- Home
- Skills
- Productivity
- AI Chat Archive — ChatGPT, Claude, Gemini & Grok to Markdown
More screenshots
Works with the AI Agents you already use
AI Chat Archive — ChatGPT, Claude, Gemini & Grok to Markdown
Copying chats by hand is slow and quietly changes your words: a typo, a stray backslash before a quote, a turn that never loaded.
$5
AI Chat Archive — ChatGPT, Claude, Gemini & Grok to Markdown
Example session with this skill installed
My ChatGPT export finally arrived: ~/Downloads/chatgpt-export.zip. Check the archive I built by reading, in ~/archive/chatgpt, against it and fill in what's missing. Don't overwrite anything.
- Read your context and instructions
- Compiled the ai chat archive
Cross-checked 2 archived chats against the export: 3 copying problems found, 1 chat only in your archive, 1 kept chat added.
Reconcile · chatgpt
| What | Found | What I did |
|---|---|---|
| chatgpt-personal-002, turn 1 | "rivr" where the export has "river" (similarity 0.99) | corrected copy written |
| chatgpt-personal-002, turn 3 | backslash quotes \" from copying | fixed in the same corrected copy |
| chatgpt-personal-002, turn 4 | never loaded on screen | filled in from the export |
| chatgpt-personal-004 | in your archive, not in the export | kept; it may be the only copy |
| survey #3 · Monthly budget formulas | marked keep, not archived yet | added as chatgpt-personal-003.md with its budget notes file |
Waiting for your OK: _reconciled-chatgpt-personal-002.md replaces the original only if you agree. The original is then renamed to _before-reconcile-chatgpt-personal-002.md, so nothing is deleted.
Checked
checkon the folder: 3 files; the only issue is the 2 backslash quotes in 002, which the corrected copy fixes.- Nothing in your archive was overwritten; the export and your chat account were only read.
Run on the bundled fictional samples (mode 3) with convert_export.py. The reading copies are fictional; the results are the commands' actual output.
Connects securely to your tools. The creator never sees your data.
About this skill
Why AI chat history is hard to keep, and what this skill does instead
Years of your work live in ChatGPT, Claude, Gemini, and Grok: drafts you wrote, final images, finished documents, decisions. The official exports are the reliable route, but they arrive as one huge JSON file, sometimes days late and sometimes without some chats or files. Copying chats by hand is slow and quietly changes your words: a typo, a stray backslash before a quote, a turn that never loaded.
This skill gives your local AI agent a complete, checkable method. It surveys every chat first, keeps only what you confirm, copies your own words exactly, keeps the final deliverables, and proves the result against the export. You get plain Markdown on your own disk, one file per conversation, readable in Obsidian, VS Code, or any Markdown viewer.
What you get
- One file per conversation worth keeping: your turns verbatim in fenced blocks, the final canvas or artifact, the files that belong to it, and searchable frontmatter (date, link, category, tags, authorship).
- A survey before anything is archived: every chat numbered oldest first, with a verdict (keep, review, check, let go) and a keep list you confirm.
- Curated, not dumped: the final image among the drafts, plus anything else worth keeping, each with a one-line reason in the report.
- Cross-checking with
reconcile: compares the archive with the export turn by turn, writes corrected copies for you to approve, lists chats the export left out, and adds kept chats that are still missing. - A report for every batch: what was archived, what is missing, and what needs your decision.
Four ways to work
| Your situation | Mode |
|---|---|
| Export in hand, simplest route | 1. Export only |
| Export in hand, files chosen by reading the chats | 2. Export with reading |
| Export on the way, or you want to start now | 3. Reading first, then the export checks and fills in |
| No export, a late one, or you prefer reading | 4. Reading only |
Export late? Start in mode 4 with your most important chats and switch to mode 3 when it arrives.
Verified on the bundled samples
The package ships fictional exports and two finished sample archives, and every command runs on them. In the mode 3 sample, reconcile caught a copying typo ("rivr"), backslash quotes, and a turn that never loaded on screen. It flagged a kept chat the export left out and added a kept chat that reading had not reached, file included. After the corrected copy was accepted, a second run matched every archived chat and check reported no issues. The reading copies in that sample are fictional; the reports in the gallery are the commands' actual output.
How every batch runs
Choose a mode → survey and confirm the keep list → pilot three → archive in batches → check → reconcile with the export → report
Your chat accounts stay read-only. Nothing is overwritten or deleted, and corrections go into separate copies until you accept them.
What's inside
- SKILL.md workflow: modes, ground rules, survey, pilot, batches, cross-check, and the final report
convert_export.py(Python standard library only, no network):inspect,survey,convert,check, andreconcilefor ChatGPT and Claude exports, plus Gemini Takeout inspect and reconcile (experimental)- Curated-reading rules: one chat at a time, page text copied rather than retyped, final versions chosen, pauses, and a stop at any limit
- Export guide for all four services, including what to do when an export is late or incomplete
- Archive format: file names, frontmatter, body, AI outputs, and files, kept simple for Obsidian and other viewers
- Eight copy-ready prompts in English and Korean, covering every mode
- Two sample archives and fictional exports to try every command first
Supported environment and safe boundaries
Works with agents that can read and write local files and run Python 3.8+, such as Claude Code, Codex, Cursor, or Gemini CLI. Reading modes need a browser the agent can use, signed in to your account by you. You request the exports from each service's own settings; the skill requests nothing unless you ask.
Everything stays in folders you choose, and the skill uploads nothing. The AI agent you connect reads the chats it archives, under its provider's data policy. Reading follows each service's limits and terms: one chat at a time, stopping at any rate limit or captcha. Independent project, not affiliated with OpenAI, Anthropic, Google, or xAI.
한국어 안내
흩어진 AI 대화, 원문 그대로 내 컴퓨터에
ChatGPT·Claude·Gemini·Grok에 쌓인 내 글과 결과물을, 읽고 검색할 수 있는 마크다운 아카이브로 정리합니다. 공식 내보내기는 정확하지만 늦게 오거나 일부가 빠질 수 있고, 화면에서 옮겨 적으면 오타나 누락이 생기기 쉽습니다. 이 스킬은 두 방법을 함께 써서 정확하게 남깁니다.
- 네 가지 방식: ① 내보내기만 ② 내보내기와 열람 대조 ③ 열람 선별 후 내보내기로 확인·보완 ④ 열람 선별만. 내보내기가 늦으면 ④로 시작하고, 도착하면 ③으로 넘어갑니다.
- 선별 보관: 조사표로 모든 대화를 먼저 훑고, 확인받은 대화만 보관합니다. 이미지는 최종본과 가치 있는 것만 남기고 그 이유를 보고서에 적습니다.
- 원문 그대로: 내가 쓴 글은 다시 타이핑하지 않고 그대로 옮깁니다. 내보내기와 대조해 오타와 누락을 찾고, 수정본은 따로 만들어 확인받습니다.
- 안전: 원래 계정은 읽기만 하고, 파일은 덮어쓰거나 지우지 않으며, 어디에도 올리지 않습니다.
필요한 것: 로컬 파일을 다루는 AI 에이전트(Claude Code·Codex·Cursor·Gemini CLI 등), Python 3.8 이상(표준 라이브러리만 사용), 공식 내보내기 파일 또는 직접 로그인한 브라우저. OpenAI·Anthropic·Google·xAI와 제휴하지 않은 독립 프로젝트입니다.
Export formats change; the converter inspects the structure first and reports what it does not recognise. Gemini cross-checking is experimental (Takeout has no chat ids) and Grok has no converter yet. Reading copies page text one chat at a time and depends on each service's pages, limits, and terms; it has not been tested against every service's current interface. Some files never arrive in exports and are listed for you to decide. The archive keeps names and contact details exactly as written, so keep it private.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 15 days ago
Requires
Permissions
File Scopes
Terminal runs the bundled converter (Python standard library, no network). It reads only the export you name and writes only new files in the archive folder you choose; it never overwrites or deletes. The browser is used only in the reading modes: your own signed-in browser through the agent's browser tool, read-only and one chat at a time. No uploads, no automatic installation, no binaries. Connected AI providers process chat content under their own policies.