pal-debug
使用 PAL MCP 进行系统调试和根本原因分析。用于复杂的错误、神秘的错误、竞争条件、内存泄漏和集成问题。在调试请求、错误调查或遇到困难问题时触发。
📖 中文导读:使用 PAL MCP 进行系统调试和根本原因分析。用于复杂的错误、神秘的错误、竞争条件、内存泄漏和集成问题。在调试请求、错误调查或遇到困难问题时触发。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。
name: pal-debug description: Systematic debugging and root cause analysis using PAL MCP. Use for complex bugs, mysterious errors, race conditions, memory leaks, and integration problems. Triggers on debugging requests, error investigation, or when stuck on difficult issues.
PAL Debug - Root Cause Analysis
Systematic debugging with hypothesis testing and expert validation through the PAL MCP server.
When to Use
- Complex bugs that aren't obvious
- Mysterious errors with unclear causes
- Race conditions or timing issues
- Memory leaks or performance problems
- Integration failures between systems
- When you've tried basic debugging and are stuck
Quick Start
Use the mcp__pal__debug tool for multi-step investigation:
# Step 1: Start investigation
result = mcp__pal__debug(
step="Investigating: API returns 500 on concurrent requests",
step_number=1,
total_steps=3,
next_step_required=True,
findings="Initial investigation - gathering context",
hypothesis="Unknown - needs investigation",
confidence="exploring",
relevant_files=["/path/to/api/handler.py"]
)
# Step 2+: Continue with continuation_id
result = mcp__pal__debug(
step="Found evidence in logs showing connection pool exhaustion",
step_number=2,
total_steps=3,
next_step_required=True,
findings="Connection pool limit reached under load",
hypothesis="Database connection pool too small for concurrent requests",
confidence="high",
continuation_id=result["continuation_id"]
)
Required Parameters
| Parameter | Type | Description |
|---|---|---|
step |
string | Current investigation narrative |
step_number |
int | Current step (starts at 1) |
total_steps |
int | Estimated total steps needed |
next_step_required |
bool | True if more investigation needed |
findings |
string | Evidence and discoveries |
Optional Parameters
| Parameter | Type | Description |
|---|---|---|
hypothesis |
string | Current root cause theory |
confidence |
enum | exploring/low/medium/high/very_high/almost_certain/certain |
relevant_files |
list | Absolute paths to relevant files |
files_checked |
list | All files examined |
issues_found |
list | Issues with severity levels |
continuation_id |
string | Continue previous session |
model |
string | Override model (default: openai/gpt-5) |
thinking_mode |
enum | minimal/low/medium/high/max |
Confidence Levels
exploring- Just starting, no theory yetlow- Early hypothesis, little evidencemedium- Some supporting evidencehigh- Strong evidence for theoryvery_high- Very confident, need verificationalmost_certain- Nearly confirmedcertain- 100% confirmed (skips external validation)
Workflow Pattern
Step 1: State the problem and initial direction
↓
Step 2: Gather evidence, form hypothesis
↓
Step 3: Test hypothesis, refine or pivot
↓
Step N: Confirm root cause, propose fix
Example: Database Connection Issue
# Start
mcp__pal__debug(
step="API returning 500 errors under load. Starting investigation.",
step_number=1,
total_steps=4,
next_step_required=True,
findings="Errors correlate with high traffic periods",
hypothesis="Resource exhaustion under load",
confidence="exploring",
relevant_files=[
"/app/api/routes.py",
"/app/db/connection.py"
]
)
Best Practices
- Start broad, narrow down - Don't assume the cause upfront
- Document everything - Track files checked, even dead ends
- Update hypothesis - Revise as new evidence emerges
- Use continuation_id - Preserve context across steps
- Set realistic steps - Adjust total_steps as complexity reveals itself