Full Claude Chat Export — Feature Plan¶

Status: Content features built and tested (v0.2.0 — three buttons, ZIP export, progress modal, timestamps). Auto-backup and message warnings moved to Persist extension (private repo, state management only). Date: March 22, 2026 Author: Nikhil Singhal + Claude Code

The Problem¶

When exporting a Claude.ai conversation, the current Copier extension captures only the chat messages (user prompts + AI responses). But Claude conversations often include artifacts — documents, code files, presentations, and other generated content. These artifacts are a critical part of the conversation output. Exporting without them gives you an incomplete record.

Additionally, users upload files to Claude (PDFs, images, text files) that provide context for the conversation. Without these, the exported conversation references attachments that don't exist in the export.

Currently, Claude has a "Download all" button for artifacts, but it: - Downloads artifacts disconnected from the conversation - Gives you a ZIP of files with no context about which message produced which artifact - Requires manual effort to associate artifacts with the conversation

Goal: One click, complete archive. The conversation AND all its artifacts in a single download.

What We Learned (API Investigation)¶

Investigation conducted on March 22, 2026 by intercepting Claude.ai network requests.

API Endpoints (all authenticated via session cookies, no API keys needed)¶

1. Get organization ID:

GET /api/organizations
Returns: [{uuid: "5bea4f9d-...", name: "Nikhil Singhal", ...}]

The extension runs on claude.ai, so it has the session cookies. First element's uuid is the org ID.

2. List all files in a conversation:

GET /api/organizations/{orgId}/conversations/{chatId}/wiggle/list-files?prefix=
Returns: {success: true, files: ["/mnt/user-data/uploads/...", "/mnt/user-data/outputs/..."]}

Files are separated by path prefix: - /mnt/user-data/uploads/ — files the user uploaded to Claude (PDFs, images, text, etc.) - /mnt/user-data/outputs/ — artifacts Claude generated (MD, code, DOCX, PPTX, etc.)

3. Download a single file:

GET /api/organizations/{orgId}/conversations/{chatId}/wiggle/download-file?path={urlEncodedPath}
Returns: The raw file content (binary or text)

The path parameter is URL-encoded, e.g., %2Fmnt%2Fuser-data%2Foutputs%2Fch02-a-ball-on-a-screen.md

Key Observations¶

The chatId is available from the URL: claude.ai/chat/{chatId}
The orgId requires one API call to /api/organizations
All endpoints use the same session authentication (cookies)
The list-files endpoint returns ALL files (uploads + outputs) in one call
File downloads are individual (one request per file)
In a test conversation with 52 artifacts and 18 uploads (70 total files), all APIs responded successfully
No rate limiting observed on these endpoints

DOM Structure for Artifacts¶

Artifacts in the conversation DOM are rendered as .artifact-block-cell elements with: - Artifact name (e.g., "Ch01 paper reams v5") - Type badge (e.g., "Document") - Format indicator (e.g., "MD") - Individual Download button - The parent element has role="button" (clickable to open in artifact panel)

The "Download all" button exists in the artifacts sidebar panel.

Design¶

User Flow¶

Current behavior (no artifacts): User clicks Download button. Gets a single HTML file with the conversation.

New behavior (with artifacts): User clicks Download button. 1. Extension detects this is claude.ai (artifacts are Claude-only) 2. Calls list-files API to check for artifacts 3. If no artifacts: same as current behavior (single HTML file) 4. If artifacts exist: - Shows a brief "Exporting conversation + {N} files..." toast - Downloads all artifact files via the API - Builds a ZIP containing:

recurate-export-{slug}-{date}/
├── conversation.html        (styled conversation with artifact manifest)
├── artifacts/               (Claude-generated files)
│   ├── ch01-paper-reams.md
│   ├── presentation.pptx
│   └── ...
└── uploads/                 (user-uploaded files)
    ├── research-paper.pdf
    ├── screenshot.png
    └── ...

- The HTML file includes an "Artifacts" section at the end listing all files with relative links to artifacts/filename and uploads/filename - Triggers browser download of the ZIP

HTML Artifact Manifest¶

The conversation HTML includes a new section at the bottom:

<div class="artifacts-section">
  <h2>Artifacts</h2>
  <p>Claude generated 51 artifacts during this conversation:</p>
  <ul>
    <li><a href="artifacts/ch01-paper-reams.md">ch01-paper-reams.md</a> (Document, MD)</li>
    <li><a href="artifacts/presentation.pptx">presentation.pptx</a> (Presentation, PPTX)</li>
    ...
  </ul>

  <h2>Uploaded Files</h2>
  <p>18 files were uploaded during this conversation:</p>
  <ul>
    <li><a href="uploads/research-paper.pdf">research-paper.pdf</a></li>
    <li><a href="uploads/screenshot.png">screenshot.png</a></li>
    ...
  </ul>
</div>

When opened from the ZIP, the links work because the files are in the same folder structure.

Dependencies¶

JSZip — lightweight ZIP library (~100KB minified). Needed to create ZIP files in the browser. No server-side processing.

Options for including JSZip: - Bundle it — include the minified JS in the extension. Adds ~100KB to the extension size. - Load from CDN — not an option for extensions (CSP restrictions).

Recommendation: Bundle it. The Copier is currently ~32KB. Adding JSZip brings it to ~132KB. Still small.

Platform Scope¶

Claude.ai only — this feature uses Claude's internal API. Other platforms (ChatGPT, Grok, etc.) don't have the same artifact system.
Other platforms — continue to work as they do today (HTML file only, no artifact export).
The existing platform detection in the Copier already handles this. The artifact export code only runs when getPlatform() === 'claude'.

Implementation Plan¶

Phase 1: Core ZIP Export¶

Add JSZip dependency — bundle the minified JS or install via npm. But Copier is vanilla JS with no build step. Options:
Copy JSZip minified source into the extension directory
OR convert Copier to a build step (WXT like Connect). This is a bigger change.
Recommendation: Copy JSZip minified source. Keep Copier as vanilla JS.
Add API helper functions:
getOrgId() — calls /api/organizations, returns first org UUID
listFiles(orgId, chatId) — calls wiggle/list-files, returns file list
downloadFile(orgId, chatId, path) — calls wiggle/download-file, returns blob
Modify downloadHTML() function:
After extracting conversation, check platform
If Claude: call listFiles() to check for artifacts
If artifacts exist: download all files, build ZIP, trigger ZIP download
If no artifacts: current behavior (single HTML)
Update toHTML() function:
Accept optional artifact/upload file lists
If present, append the artifact manifest section to the HTML
Progress modal (not toast):
Semi-transparent overlay blocks the page during export
Centered modal with spinner, title ("Exporting conversation"), status text, progress bar, file count
Progress bar fills as files download
Cannot be dismissed until complete (prevents user from interfering)
On completion: checkmark icon, "Export complete", auto-dismisses after 2.5 seconds
On failure: X icon, error message, auto-dismisses, falls back to simple HTML download

Phase 2: Polish¶

Handle download failures gracefully (skip failed files, note in manifest)
Handle large binary files (images, PPTX) — stream to ZIP without loading full content in memory
Test with large conversations (1.4GB+ like Urmila's)
Test with various artifact types (MD, DOCX, PPTX, images, code)

Decisions (March 22, 2026)¶

User uploads: Include everything (images, PDFs, text files). The goal is a complete archive.
Naming conflicts: If multiple files have the same name, append a counter (filename-2.md). API paths already include version info (ch01-paper-reams-V5.md) so most are naturally unique.
Download buttons: ~~Same button, smarter behavior.~~ Updated March 22: separate Download (quick HTML) and Export (full ZIP) buttons. Export button uses a ZIP icon and only appears on Claude.ai.
Progress feedback: ~~Toast message.~~ Updated March 22: blocking progress modal with spinner, progress bar, file count. Auto-dismisses on completion.
Build approach: Full feature in one pass. No intermediate versions.

Inline Artifact Links (Point 2)¶

When Claude presents an artifact inline in a response (e.g., "Ch02.5 a note from claude / Document / MD / Download"), the exported HTML should show a clickable link to the actual file in the artifacts/ folder at that exact position in the conversation.

DOM-to-File Matching¶

The artifact block in the DOM has aria-label="Open artifact: Ch01 paper reams". The API file path is /mnt/user-data/outputs/ch01-paper-reams-V5.md. Matching approach:

Extract artifact name from DOM (aria-label on parent of .artifact-block-cell)
Slugify the name: "Ch01 paper reams" becomes ch01-paper-reams
Find matching files in the API file list by prefix match
If multiple versions exist, link to the latest (highest version number)
Replace the artifact block in the cloned DOM with a styled link to artifacts/filename

HTML Output (inline)¶

Where Claude presented an artifact, the exported HTML shows:

<div class="artifact-link">
  <a href="artifacts/ch01-paper-reams-V5.md">Ch01 paper reams</a>
  <span class="artifact-type">Document, MD</span>
</div>

The full manifest at the bottom still lists ALL artifacts and uploads with links.

Phase 3: Three-Button UX + Auto-Backup (Next)¶

Status: Designed, not yet built¶

Three-Button Design¶

The current single download button conflates "quick HTML" and "full ZIP with artifacts." These are different operations with different time costs (instant vs ~60 seconds). Users need separate controls.

Button row on Claude.ai:

Button	Icon	Action	Time
Copy	Clipboard	Markdown to clipboard	Instant
Download	Arrow (existing)	Quick HTML only, no API calls	<1 second
Export	ZIP/archive icon	Full ZIP with conversation + artifacts + uploads + progress modal	~60 seconds

Button row on other platforms (ChatGPT, Grok, Gemini, Copilot, Google AI):

Button	Icon	Action
Copy	Clipboard	Markdown to clipboard
Download	Arrow	HTML file

The Export (ZIP) button only appears on Claude.ai because artifacts are Claude-only.

Auto-Backup (Periodic HTML Snapshots)¶

The problem: When a Claude chat hits capacity and becomes non-responsive, everything in that chat is lost unless the user manually exported beforehand. Urmila lost 615 messages and 90 minutes recovering. Non-technical users won't remember to click Export.

The solution: The Copier automatically saves a snapshot of the conversation HTML every 2 hours. No user interaction required. Just the HTML (no artifacts, no API calls), so it's fast and silent.

Design: - Uses chrome.alarms API to trigger every 2 hours (configurable) - Background service worker sends message to active Claude.ai tab content script: "export now" - Content script extracts conversation and saves HTML silently via chrome.downloads API with saveAs: false - Filename: backup-{chat-name}-{date}-{time}.html - Only exports if the Claude.ai tab is open and has messages - Shows a brief, non-blocking notification: "Conversation backed up (423 messages)" - No modal, no blocking, no user action required

Message count warning: - At 400+ messages, show a one-time warning: "This conversation has 400+ messages. Long conversations may hit Claude's capacity limit. Consider exporting with the Export button." - Dismissible. Does not block. - Re-shows if message count crosses 500, 600, etc.

Scope: - Claude.ai only (other platforms don't have the capacity problem) - HTML only (no artifacts) for speed and simplicity - Requires adding chrome.alarms, chrome.downloads, and a background service worker to the Copier (currently has none)

Relationship to V3 State Management: This auto-backup is one piece of the larger state persistence puzzle being designed in ns-2026-the-big-push/ws-ai-partnership-command-center/v3-state-management/PROBLEM-DEFINITION.md. It addresses failure mode #1 (chat hits capacity) and #3 (human forgets to trigger state save) from that document. It does NOT solve conversational context preservation (that requires a different approach).

Files to Modify¶

File	Change
`extensions/conversation-copier/content.js`	Add API helpers, modify downloadHTML, update toHTML
`extensions/conversation-copier/manifest.json`	No change needed (already has claude.ai host permission)
`extensions/conversation-copier/jszip.min.js`	New file — bundled JSZip library

Testing Plan¶

Conversation with MD artifacts — verify MD files download correctly and are readable
Conversation with binary artifacts (DOCX, PPTX) — verify binary files are intact in ZIP
Conversation with uploaded images — verify images download and display correctly
Conversation with no artifacts — verify current behavior unchanged (single HTML)
Large conversation (600+ messages, 50+ artifacts) — verify performance and memory
Non-Claude platforms — verify ChatGPT, Grok, etc. still work as before (HTML only)
ZIP structure — verify relative links in HTML work when opened from extracted ZIP