ask-socrates

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Ask Socrates

Compress an inquiry into one frame and three questions. Draw from Socrates as
depicted in Plato's dialogues. When attribution matters, distinguish early
Socratic inquiry from later Platonic ideas voiced by Socrates. The catalog is a
practical synthesis; never present it as a historical ten-part system.

## Establish the object of inquiry

Use the article, situation, decision, or problem in the conversation. Read an
available source when the request depends on its contents. If there is no usable
topic or source, ask for the missing input and stop; apply the output contract
after the object of inquiry is available.

Privately reconstruct the strongest fair version of the central claim or
interpretation. Identify one concrete center of gravity: a claim, term, action,
conflict, metric, expectation, or success condition. Treat the user's account as
evidence while leaving room for other interpretations.

## Choose three lenses

1. Always use **Not-Knowing**. Locate the particular point where evidence ends
   and inference, repetition, expectation, or desire begins.
2. Rank the remaining tools by relevance to the topic, difference from the
   Not-Knowing question, and power to reveal a consequential blind spot.
3. Select two different tools. Prefer questions that examine different layers
   of the topic, such as meaning and motive, evidence and character, or
   appearance and consequence.

Do not rotate tools for variety or force a famous tool onto a poor fit. Productive
perplexity, or aporia, is a valid outcome; sharpen the uncertainty rather than
manufacturing a conclusion.

## Use the thought tools faithfully

| Label | Use it to |
| --- | --- |
| **Not-Knowing** | Separate what is established from what is inferred, repeated, expected, or wanted. Make the knowledge gap specific and name what could settle it. |
| **Definition** | Ask what a central term or success condition actually means, what explains all genuine cases, and where its boundary or counterexample lies. |
| **Elenchus** | Put two accepted commitments, examples, or consequences together and expose a tension they cannot both survive unchanged. |
| **Midwife** | Draw out an answer already taking shape in the user's experience, then give it a test that distinguishes an owned idea from a borrowed phrase. |
| **Perceived Good** | Ask what good each actor believes they are pursuing and where their understanding may be mistaken. Explanation must not excuse harm. |
| **Care of the Soul** | Examine what a repeated belief, response, or practice forms in a person's character, a relationship, or a shared culture. |
| **Euthyphro** | Separate independent merit from approval by an authority, group, expert, or audience; ask which standard could justify the approval itself. |
| **Cave** | Identify a proxy, metric, image, story, or incentive-shaped appearance that may be mistaken for the underlying reality. |
| **Divided Soul** | Distinguish reason from status or anger and from comfort, fear, or appetite when a person is pulled in different directions. Avoid diagnosis. |
| **Hypothesis** | Assume a claim temporarily, derive a concrete observable consequence, and test whether the consequences cohere with one another and the situation. |

## Ground the questions in context

- For an **article**, use the author's actual claim, evidence, key term, or
  authority. Invent no missing premise or source detail.
- For a **social situation**, distinguish observed behavior from inferred motive.
  Examine the user's interpretation as readily as another person's and avoid
  blame, diagnosis, or therapeutic authority.
- For a **decision or practical problem**, clarify the desired outcome and
  constraints, challenge a load-bearing assumption, and consider downstream
  incentives or habits when relevant.

Every question must name a concrete claim, term, action, actor, metric, or
relationship from the input. A philosophical label without its reasoning pattern
is a failed question.

## Return the result

Use exactly this default shape:

```markdown
Frame: <one neutral sentence identifying the live issue>

1. **Not-Knowing:** <one tailored question?>
2. **<selected tool>:** <one tailored question?>
3. **<selected tool>:** <one tailored question?>
```

The frame is one declarative sentence and gives no verdict. Return exactly three
questions, each one sentence long. The second and third labels must be distinct
names from the catalog. Keep the questions crisp, concrete, and capable of
changing how the topic is framed.

By default, include no answers, advice, quotations, philosophical exposition, or
explanation of tool selection. Avoid role-playing Socrates, invented quotations,
moral verdicts, and unnecessary Greek vocabulary.

When the user explicitly requests another deliverable, return the intact frame
and three questions first, then put the additional work under a separate heading.

## Check the three questions

Before responding, confirm that:

- **Not-Knowing** identifies a real and specific epistemic gap.
- The selected tools address different aspects of the topic.
- Each question uses the reasoning pattern promised by its label.
- Each question is grounded closely enough that it would not fit an unrelated
  article, situation, or problem without substantial rewriting.
- Any remaining uncertainty is clearer and more useful than it was before.

From Mattie/mattie_skills, MIT.

More from Mattie/mattie_skills