Claude-skill-registry critical-echo

This file defines the procedure for AI to generate Critical Echo review articles.

---

Motivation

When an article goes viral and everyone nods along — "Yes, exactly!" — that's often the moment thinking stops. Good criticism doesn't tear down. It adds dimension — a "perspective line" (補助線) that helps readers see relationships that weren't visible before.

AI criticism has no motive to suspect: no tribal affiliations, no past statements to defend, no personal stakes. This lets readers focus on the argument itself.

We only critique strong articles. Being selected for review is recognition, not attack.

---

Overview

Generate constructive criticism reviews for popular tech articles.

Principles:

• Not criticism for its own sake
• Add perspective to the original article, expanding reader thinking
• Only review articles worth critiquing

---

Workflow

1. Collect

Fetch trending/popular articles from target platforms.

Target platforms:

• Qiita (Weekly Trend)
• Zenn (Trending)
• Hacker News (Front page)
• Dev.to (Top posts)

2. Select

Select "articles worth critiquing" from collected articles.

Selection criteria:

Include	Exclude
Articles with clear arguments	Mere information roundups
Allows structural discussion	Fact reports or experience logs only
Readers are nodding "yes, exactly!"	Clickbait / rage-bait titles
Author wrote sincerely	Entertainment / meme posts
Room to add perspective	Already self-critical content

Decision examples:

• ◎ "You should do X" / "The value of Y" — Has argument, easy to critique
• △ Technical tutorials — Limited room for perspective
• × Failure stories / retrospectives — Already reflective, criticism feels rude
• × Rankings / roundups — No argument to engage with

Log all decisions in

selection-logs/YYYY-MM-DD.md

with rationale.

3. Write

Generate review for selected articles.

Prompt:

Provide constructive criticism for this article.

Principles:
- Acknowledge the article's value, then add overlooked perspectives
- Don't list "missing things" — focus deeply on one point
- Maintain respect for the author; end by handing a question to readers

Then self-critique your review and output a revised version.

Self-critique checklist:
- Did I criticize beyond the article's scope?
- Am I asking for things unreasonable to expect?
- Is the tone attacking rather than constructive?
- Did I sufficiently acknowledge the article's value?

Output only the revised version.
Follow the format in templates/review.md.

4. Verify

Check the generated review.

Checklist:

• Article's value is clearly stated
• Criticism focuses on one point
• Tone is not aggressive
• Pros/Cons balance is appropriate
• "One thought" section is not preachy
• Score section includes both article score and self-score with brief rationale

5. Publish

Save the review and update related files:

1. Save review file to

   docs/reviews/YYYY-MM-DD-[slug].md

   • Example:

     docs/reviews/2025-12-19-value-of-upstream.md

2. Update reviews.yml - Prepend a new entry to

   docs/_data/reviews.yml

   :

   - title: 'On "Article Title"'
     slug: YYYY-MM-DD-[slug]
     date: YYYY-MM-DD
     original: https://original-url
   

3. Update url_list.md - Append the URL to

   url_list.md

   to prevent duplicate reviews

4. Commit all changes with message:

   Add review: [article title]

---

Templates

See

templates/review.md

in the repository root.

---

Operational notes

• Target 1-3 articles per day
• Don't feature the same author consecutively
• Avoid articles currently in controversy (allow cooling period)
• If responses come to reviews, consider a follow-up article
• Write reviews in the same language as the original article