# Plugin Patterns

> Canvas plugin architecture patterns, best practices, and implementation templates

- Skill: `canvas-medical/plugin-patterns` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add canvas-medical/plugin-patterns`
- Raw SKILL.md: https://api.skillmd.com/api/skills/canvas-medical/plugin-patterns/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: canvas-medical (https://skillmd.com/u/canvas-medical)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/canvas-medical/plugin-patterns

---


# Canvas Plugin Patterns

This skill provides architectural patterns, best practices, and implementation guidance for building Canvas plugins. It supplements the SDK reference with practical patterns learned from real-world plugins.

## When to Use This Skill

Use this skill when you need:
- Plugin architecture recommendations
- Implementation patterns and templates
- Best practices for testing, error handling, security
- Guidance on plugin complexity decisions
- Common anti-patterns to avoid
- Implementation of:
  - AWS S3
  - LLM (Anthropic Claude, OpenAI ChatGPT, Google Gemini)
  - Twilio
  - SendGrid
  - Extend.ai

## Quick Reference

Reference the `patterns_context.txt` file for detailed patterns and examples.

> **Effects are applied only after the handler returns, as a single batch capped at 64 MB (gRPC).** A handler that emits effects proportional to an unbounded queryset (all providers/patients/appointments) can exceed that ceiling, and the entire batch is **silently dropped** — the plugin logs the work as done but nothing lands. Scaling the worker does not fix this; it is a batching problem. Chunk large fan-out through a queue + cron (one small effect batch per invocation). See the **database-performance** skill (§"Canvas Execution Limits") for the full pattern, plus the write-amplification and over-hydration/memory failure modes that don't show up as N+1.