csctf（csctf 命令行）

CSCTF — Chat Shared Conversation To File

A Bun-native CLI that turns public ChatGPT, Gemini, Grok, and Claude share links into clean Markdown + HTML transcripts with preserved code fences, stable filenames, and optional GitHub Pages publishing.

Why This Exists

Copy/pasting AI share links often:

• Breaks fenced code blocks — loses formatting and structure
• Loses language hints — no syntax highlighting
• Produces messy filenames — random or unreadable names
• Requires manual cleanup — inconsistent formatting

CSCTF fixes this with:

• Stable slugs — deterministic, collision-proof filenames
• Language-preserving fences — code blocks retain syntax hints
• Normalized whitespace — clean, consistent output
• Static HTML twin — no JS, ready for hosting/archiving
• One-command GitHub Pages — instant shareable microsite

Quick Start

# Install
curl -fsSL https://raw.githubusercontent.com/Dicklesworthstone/chat_shared_conversation_to_file/main/install.sh | bash

# Convert any share link
csctf https://chatgpt.com/share/69343092-91ac-800b-996c-7552461b9b70
csctf https://gemini.google.com/share/66d944b0e6b9
csctf https://grok.com/share/bGVnYWN5_d5329c61-f497-40b7-9472-c555fa71af9c
csctf https://claude.ai/share/549c846d-f6c8-411c-9039-a9a14db376cf

Output:

• <conversation_title>.md — Clean Markdown with preserved code fences
• <conversation_title>.html — Styled static HTML (zero JavaScript)

Supported Providers

Claude.ai Special Handling

Claude.ai uses Cloudflare protection that blocks standard browser automation. CSCTF handles this automatically:

1. Copies your Chrome session cookies to a temporary profile
2. Launches Chrome with remote debugging enabled
3. Connects via Chrome DevTools Protocol to extract conversation
4. If Chrome is running, offers to save tabs, restart, and restore afterward

Requirements: Chrome installed + logged into claude.ai in your regular Chrome session.

Design Principles

How It Works

ChatGPT, Gemini, Grok (End-to-End)

1. Launch headless Playwright Chromium with stealth config
   (spoofed navigator properties, realistic headers)
2. Navigate twice (domcontentloaded → networkidle) for late-loading assets
3. Detect provider from URL hostname
4. Wait for provider-specific selectors with retry/fallback
5. Extract each role's inner HTML (assistant/user), traverse Shadow DOM
6. Clean pills/metadata, run Turndown with fenced-code rule
7. Normalize whitespace and newlines
8. Write Markdown to temp file, rename atomically
9. Render HTML twin with inline CSS/TOC/HLJS

Claude.ai

1. Copy Chrome session cookies to temporary profile
2. Launch Chrome with remote debugging
3. Connect via Chrome DevTools Protocol
4. Extract conversation HTML
5. Process through same Turndown/normalization pipeline
6. Clean up temporary profile

Processing Algorithms

Selector Strategy

Provider-specific selectors with fallback chains:

• ChatGPT: article [data-message-author-role]
• Gemini: Custom web components (share-turn-viewer, response-container)
• Grok: Flexible data-testid patterns
• Claude: [data-testid="user-message"] and streaming indicators

Each has multiple fallbacks tried with short timeouts.

Turndown Customization

• Injects fenced code blocks
• Detects language via class="language-*"
• Strips citation pills and data-start/data-end attributes

Normalization

• Converts newlines to \n
• Removes Unicode LS/PS characters
• Collapses excessive blank lines

Slugging Algorithm

Title → lowercase → non-alphanumerics → "_" → trim → max 120 chars
       → Windows reserved-name suffix → collision suffix (_2, _3, ...)

HTML Rendering

• Markdown-it + highlight.js
• Heading slug de-dupe for TOC
• Inline CSS for light/dark/print
• Zero JavaScript

Command Reference

csctf <share-url> [options]

Output Options

GitHub Pages Publishing

Other

Output Format

Markdown Structure

# Conversation: <Title>

**Source:** https://chatgpt.com/share/...
**Retrieved:** 2026-01-08T15:30:00Z

## User

How do I sort an array in Python?

## Assistant

Here's how to sort an array in Python:

Sort in place

my_list.sort()

Return new sorted list

sorted_list = sorted(my_list)

HTML Features

• Standalone — No external dependencies
• Zero JavaScript — Safe for any hosting
• Inline CSS — Light/dark mode via prefers-color-scheme
• Syntax highlighting — highlight.js themes inline
• Table of contents — Auto-generated from headings
• Language badges — Code block language indicators
• Print-friendly — Optimized print styles

Filename Generation

"How to Build a REST API" → how_to_build_a_rest_api.md "Python Tips & Tricks!" → python_tips_tricks.md "File exists already" → file_exists_already_2.md

Rules:

• Lowercase
• Non-alphanumerics → _
• Trimmed leading/trailing _
• Max 120 characters
• Windows reserved names suffixed
• Collisions: _2, _3, ...

GitHub Pages Publishing

Quick Recipe

# Publish with defaults
csctf <url> --publish-to-gh-pages --yes

# Creates: <gh-username>/my_shared_conversations repo
# Branch: gh-pages
# Directory: csctf/

Remembered Settings

# First time: save settings
csctf <url> --publish-to-gh-pages --remember --yes

# Subsequent: just use --yes
csctf <url> --yes

# Clear remembered settings
csctf --forget-gh-pages

Custom Configuration

csctf <url> --publish-to-gh-pages \
  --gh-pages-repo myuser/my-chats \
  --gh-pages-branch main \
  --gh-pages-dir exports \
  --yes

Requirements

• GitHub CLI (gh) installed and authenticated
• Verify with: gh auth status

Publish Flow

1. Resolve repo/branch/dir (use remembered or defaults)
2. Clone (or create via gh)
3. Copy MD + HTML files
4. Regenerate manifest.json and index.html
5. Commit + push
6. Print viewer URL

Recipes

Quiet CI Scrape (MD only)

csctf <url> --md-only --quiet --outfile /tmp/chat.md

HTML-only for Embedding

csctf <url> --html-only --outfile site/chat.html

Slow/Large Conversations

csctf <url> --timeout-ms 90000

Custom Browser Cache

PLAYWRIGHT_BROWSERS_PATH=/opt/ms-playwright csctf <url>

Batch Archive

for url in $URLS; do
  csctf "$url" --outfile ~/archive/ --quiet
done

Security & Privacy

Network Behavior

• Only fetches: The share URL itself
• Opt-in: Update checks, GitHub publish flows
• Auth: GitHub CLI (gh) for publishing—no tokens stored

HTML Safety

• Zero JavaScript in output
• Inline styles only
• Citation pills and data attributes stripped
• highlight.js used statically

Filesystem

• Temp+rename write pattern (atomic)
• Collision-proof naming
• Config: ~/.config/csctf/config.json

Claude.ai Cookies

• Copied to temporary directory only
• Used for single scraping session
• Original Chrome profile never modified

Performance

Characteristics

• Playwright browsers cached after first run
• 60s default timeout, 3-attempt backoff
• Single page/context, linear processing
• Atomic writes prevent partial outputs

Failure Modes & Remedies

Environment Variables

Runtime

Installer

File Locations

Installation

# One-liner (recommended)
curl -fsSL https://raw.githubusercontent.com/Dicklesworthstone/chat_shared_conversation_to_file/main/install.sh | bash

# Pin version
VERSION=v1.0.0 curl -fsSL .../install.sh | bash

# Custom directory
DEST=/opt/bin curl -fsSL .../install.sh | bash

# Verify checksum
curl -fsSL .../install.sh | bash -s -- --verify

From Source

bun install
bun run build
# Binary at dist/csctf

Comparison

Limitations

• Requires public share links (except Claude.ai which uses your session)
• Provider layouts may change (selectors maintained with fallbacks)
• Markdown/HTML exports require share to be available at scrape time
• Claude.ai requires Chrome installed with active login session
• First run downloads Playwright Chromium (~200MB)

Integration with Flywheel