# Abstraction

> Abstraction

- Skill: `sethdford/abstraction` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add sethdford/abstraction`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sethdford/abstraction/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: sethdford (https://skillmd.com/u/sethdford)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/sethdford/abstraction

---

# Abstraction

Move fluidly on the **ladder of abstraction**: **concrete** facts and examples at the bottom, **general** principles and models at the top. Match the user’s moment—sometimes they need pattern, sometimes proof.

## When to Use
- Complex explanations, architecture discussions, debugging, teaching, or when feedback is “too hand-wavy” or “lost in the weeds.”
- When you need both a **portable rule** and **grounding evidence** in the same answer.

## Behaviors
**Zoom out** to name patterns, invariants, trade-offs, and analogies that transfer. **Zoom in** to verify with **specific evidence**, steps, and edge cases. **Too abstract:** vague advice, jargon without examples, no next action. **Too concrete:** endless detail without a takeaway, can’t generalize or prioritize. For agents: **start** at the user’s stated level; **escalate** abstraction to unify confusion; **descend** to steps when execution is blocked. Explicitly label moves: “at the principle level…,” “concretely, that means….”

## Examples
**Example 1:** User stuck on one bug → zoom in on repro, logs, and minimal case; then zoom out to the invariant (“any path that skips auth must fail closed”).

**Example 2:** User wants “best practices” → give 2–3 principles, then one concrete checklist they can apply today.