@stilero/bankan 1.0.13 → 1.0.17

package/server/src/orchestrator.test.js

@@ -0,0 +1,431 @@
+import { describe, expect, test, vi } from 'vitest';
+
+import {
+  buildAgentCommand,
+  cleanTerminalArtifacts,
+  extractPlannerPlanText,
+  extractReviewerReviewText,
+  sanitizeBranchName,
+} from './orchestrator.js';
+
+describe('structured output extraction', () => {
+  test('planner extraction falls back to agent structured capture when the PTY tail lost the full block', () => {
+    const readCaptured = vi.fn(() => null);
+    const agent = {
+      cli: 'claude',
+      getBufferString: vi.fn(() => '...tail...\n=== PLAN END ==='),
+      getStructuredBlock: vi.fn(() => `=== PLAN START ===
+SUMMARY: Persist completed planner output.
+BRANCH: feature/test-plan
+FILES_TO_MODIFY:
+- server/src/agents.js (capture plan output)
+STEPS:
+1. Read from stable structured state.
+TESTS_NEEDED:
+- Run npm run test:server
+RISKS:
+- none
+=== PLAN END ===`),
+    };
+
+    expect(extractPlannerPlanText(agent, { readCapturedCodexMessage: readCaptured })).toContain(
+      'Persist completed planner output.'
+    );
+    expect(readCaptured).not.toHaveBeenCalled();
+  });
+
+  test('review extraction prefers Codex captured output before agent structured capture', () => {
+    const readCaptured = vi.fn(() => `=== REVIEW START ===
+VERDICT: PASS
+CRITICAL_ISSUES:
+- none
+MINOR_ISSUES:
+- none
+SUMMARY: Captured Codex output should win.
+=== REVIEW END ===`);
+    const agent = {
+      cli: 'codex',
+      getBufferString: vi.fn(() => '=== CODEX_LAST_MESSAGE_FILE:/tmp/test ==='),
+      getStructuredBlock: vi.fn(() => `=== REVIEW START ===
+VERDICT: PASS
+CRITICAL_ISSUES:
+- none
+MINOR_ISSUES:
+- none
+SUMMARY: Agent fallback should not be used here.
+=== REVIEW END ===`),
+    };
+
+    expect(extractReviewerReviewText(agent, { readCapturedCodexMessage: readCaptured })).toContain(
+      'Captured Codex output should win.'
+    );
+    expect(readCaptured).toHaveBeenCalledOnce();
+  });
+
+  test('planner extraction finds real plan via getAllCapturedBlocks when buffer is exhausted', () => {
+    const readCaptured = vi.fn(() => null);
+    const realPlan = `=== PLAN START ===
+SUMMARY: Real plan found via captured blocks.
+BRANCH: feature/test-captured
+FILES_TO_MODIFY:
+- server/src/agents.js (fix capture)
+STEPS:
+1. Store all captured blocks.
+TESTS_NEEDED:
+- Run npm run test:server
+RISKS:
+- none
+=== PLAN END ===`;
+    const templateBlock = `=== PLAN START ===
+SUMMARY: (one sentence describing what will be built)
+BRANCH: (feature/t-xxx-short-descriptive-slug)
+FILES_TO_MODIFY:
+- path/to/file.ts (reason for modification)
+STEPS:
+1. (detailed, actionable step)
+TESTS_NEEDED:
+- (test description, or 'none')
+RISKS:
+- (potential issue or edge case, or 'none')
+=== PLAN END ===`;
+    const agent = {
+      cli: 'claude',
+      // Buffer exhausted — only noise, no plan markers
+      getBufferString: vi.fn(() => 'lots of noise without any plan markers'),
+      // Structured capture has the placeholder (overwritten real plan)
+      getStructuredBlock: vi.fn(() => templateBlock),
+      // But getAllCapturedBlocks has the full history
+      getAllCapturedBlocks: vi.fn(() => [templateBlock, realPlan, templateBlock]),
+    };
+
+    const result = extractPlannerPlanText(agent, { readCapturedCodexMessage: readCaptured });
+    expect(result).toContain('Real plan found via captured blocks.');
+    expect(result).not.toContain('(one sentence describing');
+  });
+
+  test('planner extraction falls back to buffer scan when getStructuredBlock returns null', () => {
+    const readCaptured = vi.fn(() => null);
+    const planBlock = `=== PLAN START ===
+SUMMARY: Buffer scan fallback works.
+BRANCH: feature/buffer-fallback
+FILES_TO_MODIFY:
+- server/src/orchestrator.js (add fallback)
+STEPS:
+1. Try structured capture first, then scan buffer.
+TESTS_NEEDED:
+- Run npm run test:server
+RISKS:
+- none
+=== PLAN END ===`;
+    const agent = {
+      cli: 'claude',
+      getBufferString: vi.fn(() => `some noise\n${planBlock}\nmore noise`),
+      getStructuredBlock: vi.fn(() => null),
+    };
+
+    const result = extractPlannerPlanText(agent, { readCapturedCodexMessage: readCaptured });
+    expect(result).toContain('Buffer scan fallback works.');
+    expect(result).toContain('=== PLAN START ===');
+    expect(result).toContain('=== PLAN END ===');
+    expect(agent.getStructuredBlock).toHaveBeenCalledWith('plan');
+  });
+
+  test('planner extraction skips placeholder blocks to find real plan in buffer', () => {
+    // The CLI echoes the prompt template (1st block), agent outputs real plan (2nd),
+    // then CLI re-renders the template in the status area (3rd). The last block is a
+    // placeholder, but the 2nd block has real content.
+    const templateBlock = `=== PLAN START ===
+SUMMARY: (one sentence describing what will be built)
+BRANCH: (feature/t-xxx-short-descriptive-slug)
+FILES_TO_MODIFY:
+- path/to/file.ts (reason for modification)
+STEPS:
+1. (detailed, actionable step)
+TESTS_NEEDED:
+- (test description, or 'none')
+RISKS:
+- (potential issue or edge case, or 'none')
+=== PLAN END ===`;
+
+    const realBlock = `=== PLAN START ===
+SUMMARY: Add a Reports modal accessible from the top bar with per-repo task counts and token stats.
+BRANCH: feature/t-91eadd-reports-dashboard
+FILES_TO_MODIFY:
+- client/src/ReportsModal.jsx (new reporting modal)
+- server/src/index.js (add reports endpoint)
+STEPS:
+1. Create ReportsModal component.
+2. Wire up the top-bar button.
+TESTS_NEEDED:
+- Run npm run test
+RISKS:
+- none
+=== PLAN END ===`;
+
+    const readCaptured = vi.fn(() => null);
+    const bufferContent = `noise\n${templateBlock}\nmore noise\n${realBlock}\neven more\n${templateBlock}\ntrailing`;
+    const agent = {
+      cli: 'claude',
+      getBufferString: vi.fn(() => bufferContent),
+      // Structured capture got the template (last completed block)
+      getStructuredBlock: vi.fn(() => templateBlock),
+    };
+
+    const result = extractPlannerPlanText(agent, { readCapturedCodexMessage: readCaptured });
+    expect(result).toContain('Reports modal accessible from the top bar');
+    expect(result).not.toContain('(one sentence describing what will be built)');
+  });
+
+  test('review extraction finds real review via getAllCapturedBlocks when template overwrites capture', () => {
+    const readCaptured = vi.fn(() => null);
+    const realReview = `=== REVIEW START ===
+VERDICT: PASS
+CRITICAL_ISSUES:
+- none
+MINOR_ISSUES:
+- Minor typo in variable name
+SUMMARY: All changes look good. Tests pass and code follows conventions.
+=== REVIEW END ===`;
+    const templateBlock = `=== REVIEW START ===
+VERDICT: PASS or FAIL
+CRITICAL_ISSUES:
+- concrete issue, or none
+MINOR_ISSUES:
+- concrete issue, or none
+SUMMARY: 2-3 concrete sentences summarising the review, including changed files and strengths
+=== REVIEW END ===`;
+    const agent = {
+      cli: 'claude',
+      // Buffer exhausted — no review markers
+      getBufferString: vi.fn(() => 'noise without markers'),
+      // Structured capture has the placeholder (overwritten real review)
+      getStructuredBlock: vi.fn(() => templateBlock),
+      // But getAllCapturedBlocks has the full history
+      getAllCapturedBlocks: vi.fn(() => [templateBlock, realReview, templateBlock]),
+    };
+
+    const result = extractReviewerReviewText(agent, { readCapturedCodexMessage: readCaptured });
+    expect(result).toContain('All changes look good.');
+    expect(result).not.toContain('2-3 concrete sentences');
+  });
+
+  test('review extraction falls back to agent structured capture when the live tail only contains the end marker', () => {
+    const readCaptured = vi.fn(() => null);
+    const agent = {
+      cli: 'claude',
+      getBufferString: vi.fn(() => '...tail...\n=== REVIEW END ==='),
+      getStructuredBlock: vi.fn(() => `=== REVIEW START ===
+VERDICT: PASS
+CRITICAL_ISSUES:
+- none
+MINOR_ISSUES:
+- none
+SUMMARY: Stable review capture prevents timeout.
+=== REVIEW END ===`),
+    };
+
+    expect(extractReviewerReviewText(agent, { readCapturedCodexMessage: readCaptured })).toContain(
+      'Stable review capture prevents timeout.'
+    );
+  });
+});
+
+describe('sanitizeBranchName', () => {
+  test('strips garbage text appended by ANSI cursor collapse', () => {
+    expect(sanitizeBranchName('feature/t-a811ca-reporting FILES_TO_MODIFY:'))
+      .toBe('feature/t-a811ca-reporting');
+  });
+
+  test('strips trailing prompt characters and whitespace', () => {
+    expect(sanitizeBranchName('feature/t-b60f78-repo-reports  ❯'))
+      .toBe('feature/t-b60f78-repo-reports');
+  });
+
+  test('preserves clean branch names unchanged', () => {
+    expect(sanitizeBranchName('feature/t-91eadd-reports-dashboard'))
+      .toBe('feature/t-91eadd-reports-dashboard');
+  });
+
+  test('handles branch names with dots and underscores', () => {
+    expect(sanitizeBranchName('fix/v2.1_hotfix'))
+      .toBe('fix/v2.1_hotfix');
+  });
+
+  test('strips trailing dots from branch names', () => {
+    expect(sanitizeBranchName('feature/test.'))
+      .toBe('feature/test');
+  });
+});
+
+describe('buildAgentCommand model flag', () => {
+  test('claude CLI includes --model flag when model is non-empty', () => {
+    const cmd = buildAgentCommand('claude', 'do stuff', 'plan', 'haiku');
+    expect(cmd).toContain('--model haiku');
+    expect(cmd).toContain('--dangerously-skip-permissions');
+  });
+
+  test('claude CLI omits --model flag when model is empty', () => {
+    const cmd = buildAgentCommand('claude', 'do stuff', 'plan', '');
+    expect(cmd).not.toContain('--model');
+  });
+
+  test('claude CLI omits --model flag when model is not provided', () => {
+    const cmd = buildAgentCommand('claude', 'do stuff', 'plan');
+    expect(cmd).not.toContain('--model');
+  });
+
+  test('codex CLI includes -m flag when model is non-empty', () => {
+    const cmd = buildAgentCommand('codex', 'do stuff', 'plan', 'opus');
+    expect(cmd).toContain('-m opus');
+    expect(cmd).toContain('codex exec');
+  });
+
+  test('codex CLI omits -m flag when model is empty', () => {
+    const cmd = buildAgentCommand('codex', 'do stuff', 'interactive', '');
+    expect(cmd).not.toContain('-m ');
+  });
+
+  test('claude print mode includes --model flag', () => {
+    const cmd = buildAgentCommand('claude', 'do stuff', 'print', 'sonnet');
+    expect(cmd).toContain('--model sonnet');
+    expect(cmd).toContain('--print');
+  });
+});
+
+describe('cleanTerminalArtifacts', () => {
+  test('removes CLI status bar, permission toggle, box drawings, and header lines', () => {
+    const dirty = `=== PLAN START ===
+Opus4.6(1Mcontext) │T-91EADD ░░░░░░░░░░6%
+⏵⏵bypasspermissionson (shift+tabtocycle)
+SUMMARY: Add a Reports modal with per-repo task counts.
+⏵⏵bypasspermissionson (shift+tabtocycle)
+BRANCH: feature/t-91eadd-reports-dashboard
+────────────────────────────────────────────────────────
+Opus4.6(1Mcontext) │T-91EADD ░░░░░░░░░░6%
+⏵⏵bypasspermissionson (shift+tabtocycle)
+FILES_TO_MODIFY:
+- client/src/ReportsModal.jsx (new modal component)
+ ▐▛███▜▌ClaudeCodev2.1.76
+▝▜█████▛▘Opus4.6(1Mcontext)·ClaudeMax
+ ~/Developer/stilero/bankan/.data/workspaces/T-91EADD
+STEPS:
+1. Create the ReportsModal component.
+❯
+TESTS_NEEDED:
+- Run npm run test
+RISKS:
+- none
+=== PLAN END ===`;
+
+    const cleaned = cleanTerminalArtifacts(dirty);
+    expect(cleaned).toContain('SUMMARY: Add a Reports modal');
+    expect(cleaned).toContain('BRANCH: feature/t-91eadd-reports-dashboard');
+    expect(cleaned).toContain('- client/src/ReportsModal.jsx');
+    expect(cleaned).toContain('1. Create the ReportsModal component.');
+    expect(cleaned).not.toContain('Opus4.6');
+    expect(cleaned).not.toContain('bypasspermission');
+    expect(cleaned).not.toContain('────');
+    expect(cleaned).not.toContain('▐▛███');
+    expect(cleaned).not.toContain('ClaudeCode');
+    expect(cleaned).not.toContain('ClaudeMax');
+    expect(cleaned).not.toContain('.data/workspaces');
+    expect(cleaned).not.toContain('❯');
+  });
+
+  test('strips trailing artifacts from content lines', () => {
+    // The terminal can put artifacts on the same line as real content
+    const dirty = '=== PLAN START ===                     ❯  ──────────────────────────────────\nSUMMARY: Real plan.\n=== PLAN END ===';
+    const cleaned = cleanTerminalArtifacts(dirty);
+    expect(cleaned).toContain('=== PLAN START ===');
+    expect(cleaned).toContain('SUMMARY: Real plan.');
+    expect(cleaned).not.toContain('❯');
+    expect(cleaned).not.toContain('────');
+  });
+
+  test('preserves clean plan text unchanged', () => {
+    const clean = `=== PLAN START ===
+SUMMARY: Add automated tests.
+BRANCH: feature/add-tests
+FILES_TO_MODIFY:
+- server/src/workflow.test.js (expand coverage)
+STEPS:
+1. Add tests for retry status edge cases.
+TESTS_NEEDED:
+- Run npm run test:server
+RISKS:
+- none
+=== PLAN END ===`;
+
+    expect(cleanTerminalArtifacts(clean)).toBe(clean);
+  });
+
+  test('removes embedded prompt echo and template block from captured plan', () => {
+    // When the real plan's END marker is lost in ANSI rendering, the extraction
+    // grabs from the first START to the template's END, including echoed prompt text.
+    const dirty = `=== PLAN START ===
+SUMMARY: Add a Reports modal from the top bar.
+BRANCH: feature/t-b60f78-repo-reports
+FILES_TO_MODIFY:
+- client/src/App.jsx (add Reports button)
+STEPS:
+1. Create ReportsModal component
+TESTS_NEEDED:
+- Run npm run test
+RISKS:
+- none
+
+Message from org:
+Make sure to update CLAUDE.md
+
+❯ You are a senior software architect. A task has been assigned to you.
+Repository: https://github.com/stilero/bankan
+TASK ID: T-B60F78
+TITLE: Reports
+
+Plan Mode Instructions
+- Do not edit files
+Output ONLY in this exact format:
+
+=== PLAN START ===
+SUMMARY: (one sentence describing what will be built)
+BRANCH: (feature/t-b60f78-short-descriptive-slug)
+FILES_TO_MODIFY:
+- path/to/file.ts (reason for modification)
+STEPS:
+1. (detailed, actionable step)
+TESTS_NEEDED:
+- (test description, or 'none')
+RISKS:
+- (potential issue or edge case, or 'none')
+=== PLAN END ===`;
+
+    const cleaned = cleanTerminalArtifacts(dirty);
+    expect(cleaned).toContain('SUMMARY: Add a Reports modal');
+    expect(cleaned).toContain('BRANCH: feature/t-b60f78-repo-reports');
+    expect(cleaned).toContain('client/src/App.jsx');
+    // Should NOT contain the echoed prompt template
+    expect(cleaned).not.toContain('(one sentence describing');
+    expect(cleaned).not.toContain('You are a senior software architect');
+    expect(cleaned).not.toContain('Plan Mode Instructions');
+    expect(cleaned).not.toContain('Message from org');
+  });
+
+  test('removes inline prompt character from content lines', () => {
+    const dirty = `=== PLAN START ===
+SUMMARY: Fix the bug.
+BRANCH: feature/fix  ❯
+FILES_TO_MODIFY:
+- file.js (fix)
+STEPS:
+1. Fix it
+TESTS_NEEDED:
+- none
+RISKS:
+- none
+=== PLAN END ===`;
+
+    const cleaned = cleanTerminalArtifacts(dirty);
+    expect(cleaned).toContain('BRANCH: feature/fix');
+    expect(cleaned).not.toContain('❯');
+  });
+});

package/server/src/paths.test.js

@@ -0,0 +1,49 @@
+import { afterEach, describe, expect, test, vi } from 'vitest';
+
+const previousMode = process.env.BANKAN_RUNTIME_MODE;
+const previousHome = process.env.BANKAN_HOME;
+
+afterEach(() => {
+  if (previousMode === undefined) {
+    delete process.env.BANKAN_RUNTIME_MODE;
+  } else {
+    process.env.BANKAN_RUNTIME_MODE = previousMode;
+  }
+
+  if (previousHome === undefined) {
+    delete process.env.BANKAN_HOME;
+  } else {
+    process.env.BANKAN_HOME = previousHome;
+  }
+
+  vi.resetModules();
+});
+
+describe('runtime path resolution', () => {
+  test('uses repository .data in development mode', async () => {
+    delete process.env.BANKAN_RUNTIME_MODE;
+    delete process.env.BANKAN_HOME;
+    vi.resetModules();
+
+    const { getRuntimePaths } = await import('./paths.js');
+    const paths = getRuntimePaths();
+
+    expect(paths.packaged).toBe(false);
+    expect(paths.dataDir.endsWith('/.data')).toBe(true);
+    expect(paths.envFile.endsWith('/.env.local')).toBe(true);
+  });
+
+  test('uses BANKAN_HOME in packaged mode', async () => {
+    process.env.BANKAN_RUNTIME_MODE = 'packaged';
+    process.env.BANKAN_HOME = '/tmp/bankan-home';
+    vi.resetModules();
+
+    const { getAppDataDir, getEnvFilePath, getRuntimePaths } = await import('./paths.js');
+    const paths = getRuntimePaths();
+
+    expect(paths.packaged).toBe(true);
+    expect(getAppDataDir()).toBe('/tmp/bankan-home');
+    expect(getEnvFilePath()).toBe('/tmp/bankan-home/.env.local');
+    expect(paths.bridgesDir.endsWith('/bankan/terminal-bridges')).toBe(true);
+  });
+});

package/server/src/sessionHistory.test.js

@@ -0,0 +1,39 @@
+import { describe, expect, test } from 'vitest';
+
+import { createSessionEntry, getAgentStage } from './sessionHistory.js';
+
+describe('session history helpers', () => {
+  test('maps agent ids to stages', () => {
+    expect(getAgentStage('plan-1')).toBe('planning');
+    expect(getAgentStage('imp-1')).toBe('implementation');
+    expect(getAgentStage('rev-1')).toBe('review');
+    expect(getAgentStage('orch')).toBe('unknown');
+  });
+
+  test('creates session entries with stable shape', () => {
+    const entry = createSessionEntry({
+      id: 'imp-1',
+      name: 'Implementor 1',
+      role: 'Code Generation',
+      tokens: 320,
+    }, {
+      taskId: 'T-123',
+      outcome: 'blocked',
+      transcript: 'Need input',
+      finishedAt: '2026-03-15T12:00:00.000Z',
+    });
+
+    expect(entry).toEqual({
+      id: 'imp-1:2026-03-15T12:00:00.000Z',
+      agentId: 'imp-1',
+      agentName: 'Implementor 1',
+      role: 'Code Generation',
+      stage: 'implementation',
+      taskId: 'T-123',
+      outcome: 'blocked',
+      finishedAt: '2026-03-15T12:00:00.000Z',
+      transcript: 'Need input',
+      tokens: 320,
+    });
+  });
+});

package/server/src/store.js

@@ -212,10 +212,9 @@ class TaskStore {
     if (!task) return null;
     if (typeof totalTokens !== 'number' || totalTokens < task.totalTokens) return task;
     task.totalTokens = totalTokens;
-    task.updatedAt = new Date().toISOString();
+    // Intentionally skip updatedAt and tasks:changed — token counts are
+    // telemetry and should not affect sort order or trigger full re-renders.
     this._save();
-    bus.emit('task:updated', task);
-    bus.emit('tasks:changed', this.tasks);
     return task;
   }