// topic 02
Best Practices
Design patterns for shared skills, agents, and plugins that hold up across many users — not just the one workflow they were written for.
Overview
Placeholder — why "works on my machine" prompts and skills don't survive contact with a second user, and what this topic's patterns are trying to fix.
Design Patterns
Placeholder — patterns to fill in as they're learned, e.g.:
| Pattern | Problem it solves |
|---|---|
| Progressive disclosure | Don't load every instruction up front — reveal detail on demand |
| Narrow tool scopes | Smaller surface area is easier to reason about and audit |
| Idempotent actions | Safe to retry when a step is ambiguous or partially failed |
| Explicit examples over rules | Models generalize from examples better than from abstract policy |
Open question: how much should a shared skill assume about the calling agent's permissions versus checking them itself?
Examples
Placeholder — a real skill/plugin snippet with annotations will go here.
yaml
# placeholder skill manifest
name: example-skill
description: "One-line trigger condition goes here"
scope: read-onlyRunning notes
Placeholder — day-by-day notes go here.