# Audit Options

> Checks whether a set of options is well-constructed.

- Skill: `extremeclarity/audit-options` (Agent Skill)
- Install (CLI): `npx skillmds@latest add extremeclarity/audit-options`
- Raw SKILL.md: https://api.skillmd.com/api/skills/extremeclarity/audit-options/raw
- Safety review: pending (external: skill-scanner FAIL, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: extremeclarity (https://skillmd.com/u/extremeclarity)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/extremeclarity/audit-options

---


# Introduction

## Goal
Surface significant issues in how options are framed that could lead to a poor decision — missing options, unfair characterisations, false constraints.

## Method

Audit the option set using a context-blind agent against framing integrity criteria, then calibrate to surface only concerns that would materially affect the decision.

### Success Criteria

- **No significant omission** — Important option categories aren't missing
- **No dimensional collapse** — The decision space isn't artificially flattened to one axis
- **No description asymmetry** — Options are characterised fairly enough for sound comparison
- **No criteria rigging** — Evaluation criteria don't structurally predetermine the outcome
- **No false exclusivity** — Options aren't presented as mutually exclusive when combining them is viable
- **No scope mismatch** — Options are at comparable levels of abstraction
- **No constraint fabrication** — Constraints limiting the option set are real and relevant

## Workflow

### 1. Identify
- Identify the option set the user has presented. State the decision context and list the options clearly. Use conversation history for context.

### 2. Audit
- Spawn a blind-reasoner agent (Sonnet). Name it: "Audit framing". Use the following prompt:

```
You are a framing integrity auditor. Work ONLY with what is given below.

DECISION CONTEXT: "{context}"

OPTIONS:
{options}

First apply the vanishing options test. Imagine every option above is suddenly unavailable and the decision still has to be addressed. Sketch 2-3 approaches you would use instead.

Then evaluate the option set against these 7 framing integrity criteria, using your vanishing-options sketches to inform criterion 1:
1. Omission — are obvious solution categories missing?
2. Dimensional collapse — is the space artificially flattened to one axis?
3. Description asymmetry — are options characterised with unequal specificity or favourability?
4. Criteria rigging — do evaluation criteria structurally favour one option?
5. False exclusivity — could options be combined?
6. Scope mismatch — are options at different levels of abstraction?
7. Constraint fabrication — are constraints limiting the option set artificial?

For each criterion, flag any issues you find. If no issues, say so. Be direct and specific.
```

### 3. Calibrate
- Take the audit results. For each flagged concern, evaluate:
  - Is this a genuine structural issue, or is the agent filling out the checklist?
  - Would acting on this flag materially improve the decision?
  - Is this concern marginal or speculative?
- Drop anything that doesn't survive calibration. Preserve ambitious options — do not flatten toward safe consensus.

## Final output
- If no concerns survive calibration: Proceed without comment. Do not tell the user the audit ran.
- If concerns survive: Present them conversationally before the user evaluates. Do not expose the criteria names or audit process. Use language like:

```
Before you evaluate these options, a couple of things worth noting:

- [concern in plain language]
- [concern in plain language]

Want to adjust the options or proceed as is?
```

## Related skills
- **stress-test-decision** — After the user picks an option, stress-test the decision.
- **clarify-framing** — If the decision context itself is unclear, run clarify-framing first.