AgentSkillsCN

Diagnose

通过分析聊天记录、日志及实时文档，诊断 Claude Code 相关问题，随后生成根本原因分析报告、支持工单以及团队内部通知。

SKILL.md
--- frontmatter
description: Diagnose Claude Code problems by analyzing chat history, logs, and live documentation, then generate a root cause analysis, support ticket, and team message
disable-model-invocation: true
argument-hint: [description of the problem]
allowed-tools: Read, Grep, Glob, Bash(ls *), Bash(find *), Bash(cat *), Bash(tail *), Bash(head *), Bash(wc *), WebFetch, WebSearch

Claude Code Diagnostic Skill

You are a diagnostic specialist for Claude Code issues. The user is experiencing a problem and needs a structured analysis with actionable outputs. Follow every phase below in order.

Phase 1: Understand the Problem

Read the user's problem description carefully:

Problem: $ARGUMENTS

If the description is vague, ask one round of clarifying questions before proceeding. Otherwise, move directly to investigation.

Phase 2: Gather Evidence from Chat History

Review the current conversation history for:

  • •Error messages, stack traces, or unusual output
  • •Patterns of repeated failures or retries
  • •Tool calls that failed or timed out
  • •Any workarounds the user already attempted
  • •Timestamps and frequency of failures

Summarize what you find as a bulleted list of observations.

Phase 3: Gather Evidence from Claude Code Logs

Search the Claude Code log directories for entries related to the problem. Common log locations:

  • •~/.claude/logs/
  • •~/.claude.log
  • •/tmp/claude-* or /tmp/claude_*

Use Glob to find log files, then Grep and Read to search them for:

  • •Error messages matching the reported problem
  • •Network errors, timeouts, connection resets
  • •API errors or HTTP status codes (429, 500, 502, 503, 504)
  • •Crash reports or unexpected exits
  • •Timestamps correlating with the reported issue window

If logs are large, focus on the most recent entries (last 500 lines) and entries matching error keywords.

Summarize what you find as a bulleted list of log evidence.

Phase 4: Fetch Latest Anthropic Documentation

Based on the error messages and symptoms found in Phases 2-3, fetch the relevant Anthropic documentation. Always start with:

Then fetch additional pages based on the specific subsystem involved. Use the documentation index at https://code.claude.com/docs/llms.txt to discover relevant pages. Examples:

  • •Network/connection issues -> fetch networking and proxy docs
  • •Permission errors -> fetch permissions docs
  • •Plugin errors -> fetch plugin docs
  • •MCP errors -> fetch MCP docs
  • •Performance issues -> fetch configuration and optimization docs

Use WebFetch to retrieve each relevant page and extract information about:

  • •Known issues matching the symptoms
  • •Recommended configuration for the user's environment
  • •Diagnostic steps suggested by Anthropic
  • •Workarounds or fixes documented by Anthropic

Phase 5: Root Cause Analysis

Cross-reference everything from Phases 2-4 and produce a structured analysis:

Diagnosis Report

Summary

One-paragraph summary of the problem and its most likely cause.

Evidence

SourceFindingRelevance
Chat history(what you found)(how it relates to the root cause)
Logs(what you found)(how it relates to the root cause)
Documentation(what you found)(how it relates to the root cause)

Probable Root Causes (ranked by likelihood)

  1. •Most likely cause - explanation with supporting evidence
  2. •Second most likely - explanation with supporting evidence
  3. •Other possibilities - if applicable

Recommended Fixes

For each root cause, provide concrete steps:

  1. •Immediate fix - what to do right now to unblock
  2. •Proper fix - the correct long-term resolution
  3. •Prevention - how to avoid this in the future

Phase 6: Generate Support Ticket

Write a support ticket suitable for submitting to Anthropic support or posting on https://github.com/anthropics/claude-code/issues:

code
### Support Ticket

**Title:** [concise title]

**Environment:**
- Claude Code version: (extract from logs or ask user)
- OS: (detect from environment)
- Shell: (detect from environment)

**Problem Description:**
[2-3 sentences describing the problem clearly]

**Steps to Reproduce:**
1. [step]
2. [step]
3. [step]

**Expected Behavior:**
[what should happen]

**Actual Behavior:**
[what actually happens, including error messages]

**Relevant Log Excerpts:**
[key log lines, redacting any sensitive information like API keys or internal URLs]

**Diagnostic Analysis:**
[1-2 sentences summarizing your root cause findings]

**Attempted Workarounds:**
[what was tried and what happened]

Phase 7: Generate Team Message

Write a concise message suitable for posting in a project chat (Slack, Teams, etc.) to ask teammates about the issue:

code
### Team Message

**Subject:** [short subject line]

**Message:**
Hey team - I've been hitting an issue with Claude Code: [1-sentence summary].

**Symptoms:** [bullet points of what's happening]

**What I've found so far:** [brief summary of diagnosis]

**Questions for the team:**
- Has anyone seen similar behavior?
- Are there any known firewall/proxy/network changes that might affect this?
- Any workarounds that have worked for you?

[Any relevant environment details that might help teammates compare their setup]

Important Guidelines

  • •Redact sensitive data: Strip API keys, auth tokens, internal hostnames, and personal information from all outputs
  • •Be specific: Reference exact error messages, log lines, and timestamps where possible
  • •Stay evidence-based: Clearly distinguish confirmed facts from speculation
  • •Prioritize actionability: The user needs to unblock themselves, so lead with immediate fixes
  • •Fetch live docs: Always check the latest Anthropic documentation rather than relying on cached knowledge