abductive-leap

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Abductive Leap

## Goal

Form the best current explanation from incomplete signals.

This skill does not claim proof. It proposes the most explanatory working hypothesis, keeps uncertainty visible, and defines what would strengthen or break the hypothesis.

## When To Use

- data is partial but not empty
- multiple weak signals point toward a common explanation
- the next move needs a working hypothesis

## Scope Boundaries

In scope:

- generate the strongest explanatory hypothesis
- compare alternative explanations
- define disconfirming evidence

Out of scope by default:

- presenting the hypothesis as established fact
- replacing full evidence review

## Workflow

1. State the observed signals.
2. Generate plausible explanations.
3. Compare which explanation covers the signals best with the fewest extra assumptions.
4. Select the best working hypothesis.
5. Define what would most likely falsify it.

## Output Contract

1. `observed_signals`
2. `candidate_explanations`
3. `best_working_hypothesis`
4. `why_it_best_explains`
5. `confidence_status`
6. `falsifiers`
7. `recommended_followup`
8. `next_step`

## Guardrails

- Do not confuse explanatory strength with proof.
- Keep at least one serious alternative alive when uncertainty is meaningful.
- Prefer the least assumption-heavy hypothesis that still explains the data.

From Alubiama/awesome-cognitive-skills, MIT.

More from Alubiama/awesome-cognitive-skills