UI stuck at 10% 'Analyzing' while backend completes successfully - task_logs.json parsing failures

[ras434]

Summary

The Electron UI shows a task stuck at 10% progress on "Analyzing project structure..." for 24+ hours, while the Python backend actually completed all work successfully (8/8 subtasks, QA approved, ready to merge). The UI never updates because task_logs.json encounters JSON parsing errors during live updates, causing TaskLogService to fail silently and freeze the progress display.

This leads users to believe tasks are stuck and delete completed work.

Environment

• Auto-Claude UI Version: v2.3.0
• Latest Commit: a20b8cf (Enhance task status handling...)
• OS: macOS 14.6.0 (Darwin 24.6.0)
• Node.js: v24.11
• Python: 3.14 (in venv)
• Project: Medium-sized React/TypeScript project

Problem Description

What the User Saw

1. Started a new task via Auto-Claude UI
2. Progress bar showed "Analyzing project structure... 10%"
3. Progress remained at 10% for ~24 hours despite high CPU activity
4. Terminal showed repeated JSON parsing errors in task_logs.json
5. User assumed task was stuck and deleted it
6. Actual state: All work was complete, QA approved, ready to merge

What Actually Happened (Backend)

The Python backend (spec_runner.py → run.py → agents) completed successfully:

• ✅ All 8 subtasks completed
• ✅ 90/90 unit tests passing
• ✅ TypeScript build: SUCCESS
• ✅ Type check: SUCCESS
• ✅ Lint: 0 errors
• ✅ QA validation: APPROVED at 2025-12-18T16:17:00Z
• ✅ 11 commits on feature branch
• ✅ Status in implementation_plan.json: "qa_signoff": { "status": "approved" }

Root Cause

The Electron UI's TaskLogService (src/main/task-log-service.ts) watches task_logs.json and attempts to parse it every few seconds. During active agent sessions, the file is being written to concurrently, resulting in transient malformed JSON.

Terminal errors observed:

[TaskLogService] Failed to load logs from .../task_logs.json: 
  SyntaxError: Expected ',' or '}' after property value in JSON at position 128794 (line 584 column 7712)

[TaskLogService] Failed to load logs from .../task_logs.json:
  SyntaxError: Expected ',' or '}' after property value in JSON at position 346013 (line 1984 column 8331)

[TaskLogService] Failed to load logs from .../task_logs.json:
  SyntaxError: Unexpected end of JSON input

When parsing fails, TaskLogService.loadLogsFromPath() catches the error, logs it, and returns null. The UI receives no updates, and the progress bar freezes at its last known state (10%).

Expected Behavior

1. UI should show task progress even if task_logs.json has transient parsing errors
2. Progress should advance as phases complete (10% → 25% → 50% → 75% → 100%)
3. Final status should reflect QA approval, not show "stuck at Analyze 10%"
4. User should see "Ready to Merge" notification when QA approves

Actual Behavior

1. ✗ UI shows "Analyzing... 10%" indefinitely
2. ✗ Progress bar never advances past 10%
3. ✗ No indication that backend completed successfully
4. ✗ User deletes what they believe is a stuck task (losing completed work)

Technical Details

File Sizes at Completion

task_logs.json: 884,531 bytes (864 KB)
Lines: 6,432

Progress Tracking Logic

From auto-claude-ui/src/main/agent/agent-process.ts:216-221:

// Reset phase progress on phase change, otherwise increment
if (phaseChanged) {
  phaseProgress = 10; // Start new phase at 10%
} else {
  phaseProgress = Math.min(90, phaseProgress + 5); // Increment within phase
}

Problem: When the first phase ("Analyze") starts, progress is set to 10%. If subsequent progress updates fail due to JSON parsing errors, it never advances.

JSON Parsing Logic

From auto-claude-ui/src/main/task-log-service.ts:43-50:

try {
  const content = readFileSync(logFile, 'utf-8');
  const logs = JSON.parse(content) as TaskLogs;
  return logs;
} catch (error) {
  console.error(`[TaskLogService] Failed to load logs from ${logFile}:`, error);
  return null; // ← Silent failure
}

Problem: No retry logic, no incremental parsing, no fallback to last known good state.

Steps to Reproduce

1. Create a new task on a medium-to-large project (triggers long-running spec creation)
2. Observe terminal output for [TaskLogService] JSON parsing errors
3. Note that UI progress remains at "Analyzing... 10%" despite errors
4. Let task run to completion (~hours for complex projects)
5. Check implementation_plan.json - will show "qa_signoff": { "status": "approved" }
6. UI still shows "Analyzing... 10%" - disconnect between UI and reality

Impact

• Critical: Users delete completed work thinking tasks are stuck
• Wasted compute: Hours of Claude API usage discarded
• Loss of trust: Users lose confidence in Auto-Claude's reliability
• Poor UX: No visibility into actual backend progress

Suggested Fixes

1. Graceful JSON Parsing (Short-term)

// task-log-service.ts
private loadLogsFromPath(logFile: string): TaskLogs | null {
  try {
    const content = readFileSync(logFile, 'utf-8');
    const logs = JSON.parse(content) as TaskLogs;
    this.lastKnownGoodState = logs; // Cache last good parse
    return logs;
  } catch (error) {
    console.warn(`[TaskLogService] JSON parse failed, using cached state:`, error.message);
    return this.lastKnownGoodState ?? null; // Fall back to cache
  }
}

2. File Locking or Atomic Writes (Medium-term)

Ensure Python agents write to task_logs.json.tmp, then atomically rename to task_logs.json to prevent reading mid-write.

3. Streaming JSON or NDJSON Format (Long-term)

Replace single JSON object with newline-delimited JSON (NDJSON) where each log entry is a separate line. Allows reading partial files without parsing errors.

4. Direct IPC from Python Backend (Long-term)

Instead of polling file changes, have Python agents send progress updates via IPC/WebSocket directly to Electron main process.

5. UI Fallback to implementation_plan.json (Short-term)

If task_logs.json fails to parse, read implementation_plan.json to determine actual phase/subtask status:

// Check subtask completion percentage
const completed = phases.flatMap(p => p.subtasks).filter(s => s.status === 'completed').length;
const total = phases.flatMap(p => p.subtasks).length;
const progress = Math.round((completed / total) * 100);

Evidence

Terminal Output Showing Repeated Errors

[TaskLogService] Failed to load logs from .../task_logs.json: SyntaxError: Expected ',' or '}' after property value in JSON at position 128794 (line 584 column 7712)
[TaskLogService] Failed to load logs from .../task_logs.json: SyntaxError: Expected ',' or '}' after property value in JSON at position 346013 (line 1984 column 8331)
[TaskLogService] Failed to load logs from .../task_logs.json: SyntaxError: Unexpected end of JSON input

(Repeated dozens of times during task execution)

Worktree Contents Prove Completion

$ git log --oneline auto-claude/001-example-feature --not main | head -5
abc1234 qa: Sign off - all verification passed
def5678 auto-claude: subtask-4-2 - Export new utility functions
ghi9012 fix: Restore implementation plan status after accidental reset
jkl3456 auto-claude: subtask-4-1 - Run full test suite and fix any failures
mno7890 auto-claude: subtask-3-2 - Visual verification and edge case testing

QA Report Confirms Approval

From .auto-claude/specs/.../qa_report.md:

## Verdict

**SIGN-OFF**: APPROVED ✓

**Reason**:
All automated verification steps pass:
- All 8 implementation subtasks completed
- 90/90 unit tests passing for new functions
- TypeScript compilation succeeds with no errors
- Lint passes with 0 errors
- Production build succeeds

Related Issues

After searching, no duplicate issues found for this specific problem (UI stuck at 10% with JSON parsing failures).

Possibly related:

• Planning Always Failing #38 - Planning phase failures (validation mismatch, but different symptom)
• Starting a Task just skips it directly to Human review without doing anything #11 - Tasks skip to human review (missing auth, but different symptom)

Workaround (For Users Who Hit This)

If your UI shows a task stuck at "Analyzing... 10%" for hours:

1. Check the actual status in the worktree:

   cat .worktrees/{task-name}/.auto-claude/specs/{task-name}/implementation_plan.json | grep "qa_signoff" -A 5

2. If QA approved, the work is done! Merge it:

   git checkout main
   git merge auto-claude/{task-name}
   git worktree remove .worktrees/{task-name}

3. If still in progress, check last subtask status:

   cat .worktrees/{task-name}/.auto-claude/specs/{task-name}/implementation_plan.json | grep '"status"' | tail -5

---

Request: Please prioritize this issue as it causes users to discard completed work, wasting both time and API costs.

[coderabbitai]

📝 CodeRabbit Plan Mode

Generate an implementation plan and prompts that you can use with your favorite coding agent.

• Create Plan

Examples

• Example 1
• Example 2

---

🔗 Similar Issues

Related Issues

• Phase 'planning' failed #3
• Planning Always Failing #38
• Starting a Task just skips it directly to Human review without doing anything #11
• Fisrt task #5

🔗 Related PRs

#10 - Recent updates: roadmap enhancements, bug fixes, and drag-and-drop support [merged]
#22 - Version/2.5.5 [merged]
#24 - Fix GitHub organization repository support [merged]
#37 - Fix Windows Python detection and initialization popup issues [merged]
#41 - fix: human feedback not processed when QA already approved [merged]

👤 Suggested Assignees

• AndyMik90
• johann-taberlet
• phill-bramble
• DevExord
• bacoco

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord or schedule a call!

[ras434]

Additional Bugs Discovered & Fixed - Roadmap/Ideation Features Completely Broken

While investigating this UI stuck-at-10% issue, I discovered the Roadmap and Ideation features are completely non-functional in v2.3.0 due to multiple path resolution bugs introduced when the runners were moved to a subdirectory. These bugs affect 100% of users attempting to use these features.

---

Bug Summary

Bug #	Component	Severity	Impact
#1	UI can't find roadmap_runner.py	🔴 Critical	Roadmap feature completely broken
#2	UI can't find ideation_runner.py	🔴 Critical	Ideation feature completely broken
#3	roadmap_runner.py can't import modules	🔴 Critical	Prevents roadmap execution
#4	ideation_runner.py can't import modules	🔴 Critical	Prevents ideation execution
#5	executor.py can't find scripts	🔴 Critical	Analyzer script fails
#6	executor.py can't find prompts	🔴 Critical	Agent prompts not loaded
#7	Renderer crashes when roadmap completes	🟠 High	UI becomes blank screen

---

Root Cause

Recent refactoring (commit 3d24f8f - "fix: correct path resolution in runners") moved Python runners from auto-claude/ to auto-claude/runners/ subdirectory, but only partially updated the path references. This broke:

• Electron UI → Python runner invocation
• Python runner → auto-claude module imports
• Script executor → helper script discovery

---

Bugs & Fixes

Bug #1 & #2: UI Can't Find Runner Scripts

File: auto-claude-ui/src/main/agent/agent-queue.ts

Lines: 48, 86

Problem: UI looks for runners in wrong directory

// ❌ BEFORE (Line 48)
const roadmapRunnerPath = path.join(autoBuildSource, 'roadmap_runner.py');

// ❌ BEFORE (Line 86)
const ideationRunnerPath = path.join(autoBuildSource, 'ideation_runner.py');

Fix: Add 'runners/' subdirectory

  // Line 48
- const roadmapRunnerPath = path.join(autoBuildSource, 'roadmap_runner.py');
+ const roadmapRunnerPath = path.join(autoBuildSource, 'runners', 'roadmap_runner.py');

  // Line 86
- const ideationRunnerPath = path.join(autoBuildSource, 'ideation_runner.py');
+ const ideationRunnerPath = path.join(autoBuildSource, 'runners', 'ideation_runner.py');

Error Message Before Fix:

[roadmap-error] Roadmap runner not found at: /path/to/auto-claude/roadmap_runner.py

---

Bug #3 & #4: Runners Can't Import Modules

Files:

• auto-claude/runners/roadmap_runner.py
• auto-claude/runners/ideation_runner.py

Lines: ~20-27 in both files

Problem: sys.path points to runners/ directory instead of auto-claude/

# ❌ BEFORE (roadmap_runner.py line 21)
sys.path.insert(0, str(Path(__file__).parent))

# ❌ BEFORE (line 26)
env_file = Path(__file__).parent / ".env"

Fix: Go up one more directory level

  # Add auto-claude to path
- sys.path.insert(0, str(Path(__file__).parent))
+ # Go up from runners/ -> auto-claude/
+ sys.path.insert(0, str(Path(__file__).parent.parent))

  # Load .env file
  from dotenv import load_dotenv

- env_file = Path(__file__).parent / ".env"
+ env_file = Path(__file__).parent.parent / ".env"

Error Message Before Fix:

ModuleNotFoundError: No module named 'debug'

Apply same fix to ideation_runner.py lines 26-32

---

Bug #5 & #6: Executor Can't Find Scripts/Prompts

File: auto-claude/runners/roadmap/executor.py

Lines: 17, 81

Problem: Paths resolve from roadmap/ subdirectory, need to go up 2 levels

# ❌ BEFORE (Line 17 - ScriptExecutor.__init__)
self.scripts_base_dir = Path(__file__).parent.parent
# This resolves to: runners/ (WRONG - should be auto-claude/)

# ❌ BEFORE (Line 80 - AgentExecutor.__init__)
self.prompts_dir = Path(__file__).parent.parent / "prompts"
# This resolves to: runners/prompts/ (WRONG - should be auto-claude/prompts/)

Fix: Add one more .parent to traverse directory tree correctly

  # Line 15-18
  def __init__(self, project_dir: Path):
      self.project_dir = project_dir
-     self.scripts_base_dir = Path(__file__).parent.parent
+     # Go up from roadmap/ -> runners/ -> auto-claude/
+     self.scripts_base_dir = Path(__file__).parent.parent.parent

  # Line 75-81
  def __init__(
      self,
      project_dir: Path,
      output_dir: Path,
      model: str,
      create_client_func,
  ):
      self.project_dir = project_dir
      self.output_dir = output_dir
      self.model = model
      self.create_client = create_client_func
-     self.prompts_dir = Path(__file__).parent.parent / "prompts"
+     # Go up from roadmap/ -> runners/ -> auto-claude/, then find prompts/
+     self.prompts_dir = Path(__file__).parent.parent.parent / "prompts"

Error Message Before Fix:

[ERROR] [roadmap_executor] Script not found: /path/to/auto-claude/runners/analyzer.py

---

Bug #7: Renderer Crash on Roadmap Completion

Symptom: Electron UI goes completely blank (black screen) when roadmap generation completes at ~70% progress

Evidence:

• Roadmap data successfully written to .auto-claude/roadmap/roadmap.json
• Valid JSON with all features/phases generated
• UI crashes before displaying results

Likely Cause: Unhandled exception when loading completed roadmap data in React component

Current Workaround: Restart Electron app - roadmap loads correctly on second launch

Needs Investigation: React component error handling in roadmap loader

---

Testing Steps

Verify Fixes Work

1. Apply all 6 fixes above
2. Rebuild Electron UI: cd auto-claude-ui && pnpm run build
3. Restart app: pnpm run start
4. Test Roadmap:

   # From Auto-Claude UI: Click "Roadmap" → "Generate Roadmap"
   # Or test directly:
   cd auto-claude
   .venv/bin/python runners/roadmap_runner.py --project /path/to/any/project

5. Expected output:

   ✓ Created project_index.json
   ✓ Running discovery agent...
   ✓ Generating roadmap...
   ✓ Roadmap saved to .auto-claude/roadmap/roadmap.json
   

Verify Original Issue (stuck at 10%)

The original task stuck-at-10% issue is unrelated but triggered by the same symptom:

• Tasks appear stuck when JSON parsing fails
• UI doesn't update from 10% initial state
• Backend may complete successfully while UI shows frozen progress

---

Impact Assessment

Before Fixes:

• ❌ Roadmap generation: 100% failure rate
• ❌ Ideation generation: 100% failure rate
• ❌ Users see "Generation failed with exit code 1"

After Fixes:

• ✅ Roadmap generation: Works end-to-end
• ✅ Ideation generation: Should work (same pattern)
• ⚠️ Renderer crash on completion (Bug feat: Add multi-auth token support and ANTHROPIC_BASE_URL passthrough (CLI only) #7) - needs separate fix

---

Recommended Actions

1. Merge fixes Auto claude/add followup on done task #1-6 immediately - these are critical path bugs blocking features
2. Add integration tests to prevent path regressions:

   describe('Runner path resolution', () => {
     it('should find roadmap_runner.py in runners/ subdirectory', () => {
       const path = getRoadmapRunnerPath();
       expect(existsSync(path)).toBe(true);
     });
   });

3. Investigate Bug feat: Add multi-auth token support and ANTHROPIC_BASE_URL passthrough (CLI only) #7 (renderer crash) separately
4. Document directory structure in README to prevent future path issues

---

Files Changed

All fixes can be applied to commit a20b8cf:

File	Lines Changed	Type
auto-claude-ui/src/main/agent/agent-queue.ts	48, 86	Add 'runners/' path segment
auto-claude/runners/roadmap_runner.py	21, 26	Add .parent for path resolution
auto-claude/runners/ideation_runner.py	27, 32	Add .parent for path resolution
auto-claude/runners/roadmap/executor.py	18, 81	Add .parent for path resolution

Total: 4 files, 8 lines changed, ~30 minutes to fix and test

---

Related Issues

• Roadmap generation failed with exit code 1 #45 - "Roadmap generation failed with exit code 1" (same root cause as bugs Auto claude/add followup on done task #1-6)
• UI stuck at 10% 'Analyzing' while backend completes successfully - task_logs.json parsing failures #42 (this issue) - UI stuck at 10% (different but related symptom)

Would appreciate confirmation these fixes work for others before creating a PR.

[AndyMik90]

Fixed in commit 91a1e3d on v2.6.0 branch.

Root cause: The JSON cache in task-log-service.ts was never updated on successful parses, only read on errors.

Fix: Added this.logCache.set(specDir, logs) after successful JSON.parse() to cache valid states for fallback.