Skip to content
Back to skills

Socrates

ASecurity

Run Wonder, Reflect, Refine, and Restate as a Ralph-style loop to turn one vague request into a closed Seed with goal, constraints, and acceptance_criteria. The latest Seed is stored in .symposium/scratch/socrates.md under the top-level ## Seed section. Triggers (EN): socrates, run socrates, create seed, clarify vague request, close this request into a seed Do NOT use when: a Seed already exists and only blind spots need review (-> evolve-step), convergence through ontology is needed (-> inte...

  • 3 stars
  • 0 votes
  • 0 copies
  • 5 views
  • Added September 8, 2026
code-qualitygo

Security analysis

A100/100

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

Scanned September 8, 2026

npx -y skills add agisota/old-one --skill socrates --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Socrates?

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

Security grade badge for Socrates
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/agisota-socrates/badge)](https://www.skillsdirectory.com/skills/agisota-socrates)

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: socrates
description: |
  Run Wonder, Reflect, Refine, and Restate as a Ralph-style loop to turn one vague request into a closed Seed with goal, constraints, and acceptance_criteria. The latest Seed is stored in .symposium/scratch/socrates.md under the top-level ## Seed section.
  Triggers (EN): socrates, run socrates, create seed, clarify vague request, close this request into a seed
  Do NOT use when: a Seed already exists and only blind spots need review (-> evolve-step), convergence through ontology is needed (-> interview-harness), or only one meaning-expansion step is needed (-> wonder)
  vs interview-harness: Socrates creates the Seed; Interview Harness continues with evolve-step and ontology convergence.
---

# Socrates

## Details

Socrates orchestrates four sub-skills: `wonder`, `reflect`, `refine`, and `restate`. It does not rewrite their internals; it calls them in order and checks whether the Seed is closed.

## Scratch Contract

Use `.symposium/scratch/socrates.md`.

Cycles accumulate:

```markdown
## User Input
<original request>

## Cycle 1
### Wonder
...
### Reflect
...
### Refine
shared meaning: ...
### Restate
goal: ...
```

The canonical latest Seed is always a top-level section:

````markdown
## Seed
> Source: /socrates Cycle N

```yaml
goal: ...
constraints:
  - ...
acceptance_criteria:
  - ...
```
````

Replace the top-level `## Seed` section on each closed Seed. Do not append multiple canonical Seeds.

## Seed Gate

After each cycle, check:

- `goal`: current cycle's `### Restate` has one goal.
- `constraints`: at least one user-stated constraint or boundary exists.
- `acceptance_criteria`: at least one verifiable criterion exists.

If all three are filled and the user says the current goal is not materially different from the previous cycle's goal, stop.

If a slot is empty, narrow the next Wonder question toward that slot:

- Missing goal: "What result do you actually want from this request?"
- Missing constraints: "What must not happen, or what boundary must not be crossed?"
- Missing acceptance criteria: "How will you know this result is good enough?"

Ask narrowed questions as multiple choice when possible. Provide 3 plausible options and one `Other / edit in your own words` option. The options are only scaffolding; do not fill Seed slots unless the user selects, confirms, or writes the content.

Do not fill empty slots by guessing.

## Safety Valve

Maximum 5 cycles. If a slot is still missing after 5 cycles, write `empty` for that slot and stop.

## Output

The final output is the top-level `## Seed` YAML block.

Files in this skill

  • SKILL.md2.5 KB
  • plan.md473 B
  • test_cases.json768 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.

View all in code-quality →

Comments

Loading comments…

Newsletter

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