How do I install the Classify Agent skill?

Run npx skillmds@latest add m2ai-portfolio/classify-agent in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Classify Agent safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Classify Agent?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Classify Agent free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Classify Agent?

m2ai-portfolio (@m2ai-portfolio) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Classify Agent by m2ai-portfolio — Agent Skill

Name: Classify Agent
Author: m2ai-portfolio

Classify Agent

Classify a problem into one of four agent architectures: coding harness, dark factory, auto research, or orchestration framework. Implements the "one-question test" diagnostic from Nate Kadlac's agent taxonomy. Use when starting a new agent project, evaluating a build approach, or when someone says "what kind of agent should I build?", "classify this", or "/classify-agent".

m2ai-portfolio 87b795b 2 files · 3.8 KB Updated Sep 17, 2026

File contents

Agent Architecture Classifier

Takes a problem description and determines which of four agent architectures is the best fit, using the diagnostic question: "What are you optimizing against?"

The Four Architectures

Architecture	Governing Principle	Optimizing Against	Example
Coding Harness	Decomposition boundaries	Task isolation -- each unit of work touches minimal shared state	Claude Code, Cursor, Aider
Dark Factory	Specification-as-code	Convergence to spec -- can an agent reach the target without human input?	StrongDM's autonomous build pipeline
Auto Research	Metric + guardrail	Measurable improvement -- is there a benchmark to beat?	Karpathy's 630-line research loop
Orchestration Framework	Handoff contracts	Pipeline throughput -- do agents pass clean data to each other?	CrewAI, LangGraph, multi-agent chains

Phases

Phase 1: Gather Context

Ask the user to describe their problem in 2-3 sentences. If they've already provided it, skip this.

Phase 2: The One-Question Test

Analyze the description against the diagnostic: "What are you optimizing against?"

Map the answer:

"I need isolated tasks that don't step on each other" -> Coding Harness
"I need an agent that converges to a precise spec without hand-holding" -> Dark Factory
"I need to beat a benchmark or improve a measurable metric" -> Auto Research
"I need multiple agents to coordinate and pass work between them" -> Orchestration Framework

Phase 3: Confirmation Diagnostics

Run three follow-up questions to confirm the classification:

Decomposition test: "Can the work be split into units that don't share files/state?" (Yes = Coding Harness signal)
Spec maturity test: "Do you have a precise, testable specification?" (Yes = Dark Factory signal)
Metric test: "Is there a quantitative benchmark you're trying to beat?" (Yes = Auto Research signal)
Handoff test: "Does the output of one step become the input of another step with a different agent?" (Yes = Orchestration signal)

Phase 4: Output Classification

Report:

Primary architecture: The best fit with confidence (high/medium/low)
Runner-up: Second-best fit if signals are mixed
Anti-patterns to avoid: Common mistakes for this architecture type
Governing principle: The one rule that should guide all design decisions
Next step: What to build first

Verification

Classification maps to exactly one of the four architectures
At least 2 of 4 diagnostic signals align with the primary classification
Anti-patterns are specific to the chosen architecture, not generic advice

Source

Nate Kadlac, "There are 4 kinds of agents and you're probably using the wrong one", natesnewsletter.substack.com, 2026-03-25.

---name: classify-agentdescription: Classify a problem into one of four agent architectures: coding harness, dark factory, auto research, or orchestration framework. Implements the "one-question test" diagnostic from Nate Kadlac's agent taxonomy. Use when starting a new agent project, evaluating a build approach, or when someone says "what kind of agent should I build?", "classify this", or "/classify-agent".---# Agent Architecture ClassifierTakes a problem description and determines which of four agent architectures is the best fit, using the diagnostic question: **"What are you optimizing against?"**## The Four Architectures| Architecture | Governing Principle | Optimizing Against | Example ||---|---|---|---|| **Coding Harness** | Decomposition boundaries | Task isolation -- each unit of work touches minimal shared state | Claude Code, Cursor, Aider || **Dark Factory** | Specification-as-code | Convergence to spec -- can an agent reach the target without human input? | StrongDM's autonomous build pipeline || **Auto Research** | Metric + guardrail | Measurable improvement -- is there a benchmark to beat? | Karpathy's 630-line research loop || **Orchestration Framework** | Handoff contracts | Pipeline throughput -- do agents pass clean data to each other? | CrewAI, LangGraph, multi-agent chains |## Phases### Phase 1: Gather ContextAsk the user to describe their problem in 2-3 sentences. If they've already provided it, skip this.### Phase 2: The One-Question TestAnalyze the description against the diagnostic: **"What are you optimizing against?"**Map the answer:- "I need isolated tasks that don't step on each other" -> **Coding Harness**- "I need an agent that converges to a precise spec without hand-holding" -> **Dark Factory**- "I need to beat a benchmark or improve a measurable metric" -> **Auto Research**- "I need multiple agents to coordinate and pass work between them" -> **Orchestration Framework**### Phase 3: Confirmation DiagnosticsRun three follow-up questions to confirm the classification:1. **Decomposition test**: "Can the work be split into units that don't share files/state?" (Yes = Coding Harness signal)2. **Spec maturity test**: "Do you have a precise, testable specification?" (Yes = Dark Factory signal)3. **Metric test**: "Is there a quantitative benchmark you're trying to beat?" (Yes = Auto Research signal)4. **Handoff test**: "Does the output of one step become the input of another step with a different agent?" (Yes = Orchestration signal)### Phase 4: Output ClassificationReport:- **Primary architecture**: The best fit with confidence (high/medium/low)- **Runner-up**: Second-best fit if signals are mixed- **Anti-patterns to avoid**: Common mistakes for this architecture type- **Governing principle**: The one rule that should guide all design decisions- **Next step**: What to build first## Verification- Classification maps to exactly one of the four architectures- At least 2 of 4 diagnostic signals align with the primary classification- Anti-patterns are specific to the chosen architecture, not generic advice## SourceNate Kadlac, "There are 4 kinds of agents and you're probably using the wrong one", natesnewsletter.substack.com, 2026-03-25.

Frequently asked questions