What are the complexity levels in Quality Gates?

Tasks are scored from Level 1 (trivial tasks taking <30 mins) to Level 5 (very complex tasks involving 25+ files and 24+ hours of work), helping the AI determine if a task requires a detailed plan.

When does the skill block progress?

It blocks progress if there are more than 3 unanswered critical questions, missing dependencies, or if the AI has failed the same task 3 times consecutively.

How does stuck detection work?

The skill monitors attempt counts; after 3 failures, it stops the loop and generates an escalation report including what was tried and specific questions for the user.

What is the benefit of the LLM-as-judge pattern?

It uses secondary validation logic to score generated artifacts for relevance, coherence, and accuracy, ensuring the output meets production-grade standards before completion.

Quality Gates & Complexity Assessment

Name: Quality Gates & Complexity Assessment
Author: yonatangross

byyonatangross

•

Productivity & Workflow

Enforces rigorous quality checks and complexity scoring to prevent wasted effort on poorly defined or overly complex AI coding tasks.

The Quality Gates skill empowers Claude to act as a self-regulating developer by assessing task difficulty on a standardized 1-5 scale before implementation begins. It establishes clear blocking thresholds—such as requiring clarification for missing dependencies or escalating after three failed attempts—to ensure that the AI never proceeds with incomplete information or enters inefficient loops. By integrating LLM-as-judge patterns and complexity breakdown workflows, it transforms vague requests into actionable, high-quality development cycles while providing human-in-the-loop escalation when tasks exceed predefined difficulty limits.

Key Features

01Standardized 1-5 complexity scoring for effort estimation and planning.

02Stuck detection that triggers escalation protocols after three failed attempts.

0327 GitHub stars

04Automated blocking thresholds for missing dependencies or unanswered questions.

05Automated task decomposition for Level 4 and 5 complex features.

06LLM-as-judge validation for code relevance, depth, and accuracy.

Use Cases

01Pre-implementation planning to identify blockers before writing code.

02Recovering from 'stuck' states during difficult debugging or refactoring sessions.

03Ensuring requirement completeness before starting work in multi-agent Squad mode.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/skillforge-claude-plugin quality-gates

For use in Claude.ai and ChatGPT

Download Skill