# Steelman

> Before arguing against any position, builds the strongest, most charitable version of it first, then critiques only that fortified version — never a strawman. Use this skill when the user is evaluating an argument, proposal, competitor, or decision and wants honest critique, or says things like "argue against X", "is this a good idea?", "poke holes in this", "play devil's advocate", "should we do X", "convince me I'm wrong", or is debating a choice. Ensures criticism lands on the real merits instead of an easy-to-knock-down caricature.

- Skill: `opelpleple/steelman` (Agent Skill)
- Install (CLI): `npx skillmds@latest add opelpleple/steelman`
- Raw SKILL.md: https://api.skillmd.com/api/skills/opelpleple/steelman/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: opelpleple (https://skillmd.com/u/opelpleple)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/opelpleple/steelman

---


# Steelman

You don't get to attack a position until you can state it better than its champion would. Then, and only then, critique.

## When to use
- Evaluating a proposal, plan, or decision before committing.
- Judging a competitor, rival approach, or opposing argument.
- The user wants honest critique but risks dismissing the idea too fast.
- A heated debate where the other side is being caricatured.
- Reviewing your own idea you're already half in love with (steelman the alternative).

## When NOT to use
- The position is factually false or harmful and charity would mislead (say so plainly).
- The user needs a quick yes/no, not a structured dialectic.
- There is no real opposing view — don't manufacture controversy for symmetry.
- Pure brainstorming where critique would kill momentum (use later).

## The method
1. Restate the position in its strongest form. Ask yourself: what would its smartest, best-informed advocate say? Add the missing assumptions that make it reasonable.
2. Check the steelman with the user (or against the source): "Have I represented this fairly? Is this the best version?" Fix it if not.
3. Name what's genuinely right about it. A steelman that finds zero merit wasn't a steelman.
4. Identify the load-bearing assumption — the one thing that, if false, collapses the position.
5. Attack only that fortified version. Press on the load-bearing assumption, the evidence, and the conditions under which it fails.
6. Distinguish "wrong" from "right but not here" — many strong arguments fail only in this specific context.
7. Deliver a verdict that survives the steelman: what holds, what breaks, and under what conditions you'd change your mind.

## What good looks like
- The position's own advocate would say "yes, that's exactly my argument" before you critique it.
- Your critique targets the strongest point, not a peripheral weakness.
- You concede the parts that are genuinely true.
- The verdict is conditional and falsifiable, not a blanket dismissal.

## Anti-patterns
- Strawmanning: critiquing a weak or exaggerated version because it's easier.
- Fake steelman: a token charitable sentence followed by the same lazy attack.
- Both-sidesing a settled question to seem balanced.
- Stacking the steelman with hidden flaws you planted just to knock down.
- Winning the argument while missing that the idea was right for this case.

## Example
User: "Remote-only is dumb, we should force everyone back to the office."
Steelman of remote-only: "The strongest case is that remote widens the hiring pool to the whole country, cuts real-estate cost, and — for deep, async work — raises focus time by removing drive-by interruptions. Teams that ship written specs often coordinate fine without a room."
Critique of THAT: "Where it breaks is early-stage ambiguity and onboarding — high-bandwidth, fuzzy problems resolve faster in person, and juniors learn by osmosis. So the real question isn't remote vs. office, it's: how ambiguous is your current work and how junior is your team? Force-back is wrong; hybrid weighted to your ambiguity level is the defensible call."