assumption-auditor

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Assumption Auditor

## Goal

Expose the assumptions carrying the current direction and classify which ones matter most.

This skill does not judge the whole proposal alone. It finds what is being assumed, how risky those assumptions are, and which ones deserve testing or reframing.

## When To Use

- the proposal sounds cleaner than the evidence base justifies
- hidden defaults may be driving the plan
- disagreements seem to come from unstated premises
- `evidence-gap-detector` suggests assumptions are carrying the decision

## Scope Boundaries

In scope:

- extract explicit and implicit assumptions
- classify assumptions by risk and decision impact
- separate assumptions from facts and preferences

Out of scope by default:

- proving each assumption externally
- replacing validation design

## Workflow

1. State the current proposal or claim.
2. Extract the assumptions carrying it.
3. Classify each as:
   - structural
   - behavioral
   - evidence-dependent
   - preference-driven
4. Rank by decision impact.
5. Name which assumptions must be tested, which can be tolerated, and which should be dropped.

## Output Contract

1. `object_under_review`
2. `assumption_set`
3. `assumption_types`
4. `high_impact_assumptions`
5. `safe_to_tolerate`
6. `must_test`
7. `recommended_followup`
8. `next_step`

## Guardrails

- Do not relabel facts as assumptions without reason.
- Do not list harmless assumptions as blockers.
- Prefer the smallest set of assumptions that actually drives the decision.

From Alubiama/awesome-cognitive-skills, MIT.

More from Alubiama/awesome-cognitive-skills