Export Project

Produces a structured export package containing all project knowledge, formatted for the target platform. No skills are included in exports.

Supported platforms

Step 0 — Select platform

If the user hasn't already specified a platform, ask:

"Which platform are you exporting for — Claude.ai, Claude Code, ChatGPT, or Cursor?"

Use their answer to determine output structure in Step 2.

Step 1 — Collect static assets

python /mnt/skills/user/export-project/scripts/bundle.py \
  --staging-dir /tmp/project-export-stage

Wait for output. Note what was found (uploads count, transcripts count).

Step 2 — Write platform content

Platform: claude

Output structure:

project-export/
├── IMPORT.md
├── instructions.md
├── chat_history.md
└── uploads/           ← already staged by bundle.py

Write instructions.md from your system prompt / project instructions context:

# Project Instructions

> Exported from: <project name if known, otherwise "Claude.ai project">
> Exported at: <timestamp from meta.json>

<verbatim instructions text>

Write chat_history.md — see Chat History section below.

Write IMPORT.md:

# Project Export — Claude.ai Import Guide

Exported: <timestamp>

## Steps
1. Create a new project in Claude.ai
2. Paste `instructions.md` content into Settings → Project Instructions
3. Upload all files from `uploads/` to the project Files section
4. Upload `chat_history.md` to the project Files section
5. Start a conversation and ask Claude to summarise what it knows

## Notes
- Sensitive values (API keys, secrets) may be present — review before sharing.
- Chat history is a lossy compaction; verbatim exchanges are not preserved.

Platform: claude-code

Output structure:

project-export/
├── CLAUDE.md          ← single consolidated context file; drop in project root
└── uploads/           ← reference files, if any

Write CLAUDE.md — merge instructions + chat summary into one file:

# Project Context

> Platform: Claude Code
> Exported at: <timestamp>

## Instructions
<verbatim project instructions>

## Chat History (Compacted)
<compact summary — same format as chat_history.md below>

## Reference Files
<list any files in uploads/ with one-line description each>

No IMPORT.md needed — just drop CLAUDE.md in the project root and place uploads/ alongside it or reference them as needed.

Platform: chatgpt

Output structure:

project-export/
├── IMPORT.md
├── system_prompt.md
└── knowledge/
    ├── project_context.md
    └── chat_history.md

Write system_prompt.md — distill instructions into a ChatGPT-style system prompt:

# System Prompt

<Rewrite project instructions as a direct, second-person system prompt.
 E.g. "You are an assistant working on X. Your conventions are Y. Always do Z.">

Write knowledge/project_context.md — key project context as a knowledge document:

# Project Context

> Exported at: <timestamp>

## Overview
<1–3 paragraph summary of what this project is and its current state>

## Key Decisions
<bullet list of architectural/design decisions>

## Conventions
<bullet list of naming, style, structural conventions>

Write knowledge/chat_history.md — see Chat History section below.

Write IMPORT.md:

# Project Export — ChatGPT Import Guide

Exported: <timestamp>

## Steps
1. Open ChatGPT → Your Projects (or GPT Builder for a custom GPT)
2. Paste `system_prompt.md` content into the Instructions field
3. Upload both files from `knowledge/` as knowledge files
4. Start a conversation and ask it to summarise the project

## Notes
- ChatGPT knowledge files are searched, not always fully loaded — keep them concise.
- Sensitive values may be present — review before sharing.

Platform: cursor

Output structure:

project-export/
├── IMPORT.md
├── .cursorrules
└── context/
    └── chat_history.md

Write .cursorrules — combine instructions + key conventions into Cursor rules format:

# Project Rules

## Overview
<2–3 sentence summary of what this project is>

## Instructions
<project instructions, rewritten as imperative rules>

## Conventions
<bullet list of coding, naming, structural conventions>

## Key Decisions
<bullet list of important prior decisions the AI should respect>

Write context/chat_history.md — see Chat History section below.

Write IMPORT.md:

# Project Export — Cursor Import Guide

Exported: <timestamp>

## Steps
1. Drop `.cursorrules` into your project root
2. Optionally add `context/chat_history.md` to your project for additional background
3. Open Cursor — it will automatically pick up `.cursorrules`

## Notes
- `.cursorrules` is loaded on every request; keep it focused and avoid bloat.
- Sensitive values may be present — review before sharing.

Chat History format (all platforms)

Check /tmp/project-export-stage/transcripts_raw/ for transcript files.

If transcripts exist, produce a compact information-dense summary:

# Chat History (Compacted)

> Sessions: N  |  Compacted at: <timestamp>

## Session <date>

### Decisions
- <bullet per architectural/design decision>

### Work Done
- <bullet per completed task with key outcome>

### Key Artifacts
- <file/path created or modified — one-liner>

### Open Items
- <anything left explicitly unresolved>

Rules:

• Omit pleasantries, clarifications, tool output verbatim dumps
• Preserve: decisions, rationale, file paths, error resolutions, open questions
• Each session ≤ 50 lines unless unusually dense

If no transcripts exist, write:

# Chat History (Compacted)

No prior chat history found at export time.

Step 3 — Zip and present

python /mnt/skills/user/export-project/scripts/zip_stage.py \
  --staging-dir /tmp/project-export-stage \
  --output /mnt/user-data/outputs/project-export-<platform>.zip

Then call present_files with the zip path.

Edge cases