quality-gates_skill

Overview

This skill helps agents assess task complexity, enforce quality gates, and avoid wasted work on ill-defined or overly ambitious tasks. It produces a 1-5 complexity score, gate status (PASS/WARNING/BLOCKED), and recommended escalation or decomposition actions. Use it to catch unknowns early and require plans for high-complexity work.

How this skill works

The skill runs a pre-task gate that evaluates justified complexity, unanswered critical questions, blocked dependencies, and failed attempt counts. It calculates a complexity score (1-5) and applies blocking thresholds such as YAGNI ratio, unanswered questions > 3, missing dependencies, or 3+ failed attempts. It returns a gate status, evidence checks, and escalation templates when blocking or warning conditions are met.

When to use it

• On receiving a new task assignment or before starting implementation
• When planning multi-step or Level 4-5 features that need decomposition
• After 2-3 failed attempts or when repeatedly stuck on a task
• Before allocating work across a squad or committing to a timeline
• When requirements or dependencies seem unclear or incomplete

Best practices

• Run YAGNI check first: block if justified complexity ratio > 2.0, warn if 1.5-2.0
• Count and document critical questions; block if > 3 unanswered
• Track attempt history and escalate automatically at 3 failed attempts
• Require a written plan before proceeding on Level 4-5 tasks
• Integrate gate outcomes into context and evidence systems for auditability

Example use cases

• Pre-task validation to decide whether a feature can proceed or must be simplified
• Automatic escalation when an agent fails three times on the same approach
• Code review integration to surface anti-patterns and YAGNI violations
• Breaking a complex Epic into Level 1-3 subtasks with acceptance gates
• LLM-as-judge quality checks for relevance, coherence, and test evidence