Installable skillMIT

OpenClaw Troubleshooting

An MCP server that provides on-demand skill loading for AI coding assistants.

BY modbenderUPDATED Oct 27, 2026
INTENDED FOR◈OpenClawInstall in a listed agent.

instructions.md

---
name: openclaw-troubleshooting
description: "When users encounter OpenClaw-related issues or errors, search official docs and GitHub issues first, then provide solutions. Use for: feature malfunctions, error troubleshooting, version upgrade advice, configuration questions, and known issue verification."
---

# OpenClaw Troubleshooting

## Overview

When handling OpenClaw-related issues, follow the workflow of "check official resources first, then determine fix status."

## Workflow

### Step 1: Information Gathering

Confirm the following with the user (if not already provided):
- Current OpenClaw version (`openclaw --version`)
- Channel/platform being used (Feishu/Telegram/Discord/Slack, etc.)
- Specific problem symptoms (error messages, frequency, reproduction steps)
- Recent upgrades or configuration changes

### Step 2: Official Documentation Search

Search the official documentation for relevant content:
- URL: https://docs.openclaw.ai
- Search keywords: core issue terms (e.g., "feishu notification", "typing indicator", "cron job")

### Step 3: GitHub Issue Search

Search GitHub issues to confirm if this is a known problem:
- Issues URL: https://github.com/openclaw/openclaw/issues?q=is%3Aissue
- Search strategy (try in order):
  1. Exact search: issue keywords + channel name (e.g., "feishu reaction notification")
  2. Fuzzy search: channel name + feature module (e.g., "feishu typing")
  3. Error message search: extract key fields from error messages

### Step 4: Analyze Results and Report

Reply to the user using this structure:

```
**Problem Diagnosis**
- Symptom summary: (one sentence)
- Related GitHub issue: #[number] (if any)
- Fix status:
  - Fixed: version number + upgrade recommendation
  - Not fixed: temporary workaround + official fix progress
  - Configuration issue: provide correct configuration example
```

### Step 5: Record and Follow Up

If the issue is a known bug being fixed:
- Inform user of the expected fix version
- Record in todo list, remind user to verify after release

## Key Resources

### Quick Checklist

| Issue Type | Docs Section | GitHub Search Keywords |
|-----------|-------------|----------------------|
| Feishu notification issues | channels/feishu | feishu notification, reaction |
| Cron job problems | cron | cron job, schedule |
| Channel connection failures | onboarding | [channel] connection, webhook |
| Model invocation failures | models | [provider] error, fallback |
| Sandbox/execution issues | sandbox | sandbox, exec, security |

### CHANGELOG Check

- URL: https://raw.githubusercontent.com/openclaw/openclaw/main/CHANGELOG.md
- Purpose: Confirm whether a specific version includes a particular fix
- Search tip: Search for issue numbers (e.g., #28660) or feature keywords in CHANGELOG

## Response Template

Standard reply format for users:

---
Boss, I've checked. Here's the situation:

**Current version**: `x.x.x`
**Latest available version**: `x.x.x`
**GitHub issue**: [#number] title (if any)

**Fix status**:
- Version `x.x.x`: not fixed / fixed
- Version `x.x.x`: includes fix (CHANGELOG reference)

**Recommended solutions**:
1) Temporary workaround: (e.g., disable certain notification settings)
2) Permanent fix: (upgrade recommendation or wait for fix)

Which step would you like me to help with?
---

## Notes

- Prioritize the Fixes section in CHANGELOG to confirm fix versions
- If issue status is open, the fix is still in development
- For configuration issues, prioritize providing on-site verifiable solutions
- Record user wait decisions (e.g., "wait for next version"), follow up proactively
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.