name	chatgpt-share-dump
description	Dump ChatGPT shared conversations from https://chatgpt.com/share/... into agent-readable archives. Use when a user gives an AI agent a ChatGPT share link and asks it to continue, analyze, test, migrate, summarize, implement from, or preserve the full transcript as Markdown, JSON turns, current-state handoff context, and strict media manifests.

ChatGPT Share Dump

You operate chatgpt-share-dump on behalf of the user. The goal is to turn a ChatGPT share URL into an archive an agent can read safely as source context.

Resolve CLI

Resolve the CLI in this order:

node ${CLAUDE_PLUGIN_ROOT}/bin/chatgpt-share-dump.mjs when installed as a Claude plugin.
chatgpt-share-dump when installed by the tracked installer, GitHub install, or otherwise available on PATH.
node /absolute/path/to/bin/chatgpt-share-dump.mjs when working from a cloned repository.

Store the working command and reuse it for the rest of the session.

Workflow

Choose an output directory. Use an absolute path when possible.
Run:

$CLI --url "https://chatgpt.com/share/..." --out "/absolute/output/dir" --zip

If the share contains uploaded or generated media, rerun with the complete local assets folder:

$CLI --url "https://chatgpt.com/share/..." --out "/absolute/output/dir" --assets-dir "/absolute/assets/folder" --zip

The media policy is strict. If media is present and --assets-dir is missing, incomplete, or has files whose byte size does not match the share payload, the CLI writes transcript files and MISSING_MEDIA.md, exits non-zero, and does not create a success zip.

Output Contract

Read these files for downstream agent work:

visible-turns.json - canonical structured visible turns.
CHAT_CONTEXT_FOR_CODEX_FULL.md - current snapshot plus full transcript for agent context.
CURRENT_SNAPSHOT.md - latest-turn state and coverage.
full-transcript.md - readable Markdown transcript.
assets_manifest.json / assets_manifest.csv - media rows and local validation state.
MISSING_MEDIA.md - exact missing/invalid media rows when strict media validation fails.

Trust Rules

Do not scrape the visible ChatGPT UI; long conversations can be virtualized and incomplete.
Treat visible-turns.json as canonical for extracted visible turns.
Exclude hidden system/tool payloads, hidden code/tool calls, and hidden reasoning.
Preserve visible thought recap labels when present.
Do not claim a complete archive unless the command exits 0.
Install-source tracking is content-blind. The CLI may send one source + version beacon per install source/version; it must never send share URLs, transcript text, raw HTML, output paths, asset filenames, local usernames, or credentials.

ChatGPT Share Dump

You operate chatgpt-share-dump on behalf of the user. The goal is to turn a ChatGPT share URL into an archive an agent can read safely as source context.

Resolve CLI

Resolve the CLI in this order:

node ${CLAUDE_PLUGIN_ROOT}/bin/chatgpt-share-dump.mjs when installed as a Claude plugin.

chatgpt-share-dump when installed by the tracked installer, GitHub install, or otherwise available on PATH.

node /absolute/path/to/bin/chatgpt-share-dump.mjs when working from a cloned repository.

Store the working command and reuse it for the rest of the session.

Workflow

Choose an output directory. Use an absolute path when possible.

Run:

$CLI --url "https://chatgpt.com/share/..." --out "/absolute/output/dir" --zip

If the share contains uploaded or generated media, rerun with the complete local assets folder:

$CLI --url "https://chatgpt.com/share/..." --out "/absolute/output/dir" --assets-dir "/absolute/assets/folder" --zip

Output Contract

Read these files for downstream agent work:

visible-turns.json - canonical structured visible turns.

CHAT_CONTEXT_FOR_CODEX_FULL.md - current snapshot plus full transcript for agent context.

CURRENT_SNAPSHOT.md - latest-turn state and coverage.

full-transcript.md - readable Markdown transcript.

assets_manifest.json / assets_manifest.csv - media rows and local validation state.

MISSING_MEDIA.md - exact missing/invalid media rows when strict media validation fails.

Trust Rules

Do not scrape the visible ChatGPT UI; long conversations can be virtualized and incomplete.

Treat visible-turns.json as canonical for extracted visible turns.

Exclude hidden system/tool payloads, hidden code/tool calls, and hidden reasoning.

Preserve visible thought recap labels when present.

Do not claim a complete archive unless the command exits 0.

Install-source tracking is content-blind. The CLI may send one source + version beacon per install source/version; it must never send share URLs, transcript text, raw HTML, output paths, asset filenames, local usernames, or credentials.