socratic-duck

A Claude skill for stress-testing a single technical decision through structured interrogation, producing a persistent decision log under decisions/ in your project root.

What it does

When triggered, the skill runs a one-question-at-a-time interrogation over a single root decision and its sub-decisions (up to 3 levels deep). It investigates the codebase first, refuses to propose answers, criticizes weak reasoning after each user response, and produces a markdown decision log when the session ends.

Warning: this skill is deliberately confrontational

This is not a friendly assistant mode. The skill is designed to:

• Refuse to validate decisions until trade-offs are stated explicitly.
• Push back on weak, hand-wavy, or unverified answers.
• Stop the session and force resolution when the user contradicts themselves.
• Refuse to propose "the answer" when the user does not know, offering options with trade-offs instead.

If you want encouragement, supportive feedback, or quick alignment, do not install this skill. Use a different one or interact with Claude normally. This skill assumes you want your thinking challenged, not confirmed.

When it triggers

Explicit triggers:

• "Question this", "grill me on this", "pressure-test this".
• "Let's discuss in depth", "let's go deep on this".
• "Technical decision:" used as a prefix to introduce a decision.

Proactive trigger (asks for confirmation before running):

• The user appears to be committing to a non-trivial architectural, infrastructure, security, or stack decision without stating trade-offs.

The skill does not trigger for decisions that are local, easily reversible, and do not affect interfaces, data models, infrastructure, or external contracts.

Output

A markdown file at <project-root>/decisions/YYYY-MM-DD-<slug>.md containing context, the decision, alternatives rejected, trade-offs accepted, assumptions with revisit conditions, sub-decisions (light treatment), weak points flagged at the end of the session, deferred items, and references.

If a file with the same name already exists, the skill stops and asks how to proceed.

Credit and prior art

This skill is a variant of Matt Pocock's grill-me skill. The differences:

• Single root decision per session, with bounded depth (max 3 sub-levels).
• Mandatory codebase investigation before asking the user.
• The skill never proposes answers; only offers options with trade-offs when the user is stuck.
• Hard stop on user contradictions until they are resolved.
• Persistent output: every session produces a markdown decision log under decisions/ so the reasoning survives the session.

If you want the lighter, more open-ended version, use Pocock's original.

License and authorship

This skill is released under the MIT License (see the LICENSE file at the repository root). Original concept by Matt Pocock, whose grill-me skill is also MIT-licensed; this is an independent reimplementation of the idea, not a fork of his code.