agentsclimarketplace

Extensible by design

Skill shuymn/skills/skills/extensible-by-design

Personal agent skills

Install
npx -y skills add shuymn/skills --skill extensible-by-design

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Design philosophy for AI-native software that agents and humans can extend without bloating core. Use when designing new products, reviewing architecture for extensibility, deciding what belongs in core vs extensions, choosing between code and instruction layers, evaluating whether AI self-modification loops are possible, or when the user asks to make software self-extensible, avoid feature bloat, or apply extensibility-by-design principles.

SKILL.md

5.2 KB, as published. Nobody here has run it

Extensible by Design

Software is not a finished artifact — it is a platform that AI and humans extend over time. The design question shifts from "what features to build" to what to leave out of core and where to expose extension surfaces.

This skill is application-agnostic. It captures transferable design philosophy, not any single harness or framework.

When to Apply

  • Greenfield product or subsystem design
  • Architecture review: "Is core too fat? Are extension points missing?"
  • Choosing how to add a capability (core, plugin, config, playbook, hook)
  • Evaluating whether AI self-modification loops are possible

Quick start

User: "Review our order API for extensibility."

  1. Frame — core = order CRUD + auth; extensions = none; discounts hardcoded in service layer.
  2. Score — rate all eight principles STRONG / PARTIAL / MISSING with one sentence each.
  3. Lens — load a reference only when needed (e.g. intervention-points.md for hook placement).
  4. Deliver — scorecard + prioritized recommendations using review-output-format.md.

Eight Principles

#PrincipleOne-line test
1Minimal core, maximal extension surfaceCore runs with zero extensions; extensions add depth, not prerequisites
2Negative space defines the contract"We do not build X in core" is an explicit design decision
3Separate what runs from what guidesExecution (code) and judgment (instructions) live on different planes
4Context is a budgetOnly summaries stay resident; details load on demand
5Hooks, not hooks everywhereNamed pipeline stages with clear precedence — not arbitrary interception
6Fit-for-purpose configuration boundariesConfiguration scopes match the application's ownership model; provenance resolves conflicts when multiple scopes are justified
7Self-modification loopAn agent can author an extension, apply it, and verify without a release cycle
8Exploration over linearityBranching, retry, and alternate paths are first-class state

Workflows

Run when the user asks for an extensibility review or during design discussions.

Step 1 — Frame the system

Identify: core runtime, primary user/agent loop, existing extension mechanisms (if any), and what is currently hardcoded.

Step 2 — Score each principle

For each principle, assign STRONG, PARTIAL, or MISSING with one sentence of evidence.

For existing systems: STRONG = the system exhibits the principle in its current implementation; PARTIAL = partially present but incomplete or inconsistent; MISSING = absent.

For greenfield designs (not yet built): STRONG = the design specifies a concrete mechanism that implements the principle; PARTIAL = acknowledged in the design but the mechanism is a placeholder or underspecified; MISSING = the principle is not addressed in the design at all.

Step 3 — Apply decision lenses

Load reference files only when a lens needs depth.

LensQuestionReference
Negative spaceWhat should explicitly not live in core?negative-space-checklist.md
Two planesIs this execution logic or judgment guidance?code-vs-instructions.md
InterventionAt which pipeline stage should extensions act?intervention-points.md
Configuration scopeWhich configuration boundaries are actually justified here?configuration-scope.md

Step 4 — Recommend

For each PARTIAL or MISSING principle, propose one concrete change. Prefer:

  • Declaring negative space over adding core features
  • A new hook or justified configuration boundary over a hardcoded branch
  • An instruction-plane artifact over code when judgment changes often

Format output per review-output-format.md. Do not recommend copying any specific product's API — translate principles to the user's domain.

Quick Decision Guide

Put in core only when ALL are true:

  • Required for every session with no extension installed
  • Wrong or unsafe if left to third parties without guardrails
  • Stable contract unlikely to need per-user or per-team variation

Prefer extension surface when ANY are true:

  • Domain teams or AI agents will customize behavior
  • The capability is optional or competes with alternatives (MCP vs native plugin vs script)
  • Judgment rules change faster than execution infrastructure
  • Users need to fork, branch, or experiment without breaking others

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.