-    mockComplete.mockReset();
-  });
-
-  it('returns uncompacted when fewer than 6 messages', async () => {
-    const msgs = [userMsg('a'), assistantMsg('b'), userMsg('c'), assistantMsg('d')];
-    const result = await compactMessages(msgs, fakeModel, 'test-key', '', 'claude');
-    expect(result.compacted).toBe(false);
-    expect(result.messages).toBe(msgs);
-    expect(mockComplete).not.toHaveBeenCalled();
-  });
-
-  it('returns uncompacted when early messages < 2', async () => {
-    const msgs = [
-      userMsg('only-early'),
-      assistantMsg('r1'), userMsg('r2'), assistantMsg('r3'),
-      userMsg('r4'), assistantMsg('r5'), userMsg('r6'),
-    ];
-    const result = await compactMessages(msgs, fakeModel, 'test-key', '', 'claude');
-    expect(result.compacted).toBe(false);
-  });
-
-  it('prepends summary as separate user message when recentMessages[0] is assistant', async () => {
-    mockComplete.mockResolvedValue({
-      role: 'assistant',
-      content: [{ type: 'text', text: 'Summary of conversation.' }],
-      api: 'anthropic-messages', provider: 'anthropic', model: 'claude',
-      usage: { inputTokens: 0, outputTokens: 0 },
-    } as any);
-
-    const msgs: AgentMessage[] = [
-      userMsg('early1'),
-      assistantMsg('early2'),
-      assistantMsg('recent1'),
-      userMsg('recent2'),
-      assistantMsg('recent3'),
-      userMsg('recent4'),
-      assistantMsg('recent5'),
-      userMsg('recent6'),
-    ];
-
-    const result = await compactMessages(msgs, fakeModel, 'test-key', '', 'claude');
-    expect(result.compacted).toBe(true);
-    expect(result.messages.length).toBe(7); // 1 summary + 6 recent
-    expect((result.messages[0] as any).role).toBe('user');
-    const content = (result.messages[0] as any).content;
-    expect(typeof content === 'string').toBe(true);
-    expect(content).toContain('Summary of conversation.');
-    expect((result.messages[1] as any).role).toBe('assistant');
-  });
-
-  it('merges summary into first user message to avoid consecutive user→user', async () => {
-    mockComplete.mockResolvedValue({
-      role: 'assistant',
-      content: [{ type: 'text', text: 'This is the summary.' }],
-      api: 'anthropic-messages', provider: 'anthropic', model: 'claude',
-      usage: { inputTokens: 0, outputTokens: 0 },
-    } as any);
-
-    const msgs: AgentMessage[] = [
-      userMsg('early1'),
-      assistantMsg('early2'),
-      userMsg('recent-user-first'),
-      assistantMsg('recent2'),
-      userMsg('recent3'),
-      assistantMsg('recent4'),
-      userMsg('recent5'),
-      assistantMsg('recent6'),
-    ];
-
-    const result = await compactMessages(msgs, fakeModel, 'test-key', '', 'claude');
-    expect(result.compacted).toBe(true);
-    expect(result.messages.length).toBe(6);
-    expect((result.messages[0] as any).role).toBe('user');
-    const content = (result.messages[0] as any).content;
-    expect(content).toContain('This is the summary.');
-    expect(content).toContain('recent-user-first');
-    expect((result.messages[1] as any).role).toBe('assistant');
-  });
-
-  it('handles multimodal (array) content in first user message', async () => {
-    mockComplete.mockResolvedValue({
-      role: 'assistant',
-      content: [{ type: 'text', text: 'Multimodal summary.' }],
-      api: 'anthropic-messages', provider: 'anthropic', model: 'claude',
-      usage: { inputTokens: 0, outputTokens: 0 },
-    } as any);
-
-    const multimodalUser: AgentMessage = {
-      role: 'user',
-      content: [
-        { type: 'image', image: new Uint8Array(0), mimeType: 'image/png' },
-        { type: 'text', text: 'describe this image' },
-      ],
-      timestamp: now,
-    } as unknown as AgentMessage;
-
-    const msgs: AgentMessage[] = [
-      userMsg('early1'),
-      assistantMsg('early2'),
-      multimodalUser,
-      assistantMsg('recent2'),
-      userMsg('recent3'),
-      assistantMsg('recent4'),
-      userMsg('recent5'),
-      assistantMsg('recent6'),
-    ];
-
-    const result = await compactMessages(msgs, fakeModel, 'test-key', '', 'claude');
-    expect(result.compacted).toBe(true);
-    expect(result.messages.length).toBe(6);
-    expect((result.messages[0] as any).role).toBe('user');
-    const content = (result.messages[0] as any).content;
-    expect(Array.isArray(content)).toBe(true);
-    const parts = content as any[];
-    expect(parts[0].type).toBe('text');
-    expect(parts[0].text).toContain('Multimodal summary.');
-  });
-
-  it('falls back to hardPrune when complete() throws, then throws if pruning insufficient', async () => {
-    mockComplete.mockRejectedValue(new Error('API error'));
-
-    const msgs: AgentMessage[] = [
-      userMsg('early1'), assistantMsg('early2'),
-      userMsg('recent1'), assistantMsg('recent2'),
-      userMsg('recent3'), assistantMsg('recent4'),
-      userMsg('recent5'), assistantMsg('recent6'),
-    ];
-
-    // Small messages on a large model — hardPrune won't help, so it throws
-    await expect(
-      compactMessages(msgs, fakeModel, 'test-key', '', 'claude'),
-    ).rejects.toThrow('Context compaction failed');
-  });
-
-  it('does not split between assistant tool-call and tool result (M6)', async () => {
-    mockComplete.mockResolvedValue({
-      role: 'assistant',
-      content: [{ type: 'text', text: 'Tool summary.' }],
-      api: 'anthropic-messages', provider: 'anthropic', model: 'claude',
-      usage: { inputTokens: 0, outputTokens: 0 },
-    } as any);
-
-    const msgs: AgentMessage[] = [
-      userMsg('e1'),
-      assistantMsg('e2'),
-      userMsg('e3'),
-      assistantWithToolCall('read_file'),
-      toolResultMsg('read_file', 'file content here'),
-      userMsg('r1'),
-      assistantMsg('r2'),
-      userMsg('r3'),
-      assistantMsg('r4'),
-      userMsg('r5'),
-    ];
-
-    const result = await compactMessages(msgs, fakeModel, 'test-key', '', 'claude');
-    expect(result.compacted).toBe(true);
-
-    for (let i = 0; i < result.messages.length; i++) {
-      if ((result.messages[i] as any).role === 'toolResult') {
-        expect(i).toBeGreaterThan(0);
-        expect((result.messages[i - 1] as any).role).toBe('assistant');
-      }
-    }
+import { describe, it, expect } from 'vitest';
+import { estimateStringTokens } from '../../lib/agent/context';
+
+describe('Context estimation', () => {
+  describe('estimateStringTokens', () => {
+    it('estimates ASCII text tokens (~0.25 per char)', () => {
+      const text = 'Hello, this is a test sentence with some words.';
+      const tokens = estimateStringTokens(text);
+      // ~48 chars → ~12 tokens
+      expect(tokens).toBeGreaterThan(5);
+      expect(tokens).toBeLessThan(25);
+    });
+
+    it('estimates CJK text tokens (~1.5 per char)', () => {
+      const text = '你好世界这是测试';
+      const tokens = estimateStringTokens(text);
+      // 8 CJK chars → ~12 tokens
+      expect(tokens).toBeGreaterThan(8);
+      expect(tokens).toBeLessThan(20);
+    });
+
+    it('handles mixed CJK and ASCII', () => {
+      const text = 'Hello 你好 World 世界';
+      const tokens = estimateStringTokens(text);
+      expect(tokens).toBeGreaterThan(5);
+      expect(tokens).toBeLessThan(20);
+    });
+
+    it('returns 0 for empty string', () => {
+      expect(estimateStringTokens('')).toBe(0);
+    });
+
+    it('estimates a typical system prompt size', () => {
+      // Simulate a moderate system prompt
+      const prompt = 'You are a helpful assistant.\n'.repeat(100);
+      const tokens = estimateStringTokens(prompt);
+      // ~2800 chars → ~700 tokens
+      expect(tokens).toBeGreaterThan(500);
+      expect(tokens).toBeLessThan(1000);
+    });
+
+    it('estimates large content (20K chars) reasonably', () => {
+      const large = 'a'.repeat(20_000);
+      const tokens = estimateStringTokens(large);
+      // 20000 ASCII chars → ~5000 tokens
+      expect(tokens).toBeGreaterThanOrEqual(4000);
+      expect(tokens).toBeLessThanOrEqual(6000);
+    });
   });
 });

package/_standalone/__tests__/lib/pi-skills.test.ts

@@ -38,13 +38,13 @@ describe('pi skill integration', () => {
     expect(result).toEqual({ name: 'test-skill', description: 'useful helper' });
   });
 
-  it('scans MindOS skill directories with precedence and metadata', () => {
+  it('scans MindOS skill directories with precedence and metadata', async () => {
     writeSkill(path.join(projectRoot, 'app', 'data', 'skills'), 'mindos', '---\nname: mindos\ndescription: builtin\n---\n');
     writeSkill(path.join(projectRoot, 'skills'), 'project-helper', '---\nname: project-helper\ndescription: project builtin\n---\n');
     writeSkill(path.join(mindRoot, '.skills'), 'user-helper', '---\nname: user-helper\ndescription: user custom\n---\n');
     writeSkill(path.join(tempRoot, '.mindos', 'skills'), 'global-helper', '---\nname: global-helper\ndescription: global skill\n---\n');
 
-    const skills = scanSkillDirs({ projectRoot, mindRoot, disabledSkills: ['project-helper'] });
+    const skills = await scanSkillDirs({ projectRoot, mindRoot, disabledSkills: ['project-helper'] });
 
     expect(skills.map((skill) => skill.name)).toEqual(['mindos', 'project-helper', 'user-helper', 'global-helper']);
     expect(skills.find((skill) => skill.name === 'mindos')).toMatchObject({ source: 'builtin', editable: false, origin: 'app-builtin', enabled: true });
@@ -53,11 +53,11 @@ describe('pi skill integration', () => {
     expect(skills.find((skill) => skill.name === 'global-helper')).toMatchObject({ source: 'user', editable: true, origin: 'mindos-global', enabled: true });
   });
 
-  it('knowledge base skill takes precedence over global skill with same name', () => {
+  it('knowledge base skill takes precedence over global skill with same name', async () => {
     writeSkill(path.join(mindRoot, '.skills'), 'shared-skill', '---\nname: shared-skill\ndescription: from kb\n---\n');
     writeSkill(path.join(tempRoot, '.mindos', 'skills'), 'shared-skill', '---\nname: shared-skill\ndescription: from global\n---\n');
 
-    const skills = scanSkillDirs({ projectRoot, mindRoot });
+    const skills = await scanSkillDirs({ projectRoot, mindRoot });
     const matched = skills.filter((s) => s.name === 'shared-skill');
     expect(matched).toHaveLength(1);
     expect(matched[0].origin).toBe('mindos-user');

package/_standalone/__tests__/lib/settings-ai-client.test.ts

@@ -6,8 +6,11 @@ describe('isAiConfiguredForAsk', () => {
     expect(
       isAiConfiguredForAsk({
         ai: {
-          provider: 'anthropic',
-          providers: { anthropic: { apiKey: '***set***' }, openai: { apiKey: '' } },
+          activeProvider: 'p_anthro01',
+          providers: [
+            { id: 'p_anthro01', name: 'Anthropic', protocol: 'anthropic', apiKey: 'sk-ant-test', model: 'claude-sonnet-4-6', baseUrl: '' },
+            { id: 'p_openai01', name: 'OpenAI', protocol: 'openai', apiKey: '', model: 'gpt-5.4', baseUrl: '' },
+          ],
         },
         envOverrides: {},
       }),
@@ -18,8 +21,11 @@ describe('isAiConfiguredForAsk', () => {
     expect(
       isAiConfiguredForAsk({
         ai: {
-          provider: 'anthropic',
-          providers: { anthropic: { apiKey: '' }, openai: { apiKey: '' } },
+          activeProvider: 'p_anthro01',
+          providers: [
+            { id: 'p_anthro01', name: 'Anthropic', protocol: 'anthropic', apiKey: '', model: 'claude-sonnet-4-6', baseUrl: '' },
+            { id: 'p_openai01', name: 'OpenAI', protocol: 'openai', apiKey: '', model: 'gpt-5.4', baseUrl: '' },
+          ],
         },
         envOverrides: { ANTHROPIC_API_KEY: true },
       }),
@@ -30,8 +36,11 @@ describe('isAiConfiguredForAsk', () => {
     expect(
       isAiConfiguredForAsk({
         ai: {
-          provider: 'anthropic',
-          providers: { anthropic: { apiKey: '' }, openai: { apiKey: '***set***' } },
+          activeProvider: 'p_anthro01',
+          providers: [
+            { id: 'p_anthro01', name: 'Anthropic', protocol: 'anthropic', apiKey: '', model: 'claude-sonnet-4-6', baseUrl: '' },
+            { id: 'p_openai01', name: 'OpenAI', protocol: 'openai', apiKey: 'sk-openai-test', model: 'gpt-5.4', baseUrl: '' },
+          ],
         },
         envOverrides: {},
       }),
@@ -42,8 +51,11 @@ describe('isAiConfiguredForAsk', () => {
     expect(
       isAiConfiguredForAsk({
         ai: {
-          provider: 'openai',
-          providers: { anthropic: { apiKey: '' }, openai: { apiKey: '***set***' } },
+          activeProvider: 'p_openai01',
+          providers: [
+            { id: 'p_anthro01', name: 'Anthropic', protocol: 'anthropic', apiKey: '', model: 'claude-sonnet-4-6', baseUrl: '' },
+            { id: 'p_openai01', name: 'OpenAI', protocol: 'openai', apiKey: 'sk-openai-test', model: 'gpt-5.4', baseUrl: '' },
+          ],
         },
         envOverrides: {},
       }),
@@ -54,18 +66,26 @@ describe('isAiConfiguredForAsk', () => {
     expect(
       isAiConfiguredForAsk({
         ai: {
-          provider: 'openai',
-          providers: { anthropic: { apiKey: '' }, openai: { apiKey: '' } },
+          activeProvider: 'p_openai01',
+          providers: [
+            { id: 'p_anthro01', name: 'Anthropic', protocol: 'anthropic', apiKey: '', model: 'claude-sonnet-4-6', baseUrl: '' },
+            { id: 'p_openai01', name: 'OpenAI', protocol: 'openai', apiKey: '', model: 'gpt-5.4', baseUrl: '' },
+          ],
         },
         envOverrides: { OPENAI_API_KEY: true },
       }),
     ).toBe(true);
   });
 
-  it('treats missing provider as anthropic (error path)', () => {
+  it('treats missing provider as first entry fallback (error path)', () => {
     expect(
       isAiConfiguredForAsk({
-        ai: { providers: { anthropic: { apiKey: '' }, openai: { apiKey: '' } } },
+        ai: {
+          providers: [
+            { id: 'p_anthro01', name: 'Anthropic', protocol: 'anthropic', apiKey: '', model: 'claude-sonnet-4-6', baseUrl: '' },
+            { id: 'p_openai01', name: 'OpenAI', protocol: 'openai', apiKey: '', model: 'gpt-5.4', baseUrl: '' },
+          ],
+        },
         envOverrides: {},
       }),
     ).toBe(false);

package/_standalone/__tests__/setup.ts

@@ -23,11 +23,11 @@ export function getTestMindRoot() {
 vi.mock('@/lib/settings', () => ({
   readSettings: () => ({
     ai: {
-      provider: 'anthropic' as const,
-      providers: {
-        anthropic: { apiKey: '', model: 'claude-sonnet-4-6' },
-        openai:    { apiKey: '', model: 'gpt-5.4', baseUrl: '' },
-      },
+      activeProvider: 'p_anthro01',
+      providers: [
+        { id: 'p_anthro01', name: 'Anthropic', protocol: 'anthropic', apiKey: '', model: 'claude-sonnet-4-6', baseUrl: '' },
+        { id: 'p_openai01', name: 'OpenAI', protocol: 'openai', apiKey: '', model: 'gpt-5.4', baseUrl: '' },
+      ],
     },
     mindRoot: '',
   }),

package/_standalone/app/globals.css

@@ -70,7 +70,7 @@ body {
   --muted-foreground: #685f52;
   --accent: #d9d3c6;
   --accent-foreground: #1c1a17;
-  --destructive: oklch(0.56 0.14 24);
+  --destructive: #b33b3b;
   --destructive-foreground: #ffffff;
   --border: rgba(28, 26, 23, 0.1);
   --input: rgba(28, 26, 23, 0.12);
@@ -108,7 +108,7 @@ body {
   --muted-foreground: #8a8275;
   --accent: #2e2b22;
   --accent-foreground: #e8e4dc;
-  --destructive: oklch(0.56 0.14 22);
+  --destructive: #b33636;
   --destructive-foreground: #ffffff;
   --border: rgba(232, 228, 220, 0.08);
   --input: rgba(232, 228, 220, 0.1);
@@ -320,8 +320,8 @@ body {
 
 .modal-backdrop {
   background: rgba(10, 9, 6, 0.72);
-  backdrop-filter: blur(8px);
   -webkit-backdrop-filter: blur(8px);
+  backdrop-filter: blur(8px);
 }
 .dark .modal-backdrop {
   background: rgba(0, 0, 0, 0.65);
@@ -329,8 +329,8 @@ body {
 
 .overlay-backdrop {
   background: rgba(10, 9, 6, 0.35);
-  backdrop-filter: blur(2px);
   -webkit-backdrop-filter: blur(2px);
+  backdrop-filter: blur(2px);
 }
 .dark .overlay-backdrop {
   background: rgba(0, 0, 0, 0.4);