A reliability tool that monitors AI subagent runs by enforcing completion markers within defined deadlines.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install subagent-watchdog
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install subagent-watchdog using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Subagent Watchdog?
Subagent Watchdog is a specialized reliability tool designed for complex agentic workflows. In environments where AI agents spawn subagents to handle specific tasks, ensuring task completion is critical for system stability. This skill acts as a supervisor, monitoring the lifecycle of subagent runs through a labeling system. By integrating this into your Openclaw Skills library, you can prevent hanging processes and ensure that every autonomous task either completes successfully or triggers an immediate alert for manual intervention or automated retry.
Subagent Watchdog Use Cases
- Preventing infinite loops or stalled subagent tasks in autonomous coding workflows.
- Automating retries for failed subagents that miss their completion deadlines.
- Standardizing completion reporting across multiple concurrent Openclaw Skills subagents.
- Integrating real-time failure notifications into developer communication channels.
How Subagent Watchdog Works
- A subagent is spawned with a unique label and a specific set of instructions.
- The watchdog timer is initialized using the watch.sh runner, specifying the label and an optional timeout duration.
- The subagent executes its task and is contractually required to write a .done marker file to a specific state directory.
- If the watchdog detects the marker file before the timer expires, the process exits cleanly.
- If the deadline passes without a marker, the watchdog triggers a failure alert and exits with a non-zero code.
Subagent Watchdog Setup
To integrate this into your workflow, first define the subagent contract in your prompts to ensure they write to subagent-watchdog/state/
# Run watchdog with a 10-minute timeout
./watch.sh "data-processing-agent" 600
You can also configure default timeouts in your ~/.openclaw/openclaw.json file:
{
"caliban": {
"subagentWatchdog": {
"maxRuntimeSeconds": 600
}
}
}
Subagent Watchdog Data Schema & Taxonomy
The skill utilizes a structured state directory to track agent progress and manage configurations within the Openclaw Skills ecosystem:
| File/Directory | Description |
|---|---|
| subagent-watchdog/state/ | The central directory for all active and completed task markers. |
| A marker file written by the subagent to signal successful completion. | |
| SUBAGENT_CONTRACT.md | Documentation defining the expected behavior and output format for subagents. |
| ~/.openclaw/openclaw.json | Configuration file for global watchdog settings and messaging IDs. |
Subagent Watchdog Advanced Features
- Support for global configuration via openclaw.json for standardized timeout management across large projects.
- Integration with OpenClaw messaging for instant failure alerts via the WATCHDOG_CHAT_ID environment variable.
- Portable shell-based design that works effectively with or without the full Openclaw binary present.
- Dynamic timeout inheritance that automatically pulls from user settings if no explicit time is provided.
SKILL.md
Loading
Related Openclaw Skills
Braingit
A deterministic utility for creating automated git snapshot commits specifically for Markdown files within a repository.
Feishu Agent Relay
A framework for building multi-agent collaboration systems on Feishu by managing cross-bot user identity mapping and task relay.
index1
A persistent AI memory system providing coding agents with hybrid search, code indexing, and cognitive fact recording across sessions.
index1 Doctor
A specialized diagnostic utility that verifies the technical health, dependencies, and model configuration of index1 environments.
Config Safe
A protective workflow skill designed to safely modify OpenClaw configurations through schema validation and mandatory human-in-the-loop confirmation.
SafeSkills
A secure execution service that allows AI agents to manage EVM wallets and secrets without ever accessing raw private keys.
Featured*
