claude-watchdog

What This Skill Does

Claude Watchdog is a sophisticated telemetry and observability tool designed for power users and developers who rely on Anthropic's Claude API for mission-critical workflows. Rather than guessing why your agent might be sluggish, this skill provides precise, proactive monitoring via two primary mechanisms: a status monitor and a latency probe.

The status monitor queries status.claude.com every 15 minutes, parsing component health and incident reports to ensure you are never blindsided by outages. It is intelligent enough to filter alerts based on your monitored model, ensuring you don't receive irrelevant notifications for models you don't utilize.

The latency probe provides quantitative data by dispatching a minimal request through your local OpenClaw gateway every 15 minutes. This creates a rolling baseline, allowing the skill to flag performance degradation before a total service failure occurs. By sending notifications through Telegram, it ensures you remain informed regardless of your location, providing clear, actionable data directly to your mobile device.

Installation

Installation is streamlined via the OpenClaw hub and an interactive shell script.

1. Install via CLI: clawhub install openclaw/skills/skills/chapati23/claude-watchdog
2. Navigate to the script location and run: bash /path/to/skills/claude-watchdog/scripts/setup.sh
3. Follow the prompts to input your Telegram Bot Token and Chat ID.
4. Ensure your local OpenClaw gateway is running; the script will automatically pull the necessary auth token from your local configuration.
5. The setup script will register the required cron jobs for you automatically.

Use Cases

• High-Availability Agents: Ideal for agents running long-term tasks where downtime could result in lost time or data.
• Latency-Sensitive Infrastructure: Perfect for developers building real-time interactive apps on top of Claude who need to detect and debug performance bottlenecks immediately.
• Incident Management: Use the raw data from alerts to populate incident dashboards or notify your team's on-call engineers automatically.

Example Prompts

1. "What is the current status of Claude's latency according to the watchdog?"
2. "Update the watchdog settings to monitor only the Claude 3.5 Sonnet incidents."
3. "Show me the last 5 latency spikes detected by the watchdog."

Tips & Limitations

• Costs: While status checks are completely free, latency probes consume minimal tokens. Monitor your gateway usage logs if running at high frequency.
• Filtering: Use the FILTER_KEYWORDS configuration variable to avoid alert fatigue. For example, if you don't care about 'Artifacts' or 'Memory' status updates, exclude them to keep your Telegram feed focused on core model performance.
• Cron Precision: The 15-minute polling interval is a balance between granularity and resource usage; avoid setting this to sub-minute intervals to prevent triggering API rate limits on your host.