一键在 Manus 中运行任何 Skill

开始使用

debug

星标1

分支0

更新时间2026年1月18日 10:18

Use when encountering bugs, test failures, or unexpected behavior. Hypothesis-driven debugging.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

erikpr1994

erikpr1994/Jarvis-Code

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Systematic Debugging

Iron Law: NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST.

The Process

1. REPRODUCE  -> Trigger consistently
2. INVESTIGATE -> Gather evidence, trace data flow
3. HYPOTHESIZE -> Form ONE specific theory
4. TEST       -> Minimal change to verify
5. FIX        -> Address root cause
6. VERIFY     -> Confirm with test

Step 1: Reproduce

npm test -- --testPathPattern="auth"
# Document: exact steps, expected vs actual, full error message

Cannot reproduce? Gather more data. Don't guess.

Step 2: Investigate

Read error messages completely - line numbers, stack traces, error codes
Check recent changes: git diff HEAD~5
Trace data flow: Add logging at component boundaries

echo "=== Input to component: $INPUT ==="
# component does its thing
echo "=== Output: $OUTPUT ==="

Goal: Find WHERE it breaks, not just WHAT breaks.

Step 3: Hypothesize

Form ONE specific theory:

"Token is null because login returns before API response completes."

Not: "Something's wrong with auth" (too vague)

Step 4: Test Hypothesis

Smallest possible change:

console.log('Token before return:', token);

Confirmed? Proceed to fix.
Wrong? Form NEW hypothesis. Return to Step 3.

Step 5: Fix Root Cause

Symptom	Wrong Fix	Right Fix
Null pointer	Add null check	Fix why it's null
Timeout	Increase timeout	Fix why it's slow

Step 6: Verify

Write test that fails before fix, passes after. Proves fix works.

Decision Criteria

Situation	Action
Cannot reproduce	More data. Don't guess.
3+ failed attempts	Question architecture. Discuss with user.
"Quick fix" obvious	STOP. Follow process anyway.

Red Flags

"Let me just try..." -> STOP. Investigate first.
"Quick fix for now" -> Root cause first.
Multiple fixes at once -> One hypothesis at a time.

Integration

Pairs with: tdd (write failing test), verification (prove fix works)

同仓库更多 Skills

同仓库

git-expert

erikpr1994/Jarvis-Code

Use when working with git operations including commits, branches, worktrees, and PRs. Covers the full git workflow from feature isolation to PR submission.

2026-01-201

submit-pr

erikpr1994/Jarvis-Code

Complete PR submission pipeline with local sub-agent review before pushing, CI verification, and automated review integration. Always dispatches code-reviewer and code-simplifier for code changes, plus conditional reviewers (security, performance, dependency, accessibility, i18n, type-design, etc.) based on change type. Waits for CI with `gh pr checks --watch`. Integrates CodeRabbit and Greptile feedback.

2026-01-201

gh-cli

erikpr1994/Jarvis-Code

GitHub CLI patterns for PR reviews, comments, and API operations. Use when working with gh api commands, especially for review threads and comments.

2026-01-191

execute

erikpr1994/Jarvis-Code

Use when following a written plan or task list. Checkpoint verification at each step. Triggers - execute plan, follow plan, implement plan, next step, continue, proceed.

2026-01-191

git-worktrees

erikpr1994/Jarvis-Code

Use for parallel branch development with workspace isolation.

2026-01-191

analytics

erikpr1994/Jarvis-Code

Use when implementing analytics, event tracking, or setting up dashboards. Covers privacy-first tracking, event patterns, and common analytics tools.

2026-01-181

name	debug
description	Use when encountering bugs, test failures, or unexpected behavior. Hypothesis-driven debugging.

Systematic Debugging

Iron Law: NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST.

The Process

1. REPRODUCE  -> Trigger consistently
2. INVESTIGATE -> Gather evidence, trace data flow
3. HYPOTHESIZE -> Form ONE specific theory
4. TEST       -> Minimal change to verify
5. FIX        -> Address root cause
6. VERIFY     -> Confirm with test

Step 1: Reproduce

npm test -- --testPathPattern="auth"
# Document: exact steps, expected vs actual, full error message

Cannot reproduce? Gather more data. Don't guess.

Step 2: Investigate

Read error messages completely - line numbers, stack traces, error codes
Check recent changes: git diff HEAD~5
Trace data flow: Add logging at component boundaries

echo "=== Input to component: $INPUT ==="
# component does its thing
echo "=== Output: $OUTPUT ==="

Goal: Find WHERE it breaks, not just WHAT breaks.

Step 3: Hypothesize

Form ONE specific theory:

"Token is null because login returns before API response completes."

Not: "Something's wrong with auth" (too vague)

Step 4: Test Hypothesis

Smallest possible change:

console.log('Token before return:', token);

Confirmed? Proceed to fix.
Wrong? Form NEW hypothesis. Return to Step 3.

Step 5: Fix Root Cause

Symptom	Wrong Fix	Right Fix
Null pointer	Add null check	Fix why it's null
Timeout	Increase timeout	Fix why it's slow

Step 6: Verify

Write test that fails before fix, passes after. Proves fix works.

Decision Criteria

Situation	Action
Cannot reproduce	More data. Don't guess.
3+ failed attempts	Question architecture. Discuss with user.
"Quick fix" obvious	STOP. Follow process anyway.

Red Flags

"Let me just try..." -> STOP. Investigate first.
"Quick fix for now" -> Root cause first.
Multiple fixes at once -> One hypothesis at a time.

Integration

Pairs with: tdd (write failing test), verification (prove fix works)