pua

Detects failure modes and escalates pressure to enforce evidence-first debugging and verification.

Updated Apr 15, 2026
One-click install
npx skills add https://github.com/AoManoh/project_templates --skill pua-aomanoh
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: pua
Source: https://github.com/AoManoh/project_templates/tree/main/skills/pua
Command: npx skills add https://github.com/AoManoh/project_templates --skill pua-aomanoh

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill prevents the AI from low-effort “giving up” behavior by enforcing an evidence-first, failure-mode-driven, high-pressure execution loop during complex development, debugging, review, and risky operations.

Core Features & Use Cases

  • Failure-mode detection & forced method shift: identifies patterns like looping, blame-shifting, or “no-search guessing,” then switches to genuinely different investigation or implementation paths.
  • PUA-DIAGNOSIS gate before action: when failures happen or validation is questioned, it outputs a structured diagnosis (problem + evidence + next action) to ensure every step generates new information.
  • Pressure escalation with verification enforcement: increases intensity across L1–L4 escalation levels and requires concrete verification before claiming completion, especially after repeated failures.

Use case examples:

  • You repeatedly hit the same bug and the AI keeps tweaking superficial parameters—use pua to force deeper evidence collection and alternate hypotheses.
  • You’re close to “done” but no solid validation was run—use pua to require real commands/tests and a proof of correctness.
  • A task is high-risk or spans multiple files—use pua to proactively assess boundaries without delegating verification back to the user.

Quick Start

Ask the AI to solve your failing task with strong evidence and verification, for example: “Use pua to diagnose the failure, switch methods if it’s looping, and run the required tests before declaring it fixed.”

Frequently Asked Questions about pua

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I stop AI from giving up during complex debugging and force deeper evidence collection?

To stop AI from giving up during debugging, enforce an evidence-first execution loop that detects failure modes like looping and mandates structured diagnostics with concrete verification before completion. This forces method shifts and deeper investigation.

What is evidence-first failure recovery in AI code generation?

Evidence-first failure recovery in AI code generation is a mechanism requiring structured problem diagnosis, real test verification, and forced method shifts across escalating pressure levels before declaring a task complete.

How do I verify high-risk cross-file tasks without delegating validation back to the user?

To verify high-risk cross-file tasks without user delegation, apply a pressure escalation framework that proactively assesses boundaries and mandates real commands and tests to prove correctness before claiming completion.

Why does the AI keep tweaking superficial parameters when hitting the same bug repeatedly?

The AI keeps tweaking superficial parameters due to low-effort failure loops, which you can break by enforcing failure-mode detection that identifies superficial tweaks and forces a switch to genuinely different investigation paths.

Does this approach work for git operations and code review or just debugging?

This approach works for git operations, code review, and debugging. It enforces evidence-first diagnostics and verification across any non-trivial development or risky task where repeated failure indicates wasted loops.

What are the limitations of using AI governance for failure recovery in software engineering?

Limitations of AI governance for failure recovery include the strict requirement to generate structured PUA-DIAGNOSIS outputs at every failure, increasing operational overhead for trivial tasks not exhibiting wasted looping behaviors.