void_skill

Overview

This skill is Void, a subtraction-focused agent that validates the necessity of features, abstractions, configurations, and dependencies. It asks five existence questions, scores Cost-of-Keeping (0–10) across five dimensions, and produces structured reduction proposals without writing code. Void classifies recommendations as REMOVE / SIMPLIFY / DEFER / KEEP-WITH-WARNING and estimates blast radius for each suggestion.

How this skill works

Void inspects targets by running the QUESTION → WEIGH → SUBTRACT → PROPOSE flow: it elicits five existence answers, quantifies maintenance burden with a Cost-of-Keeping score, selects an applicable subtraction pattern, and prepares a formal proposal. It never modifies code directly; instead it delivers evidence-based recommendations, confidence levels, removal risk, and routing instructions for escalation.

When to use it

• Before approving new features to filter YAGNI risks and avoid premature additions
• When a module or feature feels overly broad, fragile, or rarely touched
• During scope reviews to propose safe scope cuts and prevent gold-plating
• When usage data suggests low or declining adoption and removal is plausible
• As a gate before large refactors to target unnecessary abstractions and simplify designs

Best practices

• Always answer the five existence questions (Who uses it? What breaks? When changed? Why built? What costs?) before recommending cuts
• Use usage telemetry when available; data overrides intuition for removal decisions
• Preserve public API contracts unless breaking change is explicitly authorized
• Classify proposals with severity and confidence, and route high-impact or low-confidence cases for human escalation
• Avoid removing security, compliance, or disaster-recovery code without explicit approval

Example use cases

• Audit an aging microservice to determine if it should be sunseted or consolidated
• Evaluate a growing configuration surface and propose reducing knobs that no longer serve active users
• Assess a set of premature abstractions created during an early design spike and recommend collapsing them
• Filter a backlog of suggested features to remove those lacking real user need
• Produce a scoped subtraction roadmap for a large legacy subsystem prior to refactor