| name | gui-report |
| description | Track and report GUI agent task performance — automatic data collection, manual report command. |
GUI Task Report
Automatic performance tracking for GUI tasks. Data collection is fully automatic, report generation is one command.
How It Works
detect_all() called → tracker auto-starts (records token baseline)
GUI functions called → counters auto-tick (screenshots, clicks, OCR, etc.)
task finished → run: tracker.py report (saves to log, prints summary)
Data collection needs zero manual effort. Only the final report command needs to be called.
Quick Start
source ~/gui-agent-env/bin/activate
python3 ~/.openclaw/workspace/skills/gui-agent/skills/gui-report/scripts/tracker.py report
python3 ~/.openclaw/workspace/skills/gui-agent/skills/gui-report/scripts/tracker.py history
What's Tracked (all automatic)
| Counter | Auto-ticked by |
|---|
| detector_calls | detect_all() |
| ocr_calls | detect_all() |
| screenshots | learn_from_screenshot() |
| learns | learn_from_screenshot() |
| clicks | record_page_transition() |
| transitions | record_page_transition() |
| workflow_level0 | quick_template_check() |
| workflow_level1 | execute_workflow() Level 1 |
| workflow_level2 | execute_workflow() Level 2 |
| workflow_auto_steps | execute_workflow() auto mode |
| workflow_explore_steps | execute_workflow() explore mode |
| image_calls | manual: tracker.py tick image_calls |
Report Output
📊 任务报告：chromium/united_com
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⏱ 耗时：2min 35s
💰 Token 消耗：
总计 +26.8k tokens
input +3.2k | output +1.8k | cache +21.8k
🔍 检测：
detect_all 8 次 | OCR 8 次 | LLM 视觉 1 次
组件总量：96（+12）
🖱 操作：
点击 5 次 | 状态转移 3 个 | 学习 3 次
🗺 导航效率：
自动模式 3 步 | 探索模式 2 步
自动率 60%
📝 记忆变化：
组件 +12（84 → 96）
状态 +1（5 → 6）
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Failsafe
If report is not called, data stays in .tracker_state.json. Next time tracker starts (next detect_all call), it auto-saves the previous session's data to the log before starting fresh. Data never lost.
Log Storage
skills/gui-report/logs/task_history.jsonl — one JSON per completed task.
