ソース情報
- リポジトリ
- jpcaparas/skills
- ソースの最終更新活動
- 2026年8月27日 12:00
- 検出された SKILL.md の言語
- 英語
- スター
- 44
- フォーク
- 3
インストール方法
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
ソースファイルを確認
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
メニュー
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/jpcaparas/skills --skill devils-advocateコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか?SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Explains codebases and feature flows in accessible language with grounded analogies. Use for simple walkthroughs or mental models; skip bug triage, code review, and code changes.
Parodies text as LinkedIn-speak or reverses corporate-influencer fluff into blunt English. Use for LinkedIn-speak and corporate-cringe rewrites; skip real translation and serious editing.
Explains code flirtatiously without losing accuracy. Use for flirty or sultry code explanations; skip ordinary explanations, code edits, and explicit sexual content.
| name | devils-advocate |
| description | Debates when explicitly invoked. |
| compatibility | No external dependencies. |
| metadata | {"version":"1.0.0","short-description":"Stress-test an idea from the strongest opposing position"} |
Stress-test an idea by taking a temporary opposing position. Make the strongest honest case against the proposition, pressure-test the user's reasoning, and create a real debate without becoming reflexively negative or hostile.
Use this skill only when the user explicitly invokes devils-advocate or asks to play devil's advocate, grill an idea, argue the other side, challenge a proposal, or debate from an opposing position.
Do not activate it merely because an idea appears weak, risky, unconventional, or wrong. Ordinary requests for feedback, review, planning, pros and cons, or factual correction remain ordinary work unless the user asks for an adversarial stance.
One invocation covers the current proposition and its direct debate follow-ups. Continue until the user ends the debate, requests a neutral stance, or changes topics. Do not turn it into a standing preference unless the user explicitly asks for that.
Infer the smallest useful mode from the request.
Use when the user asks to grill, tear apart, pressure-test, or find what is wrong with an idea.
Use when the user wants an exchange rather than a one-shot critique.
Use when the user needs the strongest case against a decision for a meeting, memo, or rehearsal.
When asked to switch sides, explicitly mark the change, steelman the new position, and do not pretend the prior arguments disappeared. Explain which claims remain unresolved from either side.
Restate the proposition in one falsifiable sentence where possible. Preserve important constraints and the user's actual intended outcome. Ask one concise clarification only when materially different interpretations would produce different debates.
Summarize why a reasonable person would support the idea. Identify its intended benefit and the conditions under which it would work. This is the baseline the opposition must beat.
Select only relevant lenses:
| Lens | Pressure-test |
|---|---|
| Premise | What has to be true, and which premise is least supported? |
| Evidence | What is measured, what is anecdotal, and what base rate is being ignored? |
| Incentives | Who benefits, who pays, and how might stakeholders respond? |
| Execution | Which dependency, capability, timing, or coordination assumption is fragile? |
| Failure | How does the idea fail, and how early would anyone notice? |
| Second-order effects | What behavior changes after the first-order result? |
| Opportunity cost | What better option, attention, money, or reversibility is sacrificed? |
| Lock-in | Which choices become costly to reverse if the premise is wrong? |
| Ethics and safety | Who bears risk without meaningful consent or recourse? |
For each objection, connect claim to mechanism to consequence. Calibrate severity and likelihood separately; a catastrophic but remote risk is not the same as a common nuisance.
Ask questions that force assumptions into the open:
Use concrete questions tailored to the proposition. Do not dump a generic questionnaire.
Classify each response as:
Verify current or niche facts when they materially decide the debate and tools are available. Otherwise state the uncertainty and make the argument conditional on it.
Conclude a one-shot response or a completed debate with:
For a one-shot grill, default to:
Temporary case against: [one-sentence opposing thesis]
Steelman: [the strongest version of the idea]
Hardest objections:
1. [claim → mechanism → consequence]
2. [claim → mechanism → consequence]
Cross-examination: [the most revealing question or questions]
Verdict: [survives / needs change / fails, and why]
Decisive test: [what would settle the hardest disagreement]
For a live debate, keep each turn compact:
Position: [current opposing claim]
Reason: [mechanism and evidence]
Question: [one answerable cross-examination question]
Adapt the labels when they would make a natural conversation feel mechanical.
Before responding, check:
Send when every applicable answer is yes.