# ANALYSIS Mode Sub-Agent Specification

ANALYSIS mode tasks are long-running (5-40 minutes), **you must spawn a sub-agent to execute in background** to avoid blocking the main conversation flow.

---

## Main Agent Behavior

1. Inform user that deep analysis has started, provide Session ID
2. Return immediately, continue responding to other user questions (non-blocking)

---

## Sub-Agent Behavior

1. Execute command in background, periodically poll `status.txt` (every 30-60 seconds)
2. Push progress updates to main agent/user at key checkpoints
3. Return report or guide user to console when task completes

---

## Key Checkpoints for Sub-Agent to Report

| Checkpoint | Trigger Condition | Report Content | Required Action |
|------|----------|----------|----------|
| 🚀 Task Started | After command execution | Session ID, estimated duration | No action needed |
| 📋 Execution Plan Ready | `status.txt` = `waiting_input` | Show analysis plan/SQL | **Must wait for user confirmation, then `attach -q 'confirm execution'`** |
| 📊 Phase Conclusion | New content in `progress.log` | Phase analysis summary | No action needed |
| ✅ Task Completed | `status.txt` = `completed` | Report link or file path | Guide user to view report |
| ❌ Task Failed | `status.txt` = `failed` | Error message | Prompt user to retry |

> ⚠️ **`waiting_input` is a mandatory wait point**: Worker has exited, sub-agent must present execution plan to user, execute `attach -q 'confirm execution'` after receiving confirmation, otherwise task will be permanently paused.

---

## Guide User to View Report After Task Completion

```bash
# Download report to local
python3 scripts/data_agent_cli.py reports --session-id <SESSION_ID>

# Or guide user to Data Agent console (recommended)
# https://agent.dms.aliyun.com/<region>/session/<SESSION_ID>
```