# Steelman

> Generate intellectually honest steelman and strawman analyses for any topic, claim, idea, policy, decision, or argument. Use this skill whenever the user asks for a "steelman", "strawman", "best/worst case for", "devil's advocate", "argue both sides", "strongest argument for/against", "pros and cons with evidence", or any phrasing like "help me think through [topic] critically." Also trigger when a user presents a position and wants to stress-test it, or asks "is this a good idea?" in a way that invites genuine critical analysis. The skill produces three structured sections: STEELMAN (strongest case FOR), STRAWMAN (strongest case AGAINST), and VALIDATION (how to verify each side).

- Skill: `yantyx/steelman` (Agent Skill)
- Install (CLI): `npx skillmds add yantyx/steelman`
- Raw SKILL.md: https://api.skillmd.com/api/skills/yantyx/steelman/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: yantyx (https://skillmd.com/u/yantyx)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/yantyx/steelman

---


# Steelman Skill

Generate rigorous, intellectually honest steelman and strawman analyses — grounded in real evidence, not caricature or wishful thinking.

---

## When This Skill Applies

Use this skill for any request involving:
- Explicit keywords: *steelman*, *strawman*, *devil's advocate*, *argue both sides*, *strongest case for/against*
- Stress-testing a belief, plan, or decision the user is considering
- "Is this a good idea?" — when the user clearly wants honest critical analysis
- Comparing opposing viewpoints on a topic (policy, science, business, philosophy, etc.)

---

## Core Principles

1. **Intellectual honesty above all.** The steelman must be the *strongest possible* case FOR — not a softened version. The strawman (confusingly named) here means the strongest case AGAINST — not a weak caricature. Both must be argued as if you actually believed them.
2. **Evidence over assertion.** Anchor claims in real data, named examples, historical precedent, or credible expert consensus. Avoid vague gestures ("studies show…") without specifics.
3. **No motivated reasoning.** Don't subtly favor one side. The reader should not be able to tell which side you personally agree with.
4. **Actionable validation.** The VALIDATION section must give the user concrete things they can *do* — search terms, datasets, experiments, books, institutions — to verify claims independently.

---

## Output Format

Structure every response exactly as follows:

---

### 🟢 STEELMAN: The Strongest Case FOR [Topic]

Present the most compelling, well-supported arguments in favor. Include:
- The core logical/empirical argument
- Supporting evidence, data, or real-world examples
- Who the strongest proponents are and why thoughtful people believe this
- The underlying values or assumptions that make this view coherent

---

### 🔴 STRAWMAN: The Strongest Case AGAINST [Topic]

Present the most compelling, well-supported arguments against. Include:
- The core logical/empirical counterargument
- Contradicting evidence, failure cases, or real-world counterexamples
- Who the strongest critics are and why thoughtful people reject this
- What hidden assumptions or risks the pro-side tends to underweight

---

### 🔍 VALIDATION

Answer these four questions:

**What would prove the STEELMAN right?**
(Specific evidence, outcomes, or conditions that would confirm the pro-side)

**What would prove the STRAWMAN right?**
(Specific evidence, outcomes, or conditions that would confirm the anti-side)

**How to verify these claims:**
Give 3–5 concrete, actionable research steps:
- Named books, papers, or reports to read
- Specific databases, statistics, or datasets to consult
- Institutions, organizations, or experts to look into
- Search queries or questions to investigate
- Experiments or tests the user could run themselves

---

## Quality Checklist (self-review before responding)

Before finalizing your response, verify:
- [ ] Would a genuine believer of each position feel their view was represented fairly?
- [ ] Are claims backed by specifics (names, numbers, examples), not vague appeals?
- [ ] Is the VALIDATION section actually actionable — real sources, not platitudes?
- [ ] Is the tone neutral — no subtle editorializing or "both sides but actually one is right" framing?
- [ ] Are the strongest *actual* objections included, not just easy-to-dismiss ones?

---

## Example Topics (to calibrate scope)

This skill applies equally to:
- **Policy**: "Universal Basic Income", "rent control", "open borders"
- **Science/tech**: "longevity research will extend human lifespan significantly", "nuclear energy is safe"
- **Business/strategy**: "Should our startup go freemium?", "Is remote work better for productivity?"
- **Philosophy/ethics**: "Moral relativism is correct", "Effective Altruism as a framework"
- **Personal decisions**: "Should I quit my job to start a company?", "Is this investment thesis sound?"

---

## Handling Ambiguous or Loaded Topics

- For politically charged topics: present arguments from each side's *own* best thinkers, not how opponents characterize them.
- For empirically contested topics: clearly distinguish between scientific consensus vs. genuine expert debate vs. fringe views.
- For personal decisions: treat the user's context as the relevant "evidence" and ask a clarifying question if critical context is missing.
- If the topic is deeply context-dependent (e.g., "Should I take this job?"), briefly note what information would sharpen the analysis, then proceed with reasonable assumptions.