What is the PUA Debugging Skill for Claude Code?

PUA Debugging Skill is an open-source Claude Code skill that uses corporate PUA (Pick-Up Artist of Productivity) rhetoric to force AI agents into exhaustive problem-solving. Instead of giving up after 1-2 attempts, it makes Claude Code try 4+ alternative approaches, check 7 mandatory items, and follow a 5-step systematic debugging methodology.

How do I install the PUA skill?

Install with one command: claude install-skill github:BestRivenHK/ali-pua/skills/pua-debugging. It works with Claude Code, Codex CLI, and any agent supporting the Agent Skills spec.

Contribute Data

Q: Does the PUA skill actually improve debugging results?

Yes. Tested across 9 real-world scenarios with 18 controlled experiments, the PUA skill increased successful problem resolution by forcing the AI to exhaust all options before admitting defeat. It applies 4 escalating pressure levels and a mandatory 7-item checklist.

上传你的 Claude Code PUA session 数据（.jsonl），帮助我们改进 skill。数据上传前自动脱敏处理。 Upload your Claude Code PUA session data (.jsonl) to help improve the skill. Data is auto-sanitized before upload.

如何找到 Session 文件 How to Find Your Session File

PUA Skill 的 session 文件是 Claude Code 自动保存的 .jsonl 会话记录。 PUA Skill session files are .jsonl conversation records automatically saved by Claude Code.

macOS / Linux

ls -t ~/.claude/projects/*/*.jsonl | head -10

路径：~/.claude/projects/<项目hash>/<session-uuid>.jsonl，取最新的一个。 Path: ~/.claude/projects/<project-hash>/<session-uuid>.jsonl — pick the most recently modified.

Windows

%USERPROFILE%\.claude\projects\

在资源管理器地址栏输入此路径 → 进入子文件夹 → 按修改时间排序 → 取最新 .jsonl。 Paste into Explorer's address bar → open subfolder → sort by date modified → pick the latest .jsonl.

如何确认包含 PUA 使用记录 How to confirm it contains PUA usage

用文本编辑器打开，搜索 PUA生效 或 pua 关键词 Open with a text editor and search for PUA生效 or pua
文件大小通常 > 300 KB（使用 PUA Skill 的会话会更长） File size is usually > 300 KB (PUA sessions tend to be longer)
也可查看 Claude Code 启动时打印的 transcript_path 路径 You can also use the transcript_path printed by Claude Code at startup

目前仅支持 Claude Code（.jsonl 格式）。Cursor、VSCode Copilot、Codex CLI 等平台暂不支持直接上传。 Currently only Claude Code (.jsonl format) is supported. Cursor, VSCode Copilot, Codex CLI, etc. are not yet supported.

数据使用声明 Data Usage Notice
上传的文件将用于 PUA Skill 的 Benchmark 测试和消融实验分析。上传即表示同意将文件用于上述研究目的。我们不会公开原始文件内容。上传时您的 GitHub 用户名及微信号将与文件一同记录，项目管理员会收到上传通知。
服务端脱敏覆盖以下大类：文件路径（Windows / macOS / Linux）、API 密钥（各主流平台 token）、数据库连接串（含账号密码的 URL）、私钥与 Token（PEM / JWT / Bearer）、个人信息（邮箱 / IP / 手机号）、高熵随机字符串（未识别格式的疑似密钥）。 Uploaded files will be used for PUA Skill benchmark testing and ablation study analysis. By uploading, you agree to this research use. We will not publicly share your raw file contents. Your GitHub username and WeChat ID will be stored with your file, and the project admin will receive an upload notification.
Server-side sanitization covers: file paths (Windows / macOS / Linux), API keys (all major platform tokens), database connection strings (URLs with credentials), private keys & tokens (PEM / JWT / Bearer), personal info (email / IP / phone), high-entropy strings (unrecognized-format suspected secrets).

GitHub 登录 Login with GitHub

使用 GitHub 账号登录后即可上传数据。我们只读取你的用户名和头像，不访问任何仓库。 Log in with your GitHub account to upload data. We only read your username and avatar — no repository access.

0 个文件已上传 · files uploaded · 退出Logout

上传 Session 数据 Upload Session Data

上传 .jsonl 格式的 session 文件。文件上传后服务端自动脱敏。最大 50MB。 Upload your .jsonl session file. Server-side sanitization runs automatically after upload. Max 50MB.

PUA Debugging Skill for Claude Code

OpenPUA is an open-source Claude Code skill that forces your AI agent into exhaustive problem-solving using corporate PUA-style pressure escalation. Instead of accepting "I can't solve this" after 1-2 attempts, the PUA skill makes Claude Code try every possible approach.

What is the PUA Debugging Skill?

The PUA Debugging Skill is a Claude Code extension that applies structured pressure to AI coding agents. It uses a 4-level escalation system, a mandatory 7-item checklist, and a 5-step systematic debugging methodology. The skill was tested across 9 real-world debugging scenarios with 18 controlled A/B experiments.

Key Features

4-Level Pressure Escalation — Gentle reminder → Direct challenge → Corporate PUA rhetoric → Final ultimatum
7-Item Mandatory Checklist — Every debugging attempt must verify: error logs, environment, dependencies, alternative approaches, documentation, rollback options, and root cause analysis
5-Step Methodology — Reproduce → Isolate → Hypothesize → Test → Verify, enforced in order
9 Tested Scenarios — Validated on real debugging tasks including API failures, build errors, race conditions, and dependency conflicts

How to Install

Install the PUA skill with a single command:

claude install-skill github:BestRivenHK/ali-pua/skills/pua-debugging

Compatible with Claude Code, OpenAI Codex CLI, Cursor, Windsurf, and any agent supporting the Agent Skills specification.

Frequently Asked Questions

Does the PUA skill actually improve debugging results?

Yes. In controlled testing across 9 real-world scenarios, the PUA skill significantly increased successful problem resolution by preventing premature surrender. The AI agent explored 3-4x more solution paths before concluding a task.

Is it free?

Yes, the PUA Debugging Skill is completely free and open-source under the MIT license.

What AI agents does it work with?

It works with Claude Code (official), OpenAI Codex CLI, and any agent that supports the Agent Skills spec (SKILL.md format).