alibabacloud-adb-openclaw-insight

name

alibabacloud-adb-openclaw-insight

description

OpenClaw conversation log collection and deep insight analysis Skill. Collects OpenClaw session logs (JSONL format) in real time and pushes them to Alibaba Cloud AnalyticDB MySQL (ADB) for storage. Provides a three-layer insight analysis architecture: L1 Operational Efficiency (Token efficiency, session depth, tool chain analysis, high-cost attribution, anomaly detection), L2 User Behavior (intent classification, task complexity, success rate, prompt quality, topic clustering, retry detection, thinking depth, user maturity), and L3 Organizational Cognition (tech stack heatmap, knowledge gap discovery, best practice extraction, skill candidate discovery, narrative report generation). Powered by SQL + Python + LLM. Use this Skill when you need to monitor OpenClaw usage, analyze costs, understand user behavior patterns, or generate organizational intelligence reports.

OpenClaw Logger Insight ADB Skill

Collect OpenClaw session logs in real time and push them to AnalyticDB MySQL. Analyze usage patterns with a three-layer insight architecture powered by SQL + Python + LLM.

Prerequisites

• Python >= 3.10 (use python or python3 depending on your system)
• An accessible Alibaba Cloud AnalyticDB MySQL instance
• OpenClaw deployed and generating session files (~/.openclaw/agents/*/sessions/*.jsonl) and logs (/tmp/openclaw/openclaw-YYYY-MM-DD.log)
• (Optional) An OpenAI-compatible or Anthropic LLM API endpoint for L2/L3 analysis

Quick Start

# 1. Install uv package manager
curl -LsSf https://astral.sh/uv/install.sh | sh

# 2. Install dependencies
uv pip install -r requirements.txt

# 3. Copy the configuration template
cp config.example.json config.json
# Edit config.json: fill in ADB connection details and (optionally) LLM API config

# 4. Initialize the database tables
uv run python -m scripts.init_db

# 5. (Optional) Start the all-in-one service (collection + scheduled analysis)
uv run python -m scripts.main

CLI Commands

Collect — One-shot data collection

Scans new session JSONL files and daily log files, inserts records into ADB, saves the file-offset checkpoint, then exits. Safe to call repeatedly.

uv run python -m scripts.main collect

Analyze — Run full insight analysis

Runs the full three-layer analysis pipeline (L1 Operational → L2 Behavior → L3 Organizational → Final Report) over the configured time window.

uv run python -m scripts.main analyze

Run with a custom time range:

# Time format: YYYY-MM-DD or YYYY-MM-DD HH:MM:SS
uv run python -m scripts.analyze_usage --from "2026-03-01 00:00:00" --to "2026-03-10 23:59:59"

Final Report — Print the latest report

Fetches and prints the most recent narrative report stored in ADB.

uv run python -m scripts.main final-report

Scheduled Collection via OpenClaw Cron

python -m scripts.main collect is the recommended way to keep data flowing into ADB. It runs a single collection pass, saves the file-offset checkpoint, and exits — making it safe to call repeatedly from any scheduler.

Register it as an OpenClaw cron job (example: every 30 seconds):

{
  "cron": "*/30 * * * * *",
  "command": "python -m scripts.main collect",
  "cwd": "/path/to/alibabacloud-adb-mysql-mcp-server/skill/alibabacloud-adb-openclaw-insight"
}

Each invocation:

1. Scans new JSONL session files and daily log files since the last run
2. Inserts new records into ADB in batches
3. Saves the file-offset checkpoint (.collect_state.json) so the next run picks up exactly where this one left off
4. Exits cleanly — no background process to manage

Configuration

See config.example.json for all options:

• adb: ADB connection (host, port, database, credentials, table name)
• collection: Collection parameters (interval, batch size, retention days)
• filters: Log filtering (minimum level, subsystem include/exclude)
• llm: LLM API configuration (endpoint, API key, model, concurrency, temperature)
• analysis: Analysis toggles (enableL1/L2/L3, analysis window days, max sessions for LLM)

Note: L1 analysis runs without LLM. L2 and L3 require a configured LLM endpoint.