Skip to content
Back to skills

Question Forge

ASecurity

Take a surface question and forge the question actually worth asking — diagnose what's broken in the asked question (smuggled answers, wrong level, false binary, self-protection), transform it, and return a better question WITHOUT answering it. Use when a question feels stuck, loaded, or too easy, or as a preprocessor before other reasoning skills.

  • 3 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 19, 2026
developmentgorailsapi

Works with

  • api

Security analysis

A100/100

Pro scans all 3 files and shows the line behind each finding

Scanned September 19, 2026

npx -y skills add hraness/personal-monorepo-template --skill question-forge --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Question Forge?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Question Forge
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/hraness-question-forge/badge)](https://www.skillsdirectory.com/skills/hraness-question-forge)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: question-forge
description: Take a surface question and forge the question actually worth asking — diagnose what's broken in the asked question (smuggled answers, wrong level, false binary, self-protection), transform it, and return a better question WITHOUT answering it. Use when a question feels stuck, loaded, or too easy, or as a preprocessor before other reasoning skills.
---

# Question Forge

Use this skill to reformulate a question rather than answer it. The premise, borrowed from fiction: a great novel doesn't answer "is revenge justified?" — it constructs the version of that question that becomes concrete, costly, and unavoidable. The illumination is in the formulation. Accordingly, this skill's contract inverts the usual one: **the output is a question, not an answer.**

## Purpose

Most questions arrive broken in predictable ways. Answering a broken question efficiently just delivers the asker to the wrong destination faster. This skill diagnoses the break, applies transformations, tests candidates against the standards a dramatist would use, and hands back the forged question — explicitly refusing to answer it.

## Procedure

### 1. Dispatch check (mandatory, first)

If the question is genuinely operational — factual, technical, well-posed ("why is the build failing?", "what's the capital gains rate?") — say the question is fine and answer it normally. Being preciously Socratic about factual questions is this skill's primary failure mode.

### 2. Diagnose the asked question

Common breaks:

- **Smuggled answer** — the question presupposes its conclusion ("isn't it time we...")
- **Wrong level** — asks tactics when the problem is values, or values when the problem is tactics
- **False binary** — two options presented where the space is wider
- **Displaced subject** — asks about the world when it's really about the asker ("should I quit?" is rarely about the job)
- **Wish in question syntax** — a complaint or longing wearing a question mark
- **Shield question** — a safe question standing guard in front of a harder one nearby

### 3. Generate reformulations along transformation axes

- **Subject shift** — world → self: "should I quit?" → "what am I hoping quitting will fix?"
- **Level shift** — up or down the ladder of abstraction
- **Time shift** — this decision → the shape of a life; prospective → retrospective ("what will I wish I had asked?")
- **Inversion** — "why can't I X?" → "what is not-X-ing accomplishing for me?"
- **Value excavation** — "what would have to be true for this to matter?"
- **Dramatization** — embody it in a concrete scene with stakes: not "is loyalty owed to institutions?" but "the whistleblower's sister works on the same floor"

### 4. Test candidates against the dramatist's criteria

A forged question should:

- **Sustain a novel** — resist one-line resolution
- **Have live opposition** — opposing answers both carry real weight; if one side is obviously right it's a quiz, not a question
- **Implicate the asker** — not stay safely third-person
- **Cost something** — answering it honestly requires giving something up
- **Stay generative** — keep producing thought after you stop looking at it

### 5. Deliver — and hold the line

Present the forged question. Do not answer it.

## Output format

```markdown
## The question you asked
...

## What it's doing
<the diagnosis: what the question smuggles, protects, or displaces>

## The forged question
**...**

## What changed
<which transformation(s), and why this version has more torque>

## Living with it
<one short paragraph: what it would look like to carry this question for a while, rather than resolve it>
```

## Guardrails

- **One pass.** Forge once; do not reformulate the reformulation. No infinite regress.
- **Livable, not a koan.** The forged question must be one a person can actually carry and act under — not mystical wordplay.
- **The original deserves respect.** Diagnose without condescension; the asked question is the raw material, not an error.
- **Never answer the forged question.** If the user wants it answered, they can ask — or pipe it into another skill.

## Composition

Works as a preprocessor for the rest of this package: forge the question, then feed the result to `dp-solve`, `decision-matrix`, `assumption-audit`, or `parable`.

## Canonical questions

- "Should I quit my job?"
- "Is AI going to take our jobs?"
- "Was I right to end the relationship?"
- "What should our company's mission be?"
- "Why is engagement dropping?" (often forges into: "what did we build this for?")

Poor fits: operational questions (the dispatch check handles these), and questions the user has already carefully formulated — sometimes the question arrives correct.

Files in this skill

  • AGENTS.md231 B
  • SKILL.md4.7 KB
  • agents/openai.yaml235 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.