A systematic debugging framework designed to identify and resolve AI agent failures, infinite loops, and resource bottlenecks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-debugger
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 agent-debugger 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 Agent Debugger?
Agent Debugger is a specialized toolset within the Openclaw Skills library built to handle the unique challenges of AI agent development. It provides a structured approach to diagnosing common autonomous agent issues such as context overflow, rate limiting, and permission errors. By implementing these debugging protocols, developers can ensure their agents remain reliable, efficient, and capable of recovering from unexpected tool failures or logic loops.
This skill serves as an essential companion for anyone building complex agentic workflows, offering both diagnostic patterns and concrete code snippets for remediation. Whether you are dealing with a stuck reasoning chain or fragmented memory, Agent Debugger provides the technical guardrails needed to stabilize agent performance and optimize token consumption.
Agent Debugger Use Cases
- Resolving infinite loops where an agent repeats the same tool call or output pattern.
- Debugging tool failures caused by schema mismatches, missing permissions, or API timeouts.
- Managing context overflow and 'context length exceeded' errors in long-running conversations.
- Implementing exponential backoff and request queuing to handle provider rate limits.
- Fixing memory issues where an agent loses track of prior decisions or fails to load MEMORY.md.
How Agent Debugger Works
- Reproduce: Document the exact steps and inputs that trigger the misbehavior or error.
- Isolate: Simplify the task, disable unnecessary Openclaw Skills, and minimize context to find the specific point of failure.
- Diagnose: Analyze logs, check current token usage, and verify tool configuration settings.
- Fix: Apply corrective measures such as adding iteration limits, retry logic, or context compression.
- Prevent: Update agent instructions with new guardrails and document the resolution in the agent's memory to avoid recurrence.
Agent Debugger Setup
To begin using the debugging protocols provided by this skill, ensure your workspace is correctly configured for Openclaw Skills development:
# Verify your workspace structure
ls -la ~/.openclaw/workspace/
# Ensure memory files are present
touch ~/.openclaw/workspace/MEMORY.md
Configure your agent with verbose thinking to expose internal reasoning during the debug phase:
{
"thinking": "verbose",
"reasoning": "on"
}
Agent Debugger Data Schema & Taxonomy
The Agent Debugger utilizes a structured approach to categorize and track agent errors. Below is the primary error mapping used by the skill:
| Error Code | Primary Cause | Remediation Strategy |
|---|---|---|
context_length_exceeded |
Excessive history/file data | Summarization & Truncation |
rate_limit_exceeded |
API frequency limits | Backoff & Queuing |
tool_not_found |
Registry mismatch | Installation & Path Check |
permission_denied |
Sandbox restrictions | Policy Adjustment |
invalid_parameters |
Schema violation | Input Validation |
Agent Debugger Advanced Features
- Iteration Limits: Define maxIterations and onLimit behaviors to break infinite loops automatically.
- Adaptive Context Compression: Dynamically use offsets and limits for large file reads to prevent context overflow.
- Custom Retry Logic: Implement granular retry strategies specifically for 5xx errors or timeouts.
- Parallel Tool Optimization: Combine multiple read operations into single calls to reduce latency.
- Checkpointing: Periodically save agent state to allow resumes during long-running tasks.
SKILL.md
Loading
Related Openclaw Skills
Task Automation
A comprehensive toolkit for automating repetitive work through scripts, scheduled jobs, and complex multi-step workflows.
Protein Mining Skill
An end-to-end automated pipeline for protein discovery that transitions from literature surveys to structural analysis and experimental prioritization.
Baidu Hot Topics Monitor
A streamlined tool to monitor Baidu's real-time trending searches, rankings, and popularity indices directly from your AI agent workflow.
Optimacros AI Modeling & Formula Expert
A specialized AI assistant for Optimacros developers focusing on complex formulas, model architecture, and enterprise-grade access security.
Crypto Real-Time USD Charting
A high-performance real-time WebSocket feed providing 1-second OHLC ticks and USD pricing for thousands of tokens across eight major blockchain networks.
Multi Search Engine
A comprehensive search aggregation skill supporting 17 global and domestic engines with advanced filtering and zero API key requirements.
Featured*
