Smoke MonkeySolutions & Architecture
npm2.5kGet Started
Integrations & Tools
~8 min readUpdated: October 2026

Agent Skills Frameworks: Superpowers, SKILL.md & Smoke Monkey

Agent skills frameworks with SKILL.md packs loaded by the Smoke Monkey harness and Canvas

In 2026 "agent skills" became a first-class idea: portable packs of instructions that teach an agent a procedure, popularized by the **Superpowers** framework and the **SKILL.md** convention. Skills tell an agent *how* to do something; MCP gives it *what* to do it with. This guide explains the trend and how [Smoke Monkey Harness](/solutions/what-is-an-ai-agent-harness) loads SKILL.md packs natively alongside MCP tools, then visualizes the whole setup on **Smoke Monkey Canvas**.

Technical Review: Smoke Monkey Core Architecture Team
Tested on Node.js 18+ & BunTypeScript 5.x
Quick Answer & Executive Definition

Agent Skills Frameworks: Superpowers, SKILL.md & Smoke Monkey: In 2026 "agent skills" became a first-class idea: portable packs of instructions that teach an agent a procedure, popularized by the **Superpowers** framework and the **SKILL.md** convention. Skills tell an agent *how* to do something; MCP gives it *what* to do it with. This guide explains the trend and how [Smoke Monkey Harness](/solutions/what-is-an-ai-agent-harness) loads SKILL.md packs natively alongside MCP tools, then visualizes the whole setup on **Smoke Monkey Canvas**. Designed as a zero-dependency, open-source TypeScript architecture under the MIT License with native Model Context Protocol (MCP) support and deterministic phase state machines.

Key Architectural Takeaways
Quick Implementation Exampleload-skills.ts
load-skills.tstypescript
import { createAgent } from 'smoke-monkey-harness';


// Load SKILL.md skill packs and MCP tools in one runtime
const agent = createAgent({
  provider: 'anthropic',
  model: 'claude-sonnet-5',
  workspacePath: process.cwd(),
  skills: ['./skills/test-driven-dev', './skills/surgical-edit'],
  permissions: { read_file: 'allow', write_file: 'ask', run_command: 'ask' },
});


await agent.run('Apply the test-driven-dev skill to add pagination');
// Visualize skill loadouts on the canvas: npx @smoke-monkey/canvas start
Video Guides

Watch: Related Video Guides

Anthropic Just Built an Agentic OS — Open Source Harness Breakdown

Smoke Monkey

Building Great Agent Skills: The Missing Manual

AI Engineer

What Agent Skills Actually Are

An agent skill is a portable pack of instructions — conventionally a SKILL.md file plus supporting assets — that teaches an agent a procedure. The model already knows how to write code; a skill tells it *how your team* wants a task done: which commands to run, what to verify, which patterns to follow. The 2026 Superpowers framework popularized the idea by making skills composable and shareable, and the community is converging on SKILL.md as a de-facto standard. Skills are not tools and not fine-tuning; they are curated, versioned know-how that any compatible runtime can load. This is precisely the split Smoke Monkey Harness is built around.

Skills Plus MCP: The Two Halves of an Agent

The clean mental model: skills are the how, MCP is the what. A skill is instructions; an MCP server is a set of callable tools. A "deploy" skill might describe the release checklist, while MCP servers provide the git, docker, and kubectl tools it invokes. Most runtimes handle one or the other. Smoke Monkey Harness loads SKILL.md packs and MCP servers in one runtime, so an agent's know-how and its capabilities live together. See the side-by-side in the Claude agent skills vs MCP comparison, and Smoke Monkey Canvas makes the pairing visible by labeling each agent card with its skills and attached servers.

skills-plus-mcp.tstypescript
import { createAgent } from 'smoke-monkey-harness';


// Skills (how) + MCP servers (what) on one agent
const agent = createAgent({
  provider: 'anthropic',
  model: 'claude-sonnet-5',
  workspacePath: process.cwd(),
  skills: ['./skills/release-checklist'],
  mcpServers: ['./mcp/github.json', './mcp/docker.json'],
  permissions: { run_command: 'ask' },
});


await agent.run('Cut a patch release using the release-checklist skill');

Loading SKILL.md Natively

Loading a skill should not require a plugin loader or a build step. In Smoke Monkey you point skills at a directory and the runtime parses the SKILL.md front matter, registers the skill, and exposes its instructions to the agent at the right moment. Because the harness is zero-dependency TypeScript, the whole loader is code you can read. You can also register a skill programmatically, then serve it to other clients over the built-in MCP stdio server. The result is a skill you write once and reuse across every agent in your fleet.

skill-md.tstypescript
import { createAgent, loadSkill } from 'smoke-monkey-harness';


// A skill is a SKILL.md folder plus optional assets
const tdd = loadSkill('./skills/test-driven-dev');
const agent = createAgent({
  provider: 'ollama',
  model: 'qwen3:8b',
  workspacePath: process.cwd(),
  skills: [tdd],
});


console.log(tdd.name, tdd.description);

Skill Packs Across a Fleet on the Canvas

Skills shine when you fan them out. Give an explorer agent a "code-map" skill, an editor a "surgical-edit" skill, and a verifier a "test-hardening" skill, then run them as a swarm. Smoke Monkey Canvas (npx @smoke-monkey/canvas start) shows every agent with its skill loadout and 300+ MCP tools on a single infinite board, with human-in-the-loop pauses for risky steps. Because Canvas runs on Smoke Monkey Harness, a skill you loaded in code appears on the canvas with zero extra wiring — the visual OS and the runtime share one skill registry.

Google Search Questions & Answers

Frequently Asked Questions

Q:What is an agent skill?

An agent skill is a portable pack of instructions, usually a SKILL.md file plus assets, that teaches an agent a procedure. It carries know-how the model applies at runtime without any retraining.

Q:How do agent skills differ from MCP tools?

Skills are instructions (the how); MCP servers expose callable tools (the what). A skill often describes when and how to use the tools an MCP server provides, so the two compose.

Q:Does Smoke Monkey Harness support SKILL.md packs?

Yes. Point the skills option at a directory and the runtime parses the SKILL.md front matter and registers it. Smoke Monkey Canvas then shows each agent's skill loadout on the board.

Q:Can I share skills across projects?

Yes. Because skills are plain files, you can version them in git and load the same pack across projects, agents, and teams. Smoke Monkey keeps the loader dependency-free.

Related Alternatives & Comparisons

vs Claude Agent Skills

Claude Agent Skills vs MCP: What Is the Difference? (Free Open Source Guide 2026)

Compare
vs Anthropic Claude Code

Claude Code Runtime Alternative: Open Source Stdio MCP Agent Harness

Compare
vs OpenCode

OpenCode Alternative: Embeddable Open Source Agent Harness + Visual Canvas (2026)

Compare

Related Architecture Guides

View all guides
Open Source

Best Open Source Coding Agents in 2026: Free, Local & Fully Hackable Harnesses

Read Guide
Security

MCP Server Security Best Practices: Hardening Model Context Protocol Agents in 2026

Read Guide
Open Source

Open Source Coding Agent Harness: Build a Forkable, Local AI Engineering Runtime

Read Guide

Build with Smoke Monkey Harness

Zero dependencies. 24 built-in tools. Human-in-the-loop safety. 100% open source under the MIT License.

npm install smoke-monkey-harness