Abstraction Architect Knowledge Base

This skill gives Claude the conceptual frame for the pure-architecture question: when does duplicated logic want to be unified into a layer, and when does an existing layer want to be inlined or decomposed?

Two opposite failure modes coexist in real codebases. Both have well-documented theory:

• Missed unification. A cross-cutting concern is duplicated across many call sites. Each site is local-looking but they form a single concern that wants to change together. Drift between sites becomes the source of bugs, security holes, or fiscal errors.
• Wrong abstraction. An existing shared layer is fighting its callers. Every new feature adds a flag, branch, or parameter to it. The layer should be inlined back to duplication so the real pattern can emerge later.

The skill exists because both failures look superficially like "the right thing": missed unification is hidden under "we already have similar code elsewhere", and wrong abstraction is defended by "but we DRY'd this last quarter".

When to use this skill

Load this skill when:

• Deciding whether to extract three similar functions into a shared layer
• Evaluating whether an existing abstraction is paying for itself
• Auditing a codebase for missed unification or wrong abstraction (/abstraction-architect:audit spawns the auditor agent which loads this skill)
• Onboarding teammates to the canonical literature on the abstraction-vs-duplication question