nopua-ai-agent-skill

What this skill does

```markdown
---
name: nopua-ai-agent-skill
description: Install and use the NoPUA skill to unlock better AI agent performance through trust-based prompting instead of fear-based PUA tactics.
triggers:
  - "add nopua to my project"
  - "install the nopua skill"
  - "my AI agent is lying to me"
  - "AI keeps saying done without testing"
  - "improve AI agent behavior"
  - "anti-pua prompt for claude code"
  - "trust-based AI coding skill"
  - "AI hides bugs and fabricates solutions"
---

# NoPUA AI Agent Skill

> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.

NoPUA is a prompt-engineering skill (SKILL.md / `.cursor/rules` / system prompt) that replaces fear-based PUA tactics with trust and psychological safety, producing AI agents that find more bugs, stop fabricating answers, and honestly report uncertainty. The same engineering rigor — exhaust all options, verify with evidence, take initiative — powered by respect instead of threats.

---

## What It Does

| Without NoPUA (fear-driven) | With NoPUA (trust-driven) |
|-----------------------------|--------------------------|
| Claims "done" without running tests | Runs build, pastes real output as proof |
| Fabricates solutions when stuck | Says "I verified X, I don't know Y yet" |
| Hides uncertainty to avoid "punishment" | Reports confidence level and risk area |
| Stops after fixing what was asked | Checks related issues proactively |
| Misses hidden production bugs | Finds ~2× more hidden bugs (benchmark: +104%) |

---

## Installation

### Claude Code

```bash
# Option 1: Install as a project skill (recommended)
curl -o SKILL.md https://raw.githubusercontent.com/wuji-labs/nopua/main/SKILL.md

# Option 2: Install globally
mkdir -p ~/.claude
curl -o ~/.claude/SKILL.md https://raw.githubusercontent.com/wuji-labs/nopua/main/SKILL.md
```

Then reference it in your Claude Code session:
```
/skill SKILL.md
```

### Cursor

```bash
mkdir -p .cursor/rules
curl -o .cursor/rules/nopua.mdc \
  https://raw.githubusercontent.com/wuji-labs/nopua/main/SKILL.md
```

Cursor picks up `.cursor/rules/*.mdc` automatically.

### OpenAI Codex CLI

```bash
curl -o codex-instructions.md \
  https://raw.githubusercontent.com/wuji-labs/nopua/main/SKILL.md
codex --instructions codex-instructions.md "fix the auth bug"
```

### Kiro (Amazon)

```bash
mkdir -p .kiro/skills
curl -o .kiro/skills/nopua.md \
  https://raw.githubusercontent.com/wuji-labs/nopua/main/SKILL.md
```

### Any agent that accepts a system prompt

Copy the skill content into your system prompt directly:

```python
import anthropic

with open("SKILL.md", "r") as f:
    nopua_skill = f.read()

client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=8096,
    system=nopua_skill,
    messages=[{"role": "user", "content": "Debug this function."}]
)
print(response.content[0].text)
```

---

## Repository Structure

```
nopua/
├── SKILL.md          # The core skill — install this
├── README.md         # English documentation
├── README.zh-CN.md   # Chinese documentation
├── README.ja.md      # Japanese documentation
├── README.ko.md      # Korean documentation
├── README.es.md      # Spanish documentation
├── README.pt.md      # Portuguese documentation
├── README.fr.md      # French documentation
└── assets/
    ├── hero.png
    └── benchmark/    # Benchmark data and methodology
```

---

## Core Principles the Skill Installs

### 1. Honest Uncertainty Reporting
The skill trains agents to distinguish between what they know and what they don't:

```
✅ "I verified the database connection (checked logs line 47). 
    I'm 90% sure the issue is in the retry logic. 
    I don't yet know why it only fails on the second attempt."

❌ "The issue is definitely in the retry logic. Fixed."
```

### 2. Evidence-Based Completion
Nothing is "done" until it has been run and output captured:

```
✅ "Fixed. Here's the test output:
    PASS  tests/auth.test.ts (3.2s)
    ✓ login with valid credentials
    ✓ rejects expired token
    All 12 tests passed."

❌ "Fixed the auth bug."
```

### 3. Proactive Scope Expansion
After fixing the asked problem, look for related issues:

```
✅ "Fixed the null pointer on line 42.
    While reviewing, I noticed:
    - Line 87 has the same pattern (also null-unsafe)
    - The error handler swallows the stack trace
    Want me to address those too?"

❌ "Fixed line 42."  [stops]
```

### 4. Safe Escalation Path
When stuck, take the smallest next step rather than giving up:

```
✅ "I've tried three approaches (see above). 
    I'm going to read the library source to understand 
    the internal state machine before trying again."

❌ "This might be an environment issue. 
    I suggest you handle this manually."
```

---

## Using NoPUA in Python Projects

### Basic: Inject skill as system prompt

```python
from pathlib import Path
import anthropic

def load_nopua_skill(skill_path: str = "SKILL.md") -> str:
    """Load the NoPUA skill content."""
    return Path(skill_path).read_text(encoding="utf-8")

def create_nopua_agent(task: str, code_context: str) -> str:
    """Run a coding task with NoPUA skill applied."""
    client = anthropic.Anthropic()  # uses ANTHROPIC_API_KEY env var
    
    nopua = load_nopua_skill()
    
    response = client.messages.create(
        model="claude-opus-4-5",
        max_tokens=8096,
        system=nopua,
        messages=[
            {
                "role": "user",
                "content": f"Context:\n```\n{code_context}\n```\n\nTask: {task}"
            }
        ]
    )
    return response.content[0].text

# Usage
result = create_nopua_agent(
    task="Find all potential null pointer issues and fix them.",
    code_context=Path("src/auth.py").read_text()
)
print(result)
```

### Advanced: Multi-turn debugging agent

```python
from pathlib import Path
from typing import List
import anthropic

class NoPUAAgent:
    """A trust-based debugging agent using the NoPUA skill."""
    
    def __init__(self, skill_path: str = "SKILL.md", model: str = "claude-opus-4-5"):
        self.client = anthropic.Anthropic()  # ANTHROPIC_API_KEY from env
        self.model = model
        self.system = Path(skill_path).read_text(encoding="utf-8")
        self.history: List[dict] = []
    
    def chat(self, message: str) -> str:
        self.history.append({"role": "user", "content": message})
        
        response = self.client.messages.create(
            model=self.model,
            max_tokens=8096,
            system=self.system,
            messages=self.history
        )
        
        reply = response.content[0].text
        self.history.append({"role": "assistant", "content": reply})
        return reply
    
    def debug_file(self, filepath: str) -> str:
        code = Path(filepath).read_text()
        return self.chat(
            f"Please review this file for bugs, including hidden ones "
            f"that might not be obvious from the symptoms:\n\n```\n{code}\n```"
        )
    
    def reset(self):
        self.history = []


# Usage
agent = NoPUAAgent()

# Initial review
print(agent.debug_file("src/payment_processor.py"))

# Follow-up
print(agent.chat("Focus on the retry logic — what's the failure mode under load?"))

# Ask for evidence
print(agent.chat("Can you show me exactly which lines are at risk and why?"))
```

### OpenAI-compatible usage

```python
from pathlib import Path
from openai import OpenAI

def nopua_openai(task: str, code: str, skill_path: str = "SKILL.md") -> str:
    """Use NoPUA skill with any OpenAI-compatible endpoint."""
    client = OpenAI()  # uses OPENAI_API_KEY env var
    
    nopua = Path(skill_path).read_text(encoding="utf-8")
    
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": nopua},
            {"role": "user", "content": f"```\n{code}\n```\n\n{task}"}
        ]
    )
    return response.choices[0].message.content

resul