test: audit and fix 18 test quality issues across 15 files

Critical fixes:
- wsl-utils: use vi.resetModules for fresh import, error path now tested
- workflow-run-store: simplify prune test, assert exactly 20 remain
- workflow-store: remove dead testStore.set with matcher-as-key
- log-adapters: fresh CodexAdapter per test to prevent state bleed
- cost-tracker: assert IPC send instead of weak call count

Weak/trivial tests removed:
- workflow-engine: delete constant equality + type-only continueOnError tests
- skill-scanner: replace no-op cache tests with behavioral verification
- project-store: assert decrypted value, not just function call
- appStore: replace toBeTruthy with type+format check on session IDs

Redundancies consolidated:
- agents: remove 3 individual mapping tests (loop test covers all)
- node-runners-skill: consolidate 3 identical tests into it.each

Edge cases added:
- pty-manager: multi-line data chunk for activity line buffer
- safeFitAndResize: single-zero-dimension cases (cols=0, rows=0)
- workflow-engine-execution: workflow-level event on idle timeout

Docs updated:
- README: add cost tracking + worktree isolation features, update test count
- USER-GUIDE: add cost tracking + worktree isolation sections, bump to v4.8.0

Net: 614 → 610 tests (removed 7 bad, added 3 meaningful)

Co-Authored-By: Rooty

Author: Wintersta7e

README.md

@@ -6,7 +6,7 @@ A desktop terminal manager for WSL AI coding agents. Launch, manage, and orchest
 ![React](https://img.shields.io/badge/React-19-61DAFB?logo=react)
 ![TypeScript](https://img.shields.io/badge/TypeScript-5-3178C6?logo=typescript)
 ![License: Elastic-2.0](https://img.shields.io/badge/License-Elastic--2.0-blue.svg)
-![Tests](https://img.shields.io/badge/Tests-499_passing-brightgreen)
+![Tests](https://img.shields.io/badge/Tests-610_passing-brightgreen)
 ![CI](https://github.com/Wintersta7e/agentdeck/actions/workflows/ci.yml/badge.svg?branch=main)
 
 ## Why AgentDeck?
@@ -28,6 +28,8 @@ AgentDeck provides a unified desktop environment for working with AI coding agen
 - **Project Management** - Configure projects with paths, agents, prompt templates, and stack badges
 - **Split Terminal Views** - 1/2/3 pane layouts with independent sessions
 - **Agentic Workflows** - Visual node-graph editor with conditions, loops, variables, and execution history
+- **Cost/Token Tracking** - Live per-session cost and token usage for Claude Code and Codex
+- **Git Worktree Isolation** - Per-session git branches with Keep/Discard review flow
 - **8 Themes** - 4 dark + 4 light themes with smooth view transitions
 - **Command Palette** - Quick access to projects, sessions, templates, and tools
 
@@ -76,6 +78,21 @@ AgentDeck provides a unified desktop environment for working with AI coding agen
 - **Agent Visibility** - Toggle which agents appear on the home screen
 - **Keyboard Shortcuts** - Full shortcut reference (Ctrl+/)
 
+### Cost/Token Tracking
+
+- **Live Cost Badge** - Per-session USD cost and token count in the pane topbar
+- **Claude Code Support** - Parses JSONL session logs with cache-aware pricing (write 1.25x, read 0.1x)
+- **Codex CLI Support** - Parses JSONL rollout logs with per-model pricing maps
+- **Tooltip Breakdown** - Hover for input, output, cache read, and cache write token counts
+- **Automatic Discovery** - Finds active log files via WSL polling, no configuration needed
+
+### Git Worktree Isolation
+
+- **Per-Session Branches** - Each agent session gets its own git worktree and branch
+- **Branch Badge** - Active branch shown in pane topbar
+- **Review Flow** - Inspect changes on close, then Keep (merge) / Discard / Cancel
+- **Automatic Cleanup** - Orphaned worktrees pruned on startup
+
 ### Agent Updates
 
 - **Version Checking** - Startup notification when agent updates are available
@@ -133,7 +150,7 @@ npm run dev
 # Build for production (validates TypeScript)
 npm run build
 
-# Run tests (499 tests)
+# Run tests (610 tests)
 npm test
 
 # Lint code (zero-warning policy)
@@ -158,13 +175,17 @@ Output: `dist/AgentDeck-{version}-portable.exe` (~89 MB)
 src/
 ├── main/                    # Electron main process
 │   ├── index.ts             # App lifecycle, IPC handler registration
-│   ├── ipc/                 # 6 IPC modules (pty, window, agents, projects, workflows, utils)
+│   ├── ipc/                 # 8 IPC modules (pty, window, agents, projects, workflows, skills, worktree, utils)
 │   ├── pty-manager.ts       # node-pty: spawn, resize, kill, activity parsing
 │   ├── workflow-engine.ts   # Edge-activation scheduler DAG execution
 │   ├── edge-scheduler.ts    # Pure scheduler: ready queue, branching, skip, loop reset
 │   ├── variable-substitution.ts # {{VAR}} replacement in workflow nodes
 │   ├── workflow-run-store.ts # Execution history persistence
 │   ├── agent-updater.ts     # Agent version checking and updating via WSL
+│   ├── log-adapters.ts      # Claude + Codex JSONL cost/token parsing
+│   ├── cost-tracker.ts      # Log file discovery, tailing, and IPC push
+│   ├── git-port.ts          # Git command abstraction (WSL)
+│   ├── worktree-manager.ts  # Per-session git worktree lifecycle
 │   └── project-store.ts     # electron-store: CRUD + safeStorage for API keys
 ├── preload/
 │   └── index.ts             # contextBridge: safe IPC surface (window.agentDeck)
@@ -193,7 +214,7 @@ src/
 | [node-pty](https://github.com/microsoft/node-pty) | Pseudo-terminal (WSL sessions) |
 | [Zustand](https://zustand-demo.pmnd.rs) | State management |
 | [React Flow](https://reactflow.dev) | Visual workflow node editor |
-| [Vitest 4](https://vitest.dev) | Testing framework (499 tests) |
+| [Vitest 4](https://vitest.dev) | Testing framework (610 tests) |
 | [ESLint 9](https://eslint.org) | Linting (flat config, zero-warning policy) |
 
 ## Documentation

docs/USER-GUIDE.md

@@ -1,6 +1,6 @@
 # AgentDeck User Guide
 
-> **Version**: 4.7.0
+> **Version**: 4.8.0
 
 AgentDeck is a desktop command center for managing AI coding agents through WSL2 terminals. This guide covers every feature from first launch to advanced workflow automation.
 
@@ -21,10 +21,12 @@ AgentDeck is a desktop command center for managing AI coding agents through WSL2
 11. [Agentic Workflows](#agentic-workflows) (conditions, loops, variables, import/export, history)
 12. [Workflow Roles](#workflow-roles)
 13. [Command Palette](#command-palette)
-14. [Agent Updates](#agent-updates)
-15. [Themes](#themes)
-16. [Keyboard Shortcuts](#keyboard-shortcuts)
-17. [Troubleshooting](#troubleshooting)
+14. [Cost/Token Tracking](#costtoken-tracking)
+15. [Git Worktree Isolation](#git-worktree-isolation)
+16. [Agent Updates](#agent-updates)
+17. [Themes](#themes)
+18. [Keyboard Shortcuts](#keyboard-shortcuts)
+19. [Troubleshooting](#troubleshooting)
 
 ---
 
@@ -541,6 +543,67 @@ In the Themes sub-menu, arrow keys give a live preview of each theme. Press Ente
 
 ---
 
+## Cost/Token Tracking
+
+AgentDeck tracks token usage and estimated cost for **Claude Code** and **Codex CLI** sessions in real-time.
+
+### How It Works
+
+When a session starts, AgentDeck discovers the agent's JSONL log file in WSL and tails it every 3 seconds. Token usage and cost are parsed from the log entries and pushed to the UI.
+
+- **Claude Code**: Reads `~/.claude/projects/` session logs. Pricing uses the model ID (opus/sonnet/haiku) with cache-aware rates — cache writes cost 1.25x, cache reads cost 0.1x of the base input rate.
+- **Codex CLI**: Reads `~/.codex/sessions/` rollout files. Pricing uses per-model maps (gpt-4o, o3, gpt-5.3, etc.).
+
+### The Cost Badge
+
+A **Zap icon** appears in the pane topbar showing:
+- **USD cost** (e.g. `$0.18`) — computed from model pricing
+- **Total tokens** (e.g. `28.8k tokens`) — all tokens processed (input + cache + output)
+
+**Hover** the badge for a tooltip breakdown: input, output, cache read, and cache write tokens.
+
+The cost and token count are always consistent — both reflect the same set of tokens, so the numbers make sense together.
+
+### Notes
+
+- Cost tracking is automatic — no configuration needed
+- Only Claude Code and Codex are supported (other agents show no badge)
+- On the first turn, Claude shows high token counts due to system prompt caching — this drops dramatically on subsequent turns
+- Cost data is per-session and resets when the session closes
+
+---
+
+## Git Worktree Isolation
+
+Each agent session can run in its own **git worktree** — an isolated copy of the repository on a separate branch. This prevents agents from interfering with each other or with your working directory.
+
+### How It Works
+
+When you open a session for a project that is a git repository, AgentDeck automatically creates a worktree:
+- A new branch is created (e.g. `agentdeck/session-abc`)
+- The agent works in the worktree directory, not your main working copy
+- Changes are isolated until you decide to keep or discard them
+
+### Branch Badge
+
+When a session has an isolated worktree, a **branch badge** appears in the pane topbar showing the branch name. The status bar also shows a worktree indicator.
+
+### Closing a Session
+
+When you close a session with an active worktree, AgentDeck inspects the worktree for changes:
+
+- **No changes**: Worktree is silently cleaned up
+- **Has changes**: A dialog appears with three options:
+  - **Keep** — Preserve the branch and worktree for later merging
+  - **Discard** — Delete the worktree and branch (changes are lost)
+  - **Cancel** — Keep the session open
+
+### Cleanup
+
+Orphaned worktrees (from crashes or abrupt exits) are automatically pruned on startup.
+
+---
+
 ## Agent Updates
 
 AgentDeck checks for agent updates on startup and shows a toast notification when updates are available.

src/main/__tests__/node-runners-skill.test.ts

@@ -16,17 +16,12 @@ describe('skill prefix extraction', () => {
     expect(extractSkillPrefix('project:deploy', 'codex')).toBe('$deploy ')
   })
 
-  it('returns null for non-codex agent (claude-code)', () => {
-    expect(extractSkillPrefix('global:lint-fix', 'claude-code')).toBeNull()
-  })
-
-  it('returns null for non-codex agent (aider)', () => {
-    expect(extractSkillPrefix('global:lint-fix', 'aider')).toBeNull()
-  })
-
-  it('returns null for non-codex agent (goose)', () => {
-    expect(extractSkillPrefix('global:lint-fix', 'goose')).toBeNull()
-  })
+  it.each(['claude-code', 'aider', 'goose'] as const)(
+    'returns null for non-codex agent (%s)',
+    (agent) => {
+      expect(extractSkillPrefix('global:lint-fix', agent)).toBeNull()
+    },
+  )
 
   it('returns null when skillId is undefined', () => {
     expect(extractSkillPrefix(undefined, 'codex')).toBeNull()

src/main/__tests__/skill-scanner.test.ts

@@ -5,6 +5,7 @@ import {
   parseFrontmatter,
   invalidateAllCaches,
   scanSkillDirectory,
+  getProjectSkills,
 } from '../skill-scanner'
 
 vi.mock('child_process', () => ({
@@ -259,16 +260,32 @@ describe('parseFrontmatter', () => {
 // ── Cache helpers ──────────────────────────────────────────────────
 
 describe('invalidateAllCaches', () => {
-  it('resets state without errors', () => {
-    // Should not throw even when caches are already empty
-    expect(() => invalidateAllCaches()).not.toThrow()
-  })
+  it('forces a fresh WSL call after invalidation', async () => {
+    const output = [
+      '---SKILL-BLOCK---',
+      '/home/user/project/.agents/skills/deploy/SKILL.md',
+      'deploy',
+      '---',
+      'name: deploy',
+      'description: Deploy to prod',
+      '---',
+    ].join('\n')
 
-  it('can be called multiple times', () => {
-    invalidateAllCaches()
-    invalidateAllCaches()
+    mockWslOutput(output)
+
+    // First call populates the project cache
+    await getProjectSkills('/home/user/project', 'Ubuntu')
+    const callsAfterFirst = vi.mocked(cp.execFile).mock.calls.length
+
+    // Second call should hit the cache — no new execFile call
+    await getProjectSkills('/home/user/project', 'Ubuntu')
+    expect(vi.mocked(cp.execFile).mock.calls.length).toBe(callsAfterFirst)
+
+    // Invalidate and call again — must trigger a fresh WSL call
     invalidateAllCaches()
-    // No error means success
+    mockWslOutput(output)
+    await getProjectSkills('/home/user/project', 'Ubuntu')
+    expect(vi.mocked(cp.execFile).mock.calls.length).toBeGreaterThan(callsAfterFirst)
   })
 })
 

src/main/cost-tracker.test.ts

@@ -211,7 +211,10 @@ describe('file discovery', () => {
     // Tailing poll at 3s after discovery
     await vi.advanceTimersByTimeAsync(3000)
 
-    expect(callCount).toBeGreaterThanOrEqual(3)
+    expect(win.webContents.send).toHaveBeenCalledWith(
+      'cost:update',
+      expect.objectContaining({ sessionId: 's1' }),
+    )
 
     tracker.destroy()
   })

src/main/log-adapters.test.ts

@@ -1,4 +1,4 @@
-import { describe, it, expect } from 'vitest'
+import { describe, it, expect, beforeEach } from 'vitest'
 import {
   formatTokens,
   formatCost,
@@ -83,9 +83,7 @@ describe('ClaudeAdapter', () => {
     const dirs = adapter.getLogDirs('/home/rooty/my-project')
     const second = dirs[1]
     if (!second) throw new Error('Expected second dir')
-    expect(second).toContain('~/.claude/projects/')
-    // fallback dir should not contain the path slug
-    expect(second).not.toContain('rooty')
+    expect(second).toBe('~/.claude/projects/')
   })
 
   it('getFilePattern returns "*.jsonl"', () => {
@@ -219,7 +217,11 @@ describe('ClaudeAdapter', () => {
 // ---------------------------------------------------------------------------
 
 describe('CodexAdapter', () => {
-  const adapter = createCodexAdapter()
+  let adapter: ReturnType<typeof createCodexAdapter>
+
+  beforeEach(() => {
+    adapter = createCodexAdapter()
+  })
 
   it('has agent "codex"', () => {
     expect(adapter.agent).toBe('codex')

src/main/project-store.test.ts

@@ -150,6 +150,7 @@ describe('createProjectStore', () => {
     expect(safeStorage.decryptString).toHaveBeenCalled()
     // The value should come back decrypted
     expect(projects[0]?.envVars?.[0]?.key).toBe('API_KEY')
+    expect(projects[0]?.envVars?.[0]?.value).toBe('secret123')
   })
 
   it('handles encryption unavailable gracefully', async () => {

src/main/pty-manager.test.ts

@@ -389,6 +389,23 @@ describe('createPtyManager', () => {
         expect.objectContaining({ type: 'think' }),
       )
     })
+
+    it('handles multi-line data chunks correctly', () => {
+      const win = makeMockWindow()
+      const mgr = createPtyManager(win)
+
+      mgr.spawn('s1', 80, 24)
+      const cb = ptyInstances[0]?.onData[0]
+      // Send a chunk with multiple lines at once
+      cb?.('Read file1.ts\nWrite file2.ts\n')
+      vi.runAllTimers()
+
+      // Should have emitted activity for both lines
+      const calls = vi
+        .mocked(win.webContents.send)
+        .mock.calls.filter((c) => c[0] === 'pty:activity:s1')
+      expect(calls.length).toBeGreaterThanOrEqual(2)
+    })
   })
 
   describe('line buffer cap', () => {

src/main/workflow-engine-execution.test.ts

@@ -276,6 +276,7 @@ describe('concurrent execution', () => {
     engine.run(wf)
     await tick()
 
+    // MAX_TIER_CONCURRENCY = 5 (defined in workflow-engine.ts)
     expect(mockSpawn).toHaveBeenCalledTimes(5)
 
     // Close the first child — the 6th node should now spawn
@@ -536,6 +537,10 @@ describe('error scenarios', () => {
     const errors = getEvents(sendSpy, 'wf-idle', 'node:error')
     expect(errors).toHaveLength(1)
     expect(String(errors[0]?.message)).toContain('idle')
+
+    // Verify workflow-level terminal event was emitted (stopped due to node failure)
+    const workflowStopped = getEvents(sendSpy, 'wf-idle', 'workflow:stopped')
+    expect(workflowStopped).toHaveLength(1)
   })
 
   it('kills agent after absolute timeout (node.timeout)', async () => {

src/main/workflow-engine.test.ts

@@ -1,12 +1,6 @@
 import { describe, it, expect, beforeEach } from 'vitest'
-import {
-  stripAnsi,
-  shellQuote,
-  topoSort,
-  validateWorkflow,
-  AGENT_IDLE_TIMEOUT,
-} from './workflow-engine'
-import type { WorkflowNode } from '../shared/types'
+import { stripAnsi, shellQuote, topoSort, validateWorkflow } from './workflow-engine'
+
 import { makeWorkflowNode, makeWorkflowEdge, makeWorkflow, resetCounter } from '../__test__/helpers'
 
 beforeEach(() => {
@@ -728,31 +722,3 @@ describe('validateWorkflow', () => {
     expect(result.errors.length).toBeGreaterThanOrEqual(2)
   })
 })
-
-// ── AGENT_IDLE_TIMEOUT ────────────────────────────────────
-
-describe('AGENT_IDLE_TIMEOUT', () => {
-  it('is exported and equals 300000 (5 minutes)', () => {
-    expect(AGENT_IDLE_TIMEOUT).toBe(300_000)
-  })
-})
-
-// ── continueOnError type ──────────────────────────────────
-
-describe('WorkflowNode continueOnError', () => {
-  // T1: Verify the type accepts continueOnError flag
-  it('accepts continueOnError as optional boolean on WorkflowNode', () => {
-    const node: WorkflowNode = makeWorkflowNode({
-      id: 'n-coe',
-      type: 'shell',
-      command: 'npm test',
-      continueOnError: true,
-    })
-    expect(node.continueOnError).toBe(true)
-  })
-
-  it('defaults continueOnError to undefined when not set', () => {
-    const node: WorkflowNode = makeWorkflowNode({ id: 'n-default' })
-    expect(node.continueOnError).toBeUndefined()
-  })
-})