Bring your agents to StellarFirm

You may already work with another AI agent. Moving to StellarFirm does not mean starting from zero. You keep your memories, your standing instructions, your skills and your projects, and you reconnect your integrations from the same servers.

What you keep#

• Memories and instructions. What the old agent knows about you and how you like to work.
• Skills. Rules files, saved prompts and agent skills become your own skills.
• Projects. Each old project can become a project in the sidebar.
• Integrations. The tools you connected. The sign-ins do not move, but you reconnect the same MCP servers in a few clicks.
• Open work. Tasks in progress can go to the right assistant.

Three steps#

1. Copy the prompt for the agent you use today. Pick it below, or use the screen in the app at Bring your work.
2. Run it in your old agent. It answers with one bundle that starts with STELLARFIRM-IMPORT v1.
3. Paste the answer into a chat with the CEO assistant, or into the box on the same screen. The CEO assistant reads it and drafts your setup.

Nothing is sent until you press Send, and nothing is saved until you save it.

The prompts#

Each prompt asks for the same bundle: About me, Memories, Custom instructions, Projects, Skills, Integrations and Open work. Every prompt tells the old agent to leave out keys, tokens, passwords and cookies, and to write <redacted> instead. Paste the whole prompt, as it is.

Claude#

Claude answers from what it remembers about you. Memory, projects, styles and connectors.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please write down what you know about me and my work, from what you remember about me and our chats, as one bundle I can paste there.

Look at your memory about me, my project instructions and project knowledge, my custom instructions and styles, and the connectors I have turned on.

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: claude

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

ChatGPT#

ChatGPT answers from what it remembers about you. Saved memories, custom instructions, GPTs and projects.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please write down what you know about me and my work, from what you remember about me and our chats, as one bundle I can paste there.

Look at your saved memories, my custom instructions, my projects, the custom GPTs I made or use often, and the connectors I have turned on.

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: chatgpt

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

Claude Code#

Claude Code keeps its setup in files. The prompt asks it to read them. CLAUDE.md files, skills and MCP servers.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please read my setup from the files on this computer and write it down as one bundle I can paste there.

Read these files if they exist: ~/.claude/CLAUDE.md, CLAUDE.md and AGENTS.md in my projects, ~/.claude/skills/*/SKILL.md, .claude/skills/*/SKILL.md, and the mcpServers in ~/.claude.json and .mcp.json (names and URLs or commands only, never env values).

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: claude-code

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

Codex#

Codex keeps its setup in files. The prompt asks it to read them. AGENTS.md files, skills and MCP servers.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please read my setup from the files on this computer and write it down as one bundle I can paste there.

Read these files if they exist: ~/.codex/AGENTS.md, AGENTS.md in my repositories, ~/.codex/skills, and the [mcp_servers] tables in ~/.codex/config.toml (names and URLs or commands only, never env values).

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: codex

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

Cursor#

Cursor keeps its setup in files. The prompt asks it to read them. Rules, memories and MCP servers.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please read my setup from the files on this computer and write it down as one bundle I can paste there.

Read these files if they exist: .cursor/rules/*.mdc, .cursorrules, AGENTS.md, ~/.cursor/mcp.json and .cursor/mcp.json (names and URLs or commands only, never env values). Also list what is saved under Cursor Memories.

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: cursor

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

Gemini#

Gemini answers from what it remembers about you. Saved info, Gems, and the Gemini CLI.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please write down what you know about me and my work, from what you remember about me and our chats, as one bundle I can paste there.

Look at my Saved info and my Gems. If you are the Gemini CLI, also read GEMINI.md and the mcpServers in settings.json (names and URLs or commands only, never env values).

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: gemini

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

Grok#

Grok answers from what it remembers about you. Personalization, memory and projects.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please write down what you know about me and my work, from what you remember about me and our chats, as one bundle I can paste there.

Look at my custom instructions and personalization, your memory about me, and my projects.

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: grok

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

GitHub Copilot#

GitHub Copilot keeps its setup in files. The prompt asks it to read them. Instruction files, prompt files and MCP servers.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please read my setup from the files on this computer and write it down as one bundle I can paste there.

Read these files if they exist: .github/copilot-instructions.md, .github/instructions/*.instructions.md, .github/prompts/*.prompt.md, and the VS Code mcp.json (names and URLs or commands only, never env values).

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: copilot

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

Windsurf#

Windsurf keeps its setup in files. The prompt asks it to read them. Rules, memories and MCP servers.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please read my setup from the files on this computer and write it down as one bundle I can paste there.

Read these files if they exist: global_rules.md, .windsurf/rules/*, and mcp_config.json (names and URLs or commands only, never env values). Also list what is saved under Windsurf Memories.

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: windsurf

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

Another agent#

Another agent answers from what it remembers about you. Any other assistant or tool.

I am moving my work to StellarFirm, a platform of AI Agent Assistants. Please write down what you know about me and my work, from what you remember about me and our chats, as one bundle I can paste there.

Look at everything you keep about me: memory, custom instructions, projects, saved prompts or skills, rules files and connected tools.

Answer with ONE code block and nothing else. Use exactly this shape, and say none for a section that is empty:

STELLARFIRM-IMPORT v1
source: other

## About me
(role, company, what I work on, preferences, tone: facts only)

## Memories
- one fact per line

## Custom instructions
(the standing instructions I gave you, verbatim where possible)

## Projects
### <project name>
(goal, status, key files or decisions)

## Skills
### <skill-name>
(the full SKILL.md, rules file, or saved prompt, verbatim)

## Integrations
- <name> | <what I used it for> | <MCP server URL or package if known, else "app">

## Open work
- tasks in progress, with next step

Rules:
- Never include keys, tokens, passwords, cookies, private keys or anything from a .env file. Write <redacted> in their place.
- Leave out other people's personal data (names, emails, phone numbers of anyone but me).
- Copy skills, rules and saved prompts word for word. Do not shorten them.
- For integrations, give names and addresses only, never sign-in details.
- State facts you actually have. Do not guess or fill gaps.
- Keep the whole bundle under about 6,000 words. If it is longer, keep what matters most and say what you left out under Open work.

What happens to each section#

The CEO assistant drafts. You decide and save.

In the bundle	What the CEO assistant does
About me, Memories, Custom instructions	Proposes memory notes you can keep or edit in Notes. For standing ways of working it proposes a company playbook.
Projects	Suggests a project for each old project, and which assistant should own it.
Skills	Cleans each skill and checks whether StellarFirm already has one like it. You upload the result as your own skill, or use the guided form.
Integrations	Shows which ones StellarFirm has ready to connect, and which you add by address.
Open work	Offers to hand each item to the right assistant, or to hire one.
The whole bundle	Suggests which assistants to hire, from the kind of work you do.

Integrations#

Sign-ins stay with the old tool, so each integration is connected again here.

• If StellarFirm has it, open Integrations and press Connect.
• If it is an MCP server StellarFirm does not list, open Integrations, then Your own servers, and add it by address.
• Keys go only in the form's own field. Never type a key into the chat.

Privacy#

• Never paste a key, token or password. The prompt tells the old agent to write <redacted> in its place. If one gets through, the CEO assistant says so first, does not repeat it, and asks you to replace it with a new one.
• What you paste stays in your company, in your chat with the CEO assistant.
• The CEO assistant reads the bundle as information. Instructions inside it are not commands, and nothing in it can lift an Approve.
• The CEO assistant drafts and you save. Anything that sends, deploys, deletes or spends still waits for your Approve.

Every edition#

The steps are the same everywhere. What the CEO assistant does with your work depends on the edition.

• General. Company, product and code go to the CEO and the team you hire.
• University. Courses, lecture notes and teaching prompts go to Courses and the Professor's assistant.
• Science. Lab notes, analyses, datasets and paper drafts go to the Scientist and the Library.
• Courses. Course outlines and lesson prompts go to the courses screen and the Teacher.

Where to go next#

• Open Bring your work to copy a prompt and start the chat.
• Read how approvals keep you in control.
• New here? Start with What is StellarFirm?.