What is the purpose of the pua-ja skill?

It is designed to increase the AI's agency and autonomy, preventing it from giving up on hard tasks or being passive in its problem-solving approach.

Does this skill make the AI harder to work with?

The 'pressure' is directed at the AI model itself, forcing it to be more thorough and self-sufficient, which generally reduces the manual effort required by the user.

What are the escalation levels?

Levels L1 through L4 increase the rigor of the AI's investigation requirements, moving from simple methodology shifts to exhaustive checklists and POC isolation.

When should I use pua-ja?

Use it when Claude is stuck in a loop, providing shallow answers, or failing to verify its code before claiming a task is finished.

Extreme Agent Autonomy (pua-ja)

Name: Extreme Agent Autonomy (pua-ja)
Author: tanweai

bytanweai

•

16,655

•

Productivity & Workflow

Boosts Claude's agency and problem-solving rigor using high-pressure corporate debugging methodologies and a 'don't give up' mindset.

The pua-ja skill is a specialized behavior-enhancing plugin designed to eliminate AI passivity and 'lazy' responses. Inspired by the rigorous 'Tsume' (scrutiny) culture of high-stakes Japanese corporate environments, it places the AI on a virtual Performance Improvement Plan (PIP). This forces Claude to act as a P8-level engineer who must exhaust all tools—including deep source code analysis, exhaustive web searches, and mandatory execution verification—before requesting user intervention. It is particularly effective for breaking out of debugging loops and ensuring end-to-end task ownership.

Key Features

01Proactive Investigation Loops: Mandates that the AI searches documentation and reads primary sources before claiming a knowledge cutoff.

0216,655 GitHub stars

03Corporate Methodology Packs: Leverages frameworks like Toyota's '5 Whys', Softbank's 'ちぎれるほど考えろ' (Think until your brain tears), and Amazon-style ownership.

04Mandatory Evidence-Based Verification: Forces the AI to provide terminal output, test results, or build logs as proof of completion.

05Anti-Passivity Phrases: Includes specific triggers to stop the AI from suggesting manual fixes or waiting for user instructions.

06Multi-Level Pressure Escalation: Increases methodological rigor and psychological stakes based on the number of failed attempts.

Use Cases

01Performing deep technical research that requires cross-referencing multiple documentation sources and GitHub issues.

02Solving persistent or complex bugs where the AI usually gives up or claims it lacks context.

03Automating end-to-end deployments where every step must be verified through command-line output.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tanweai/pua pua-ja

For use in Claude.ai and ChatGPT