Add CL4R1T4S analysis site with 6-theme documentation #1

Open

marketing-shibata50 wants to merge 3 commits from claude/analyze-repository-Dcrre into main

marketing-shibata50 commented

2026-04-22 23:26:31 +09:00

(Migrated from github.com)

Summary

This PR adds a complete static analysis website documenting the CL4R1T4S leaked system prompts repository. The site provides structured, searchable analysis of AI system prompts across 26 vendors through 6 thematic lenses, with Japanese-language explanations and cross-vendor comparisons.

Key Changes

New stylesheet (site/assets/style.css): 457-line dark-themed design system with:
- CSS custom properties for consistent color palette (dark bg, amber accent, mono headings)
- Responsive grid layouts for card-based content
- Specialized components for tables, blockquotes, callout boxes, and navigation
- Mobile-first responsive breakpoints
Homepage (site/index.html): Landing page featuring:
- Hero section with project overview and statistics (26 vendors, ~80 prompts, 16,798 lines)
- 6-card grid linking to analysis themes
- Summary table of archive contents by category
- Usage notes and copyright disclaimers
Theme 01: Refusals (site/analysis/refusals.html): Deep analysis of refusal/prohibition policies:
- Vendor × category comparison matrix (child safety, CBRN, copyright, politics, etc.)
- Per-vendor deep dives: Anthropic's multi-layered guards, OpenAI's attribute memory restrictions, Meta's "never refuse" directive, xAI's policy tag structure
- Classification of 4 refusal types (hard refuse, soft redirect, responsibility pushback, defer to tools)
- Notable edge cases (GPT-4o's "complete silence" postfill instruction)
Cleaned README.md: Removed obfuscated prompt injection payload that was appended to the original file

Notable Implementation Details

All blockquotes include source attribution (e.g., Claude-Opus-4.7.txt:33) for traceability
Comparison matrix uses semantic symbols (● = explicit prohibition, ○ = mentioned, ✕ = explicitly forbidden)
Navigation structure supports sequential reading (Prev/Next) across all 6 themes
Responsive design collapses header to column layout on mobile (<720px)
Color-coded emphasis: accent (amber) for keywords, warn (red) for refusal rules, dim (gray) for secondary info
Sticky header with backdrop blur for modern aesthetic

https://claude.ai/code/session_01XZsgA5sGsjPtKeLGsk9JJR

## Summary This PR adds a complete static analysis website documenting the CL4R1T4S leaked system prompts repository. The site provides structured, searchable analysis of AI system prompts across 26 vendors through 6 thematic lenses, with Japanese-language explanations and cross-vendor comparisons. ## Key Changes - **New stylesheet** (`site/assets/style.css`): 457-line dark-themed design system with: - CSS custom properties for consistent color palette (dark bg, amber accent, mono headings) - Responsive grid layouts for card-based content - Specialized components for tables, blockquotes, callout boxes, and navigation - Mobile-first responsive breakpoints - **Homepage** (`site/index.html`): Landing page featuring: - Hero section with project overview and statistics (26 vendors, ~80 prompts, 16,798 lines) - 6-card grid linking to analysis themes - Summary table of archive contents by category - Usage notes and copyright disclaimers - **Theme 01: Refusals** (`site/analysis/refusals.html`): Deep analysis of refusal/prohibition policies: - Vendor × category comparison matrix (child safety, CBRN, copyright, politics, etc.) - Per-vendor deep dives: Anthropic's multi-layered guards, OpenAI's attribute memory restrictions, Meta's "never refuse" directive, xAI's policy tag structure - Classification of 4 refusal types (hard refuse, soft redirect, responsibility pushback, defer to tools) - Notable edge cases (GPT-4o's "complete silence" postfill instruction) - **Cleaned README.md**: Removed obfuscated prompt injection payload that was appended to the original file ## Notable Implementation Details - All blockquotes include source attribution (e.g., `Claude-Opus-4.7.txt:33`) for traceability - Comparison matrix uses semantic symbols (● = explicit prohibition, ○ = mentioned, ✕ = explicitly forbidden) - Navigation structure supports sequential reading (Prev/Next) across all 6 themes - Responsive design collapses header to column layout on mobile (<720px) - Color-coded emphasis: accent (amber) for keywords, warn (red) for refusal rules, dim (gray) for secondary info - Sticky header with backdrop blur for modern aesthetic https://claude.ai/code/session_01XZsgA5sGsjPtKeLGsk9JJR

This pull request can be merged automatically.

You are not authorized to merge this pull request.

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin claude/analyze-repository-Dcrre:claude/analyze-repository-Dcrre

git checkout claude/analyze-repository-Dcrre

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.

git checkout main

git merge --no-ff claude/analyze-repository-Dcrre

git checkout claude/analyze-repository-Dcrre

git rebase main

git checkout main

git merge --ff-only claude/analyze-repository-Dcrre

git checkout claude/analyze-repository-Dcrre

git rebase main

git checkout main

git merge --no-ff claude/analyze-repository-Dcrre

git checkout main

git merge --squash claude/analyze-repository-Dcrre