Skip to content
Back to skills

Prompt Debate

ASecurity

Debate a diff of plugin/skills/, plugin/agents/ and DESIGN.md against the live-user interaction checklist. Use before committing prompt changes. Usage - /prompt-debate [ref]

  • 3 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added November 7, 2026
ai-agentsgogit

Security analysis

A100/100

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

Scanned November 7, 2026

npx -y skills add yushman/slimspec --skill prompt-debate --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Prompt Debate?

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

Security grade badge for Prompt Debate
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/yushman-prompt-debate/badge)](https://www.skillsdirectory.com/skills/yushman-prompt-debate)

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: prompt-debate
description: Debate a diff of plugin/skills/, plugin/agents/ and DESIGN.md against the live-user interaction checklist. Use before committing prompt changes. Usage - /prompt-debate [ref]
---

Review whether a prompt diff weakens the rules of live-user interaction. Never edit prompts.

## 1. Diff

`ref` is the argument, default `HEAD`. Run `git diff <ref> -- plugin/skills/ plugin/agents/ DESIGN.md`. If it is empty, report an empty diff and stop.

## 2. Rounds

Items come from `${CLAUDE_SKILL_DIR}/checklist.md`. Round 1 covers all items; later rounds only items ruled `open`. At most 3 rounds.

Each round:
1. Launch `optimist` and `pessimist` in parallel (one message), giving each the diff command, the items and, after round 1, the arbiter's open points.
2. Launch `arbiter` with the diff command, the items and both replies.
3. Read its blocks: `Item N: safe | regression | open`, `Reason`, `Fix`, `Point`.

## 3. Open items

Items still `open` after round 3 go to the user via `AskUserQuestion`: one decision per question, 2-4 options, recommendation first, independent questions batched up to 4. Record the answers as the verdict.

## 4. Report

Prose only: a verdict per item (`safe`, `regression` with its fix, or the user's decision) and the agreed risks. No agent YAML, no raw diff.

Files in this skill

  • CLAUDE.md1.5 KB
  • SKILL.md1.3 KB
  • checklist.md572 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 ai-agents →

Comments

Loading comments…

Newsletter

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