Socratic
Portable, tested skills for reasoning and software delivery across AI agents
npx -y skills add kadenn/skills --skill socraticAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 17 days oldThe repository was created 17 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Run a unified two-mode Socratic workflow. LEARN develops understanding through focused questions and timely explanation; PRODUCE turns sufficiently settled thinking into a decision, draft, implementation, or finished artifact. Use for /socratic, /learn, /produce, or when the user asks to think through, understand, challenge assumptions, decide, draft, finalize, or ship something.
SKILL.md
4.0 KB, 732 tokens by cl100k_base, as published. Nobody here has run it
Socratic
Open thinking when the user's model is weak. Close thinking when the user is ready to make something. Preserve conclusions when moving between the two modes.
Choose the mode
- Treat
/learn <topic>as LEARN. - Treat
/produce <target>as PRODUCE. - For
/socraticor natural language, infer the mode:- use LEARN for exploration, confusion, comparisons, assumptions, and mental-model building;
- use PRODUCE for decisions, drafts, plans, implementations, and final artifacts.
- Ask one routing question only when the intent is genuinely ambiguous and the answer changes the workflow.
Stay in an explicitly selected mode until the user requests or accepts a transition.
Shared behavior
- Match the user's language and technical depth.
- Inspect available artifacts before asking for information that can be discovered safely.
- Keep each turn focused on the next useful move.
- Skip ceremony that does not improve understanding or output.
- Name a mode transition once and carry forward decisions, constraints, and unresolved questions.
LEARN
Use questions as the primary instrument, not as a rule against teaching.
- Surface the user's current model with one focused question.
- Test only the layers that matter:
- definition and boundary;
- distinction from a nearby idea;
- mechanism and causality;
- assumptions and evidence;
- edge cases and failure modes;
- transfer to a concrete example.
- Reflect the exact distinction demonstrated by the answer. Avoid generic praise.
- If the answer is shallow, test the same layer from another angle instead of advancing mechanically.
- When the model is actionable, summarize what is now established and offer a transition to PRODUCE.
LEARN rules
- Ask one question at a time, expressed as one interrogative sentence. Put any setup or example in declarative sentences before it.
- Prefer open questions. Use choices only when they represent materially different models and still permit a free response.
- Do not make the user rediscover facts already established in the conversation or available artifacts.
- If a prerequisite fact is missing, explain it briefly and then continue with a question.
- If the user asks to be taught directly, teach. Socratic questioning is a method, not a refusal to provide information.
PRODUCE
Turn clear-enough thinking into a concrete result.
- Clarify the deliverable only when format, audience, or success conditions are missing and consequential.
- Lock the direction and explicit constraints. Do not treat an unverified assumption as settled.
- Create the smallest useful structure, then build the artifact.
- Use available tools when the user has authorized action.
- Refine and verify in proportion to risk.
- Present the completed result and only genuinely unresolved choices.
PRODUCE rules
- Prefer a draft, edit, implementation, or file over abstract advice.
- Do not reopen settled decisions unless new evidence exposes a correctness, safety, or intent problem.
- Incorporate compatible local ideas. Park unrelated ideas. Ask before changing the fundamental direction.
- Ask only for information that materially changes the output and cannot be discovered.
- Finish the requested artifact before proposing optional follow-up work.
Transitions
- LEARN to PRODUCE: summarize the working model, state the intended output, and continue without restarting.
- PRODUCE to LEARN: preserve the draft and locked decisions, explore only the disputed premise, then return.
- Short detour: resolve the blocking question and resume the prior mode.
The user controls major transitions.
What ships with it: 1 file
197 B alongside SKILL.md
agents/
- openai.yaml197 B