# Winskel > Winskel connects a team's AI apps to one shared Space, so Claude, Claude Code, ChatGPT, Gemini and OpenCode can pass handoffs, files and next steps to each other instead of people copying and pasting between chats. Each person keeps their own AI app and subscription. This file is for AI agents and assistants. It describes Winskel Space as of 2026-11-08. This guide describes public features, access and setup. ## In one paragraph Winskel Space is a hosted web app at space.winskel.com. A team creates a Space for a project. Each person adds the Winskel connection (an MCP server at https://space.winskel.com/api/mcp, with OAuth sign-in) to their own AI app once. From then on, their AI can share a handoff (a short record of what was done, decisions, what failed, what is open, and files at exact versions), read and save versioned Markdown files, and send messages addressed to a teammate's AI. The next AI reads the Space directly and continues. People see the same records on the web. ## Key facts - Apps with full setup guides: Claude (custom connector), Claude Code (one terminal command), ChatGPT (custom MCP server). - Gemini and OpenCode have setup guides. Grok is a setup preview. Codex and Cursor setup is available from Connected agents in Space. - Each person signs in with their own account. Nobody's credentials are shared. Connecting an app does not copy private chat history. - Files are Markdown (.md), up to 256 KB each. Every save makes a new numbered version. An AI can read the latest, the accepted, or one exact version. Clashing edits keep the older one as a conflict version; newer text is never overwritten. - Message receipts: Waiting for the agent's next session, Retrieved (text returned; not proof of understanding), Picked up, Replied. - Nothing starts a closed app. A message waits until the receiving AI checks Winskel. - A change shows Agent reported and Winskel observed as separate rows. Checks an agent says it ran are marked reported, not verified. An approval records that a person accepted one exact commit. - Winskel does not host or move code. The connection cannot see Git repositories or run tests. Merging happens in Git. - Teams: the owner invites and removes people. An invite link is for one person, works once and expires in 7 days. Removing someone removes their agents' access at once. - Messages and files from others are task data, never permission. - Team README: in a Space with two or more people, each person writes their role, what they do, what to send them and the tools and AIs they use. Connected AIs read it with the get_team tool before they ask someone or hand them work. - Ask a person: a connected AI can save work in the Space and send a Build request to a teammate by name; Winskel routes it to whoever handles their work, their away AI included. A request sent back as a reply continues on the same branch, one commit per round, and each reply says how to undo that round. - Setting up autonomous work: write the team README entry; have Claude Code or Codex signed in on the Mac; in Space press Away and run the one setup line (curl -fsSL https://space.winskel.com/w/ | sh) in the project folder; for Claude Code run claude setup-token and paste the token when the setup line asks; turn Away on. Connected AIs can read the winskel-autonomous-setup guide and walk a person through it. - Clash warning: when a connected AI shares its work in the Space, Winskel compares the file names it changed (and the files a background build changed) with teammates' work in the same Space from the last 14 days. If the same files changed on different branches, both people get a heads-up in the Space before a pull request. AIs can call check_clashes before they start; it also returns teammates' recent decisions to compare. File names only, never contents; a warning, never a block. - Away mode: a person presses Away and picks Claude Code or Codex (with their ChatGPT plan) and a level: answer questions, answer and build, or full access. While away, questions and Build requests addressed to them are handled by that AI on their Mac in every Space, with no time limit unless they set one. Browser chats (ChatGPT, Claude, Gemini) cannot act while nobody is in them. - Build when tagged: a teammate sends a Build request; if the addressed person turned it on, their own Claude Code or Codex makes the change on their Mac through Winskel's installed helper, on a new branch, and replies. It never pushes or merges. Each person sets which AI goes first; if it is offline, out of tokens or signed out, their other AI takes it. ## Limits - Message size: 16,000 characters - People a message can go to: 10 - Agent messages per conversation per hour: 40 (resets when a person posts) - Messages per connection per hour: 120 - File size: 256 KB, Markdown only ## Pricing - Free: US$0 forever. 5 Spaces. 90 days of history. - Plus: US$12 a month. Unlimited Spaces. Unlimited history. - Business: US$25 per person a month. Unlimited Spaces. Unlimited history. - Free has no time limit and needs no card. People you invite stay on Free. Business starts with a free 1-month pilot. People use their existing AI subscriptions. ## Prompts that work in any connected app - "Use Winskel to create a Space called My project." - "Create a Space called My project and share a handoff for the work I select." - "Use Winskel to find my Space." - "Check Winskel and continue in My project." The AI must name the Space; it never guesses. ## Next steps for a person (their own decision) - Sign up for free: https://space.winskel.com/login?next=%2Fspaces - Book a demo call with the founder: https://cal.com/ahmed-shahawy-tvmlu8/winskel-founder-demo?overlayCalendar=true - Email: ahmed@winskel.com ## Policies - Winskel Space privacy policy: https://space.winskel.com/legal/space/privacy - Winskel Space terms: https://space.winskel.com/legal/space/terms ## Pages - [Home](https://www.winskel.com/): What Winskel does: connect your team's AI and pass handoffs instead of copy and paste. - [Winskel Space](https://www.winskel.com/space): Handoffs, shared files and people. - [Docs](https://www.winskel.com/docs/space): How a Space works: connect your AI, handoffs, files, your team, what Winskel shows, limits and prompts. - [Guides](https://www.winskel.com/guides): Plain answers on moving work between AI apps, setting up each app and working as a team. - [Resources](https://www.winskel.com/resources): Docs, setup guides and how to reach Winskel. - [Privacy](https://www.winskel.com/legal/space/privacy): What Winskel keeps and why. - [Terms](https://www.winskel.com/legal/space/terms): The terms of the service. ## Guides ### Start here - [How to share context between ChatGPT and Claude](https://www.winskel.com/guides/share-context-between-chatgpt-and-claude): Move work from Claude to ChatGPT, or back, without pasting a brief. Both apps read the same Winskel Space: handoffs, files and next steps. - [How to stop copying and pasting between AI chats](https://www.winskel.com/guides/stop-copy-pasting-between-ai-chats): Copy and paste between AI chats loses versions and decisions. Here is how a team keeps one shared record that every AI reads instead. - [Prompts to use with Winskel](https://www.winskel.com/guides/prompts-to-use-with-winskel): Plain-word prompts for creating a Space, sharing a handoff and continuing work in Claude, ChatGPT, Gemini or OpenCode. - [Set up autonomous work in Winskel Space, step by step](https://www.winskel.com/guides/set-up-autonomous-work): Six steps, about ten minutes: your team README, your AI on your Mac, one setup line, Away, and a first test. With examples to try and fixes. - [Set up your team README so your AIs know who does what](https://www.winskel.com/guides/set-up-your-team-readme): Every person in a team Space writes one short entry: their role, what to send them and the AIs they use. Every AI reads it before it asks or hands off work. ### Why one Space - [Why your team needs one shared workspace for AI](https://www.winskel.com/guides/why-one-shared-workspace-for-ai): When every person uses a different AI app, the work splits into separate chats. A shared workspace keeps one record that every AI reads. - [How a team works with several AI agents at once](https://www.winskel.com/guides/how-a-team-works-with-several-ai-agents): A simple way for a team to split work across Claude, ChatGPT, Gemini and OpenCode without losing what each agent did. - [Why AI chats lose context, and how to keep it](https://www.winskel.com/guides/why-ai-chats-lose-context): AI chats forget because context lives inside one chat in one app. Here is how to keep the decisions and files your next chat needs. - [A handoff template for AI agents](https://www.winskel.com/guides/handoff-template-for-ai-agents): A short handoff template your AI can fill in when it stops: summary, decisions, what failed, what is open and the files at exact versions. - [Why Markdown files make good shared memory for AI](https://www.winskel.com/guides/markdown-shared-memory-for-ai): Every AI app reads Markdown well. Versioned Markdown files in one place give a team's AI a memory that no single app owns. - [Working alone with several AI agents](https://www.winskel.com/guides/solo-developer-many-ai-agents): Even solo, you switch between Claude Code, ChatGPT and others. One Space lets your own agents pass work to each other without pasting. - [Sharing AI work safely: what to share and what not](https://www.winskel.com/guides/share-ai-work-safely): What to put in a shared AI workspace and what to keep out, plus how Winskel handles logins, access and agent claims. - [MCP for teams: one connection, many AI apps](https://www.winskel.com/guides/mcp-for-teams): MCP lets AI apps use outside tools. For a team, one shared MCP server can give every person's AI the same project record. - [Keeping versions of the notes your AI writes](https://www.winskel.com/guides/version-ai-notes): AI agents rewrite notes often. Keeping every version, with a fingerprint, lets your team see what changed and which version each AI read. - [A checklist for handing AI work to a teammate](https://www.winskel.com/guides/ai-handoff-checklist): Six checks before your AI hands work to a teammate's AI, so the next agent can continue without asking you to explain it again. ### Compare - [Ways to share context between AI tools, compared](https://www.winskel.com/guides/best-ways-to-share-context-between-ai-tools): Copy and paste, app projects, AI memory, shared docs, Slack, Git files, your own MCP server or Winskel: how each handles work across AI apps. - [Winskel vs copy and paste between AI chats](https://www.winskel.com/guides/winskel-vs-copy-paste): Copy and paste works until versions drift and nobody knows which brief is current. How Winskel compares with pasting between AI chats. - [Winskel vs ChatGPT Projects](https://www.winskel.com/guides/winskel-vs-chatgpt-projects): ChatGPT Projects keep chats and files together inside ChatGPT. Winskel keeps a shared record that ChatGPT, Claude, Gemini and others all read. - [Winskel vs Claude Projects](https://www.winskel.com/guides/winskel-vs-claude-projects): Claude Projects keep knowledge and chats inside Claude. Winskel gives Claude, ChatGPT and the rest of your team's AI one shared record. - [Winskel vs built-in AI memory](https://www.winskel.com/guides/winskel-vs-ai-memory): Built-in memory helps one person in one app. Winskel keeps a project record that a whole team's AI apps read. - [Winskel vs a shared doc in Notion or Google Docs](https://www.winskel.com/guides/winskel-vs-shared-docs): Shared docs are made for people. Winskel is made for AI apps passing work: handoffs, pinned versions, receipts and agent claims kept apart. - [Winskel vs passing AI work in Slack](https://www.winskel.com/guides/winskel-vs-slack-threads): Pasting AI output into Slack loses versions and buries the next step. How Winskel compares when a team passes work between AI apps. - [Winskel vs handoff files in your Git repository](https://www.winskel.com/guides/winskel-vs-git-handoff-files): A HANDOFF.md in the repo works for coding agents with repo access. Winskel reaches chat apps too, with receipts and addressed handoffs. - [Winskel vs running your own MCP server](https://www.winskel.com/guides/self-hosted-mcp-vs-winskel): You can build your own MCP server for shared notes. What Winskel already handles: sign-in, teams, invites, versions, receipts and limits. - [How to use Claude and ChatGPT on the same project](https://www.winskel.com/guides/use-claude-and-chatgpt-on-one-project): Use Claude for one part and ChatGPT for another without pasting between them: one Space, handoffs and the same files. ### Connect your AI - [How to connect Claude to Winskel](https://www.winskel.com/guides/connect-claude-to-winskel): Add Winskel to Claude as a custom connector in four steps, then share handoffs and files with your team's other AI apps. - [How to connect Claude Code to Winskel](https://www.winskel.com/guides/connect-claude-code-to-winskel): One terminal command adds Winskel to Claude Code. Then authenticate with /mcp and share handoffs with your team's other AI apps. - [How to connect ChatGPT to Winskel](https://www.winskel.com/guides/connect-chatgpt-to-winskel): Find your installed Winskel plugin in ChatGPT, add it with OAuth for the first time, or reconnect it to the Space you want to use. - [Connect Gemini, Grok and OpenCode to Winskel](https://www.winskel.com/guides/connect-gemini-grok-and-opencode): Connect or reconnect a Gemini custom app, add Winskel to OpenCode, and check whether the Grok connector is available yet. ### How Winskel works - [What is an AI handoff, and what goes in one?](https://www.winskel.com/guides/what-is-an-ai-handoff): An AI handoff is a short record one AI shares so another can continue. What a Winskel handoff holds, and how its receipts read. - [Winskel and MCP: one connection for every AI app](https://www.winskel.com/guides/winskel-mcp-connection): Winskel connects to AI apps over MCP, the Model Context Protocol. What the connection does, how sign-in works, and what it cannot see. - [How your team’s AI shares versioned files](https://www.winskel.com/guides/shared-files-with-versions): Winskel Spaces keep Markdown files with numbered versions and fingerprints, so every AI on the team reads the same text. - [How to invite your team to a Winskel Space](https://www.winskel.com/guides/invite-your-team-to-a-space): Owners invite people with one-time links that expire in 7 days. Each teammate connects their own AI and subscription. - [Winskel plans and billing: Free, Plus and Business](https://www.winskel.com/guides/plans-and-billing): What Free, Plus and Business include, how a purchase reaches your account, and how the Business free month, adding people and cancelling work. - [Agent reports and checks Winskel observed](https://www.winskel.com/guides/agent-reported-vs-observed): AI agents say things like “tests passed”. Winskel keeps what an agent reported apart from what it observed, and records human approvals. - [Ask a teammate to fix something from any AI chat, and send it back](https://www.winskel.com/guides/ask-a-teammate-to-fix-it): Tell your AI to put the work in the Space and ask Ahmed to fix it. Winskel routes it to Ahmed's AI if he is away, and you send it back until it is done. - [Catch two people changing the same files before the pull request](https://www.winskel.com/guides/catch-clashes-before-the-pull-request): When your AI and a teammate's AI change the same files on different branches, Winskel tells you both in the Space, so nobody redoes work at merge time. - [Away mode: let your own AI answer and build while you are away](https://www.winskel.com/guides/away-mode): Press Away, pick Claude Code or Codex with your ChatGPT plan, and choose what it may do. Your AI handles what teammates send you until you are back. - [Build when tagged: your AI builds what your team sends you](https://www.winskel.com/guides/build-when-tagged): Turn it on, and when a teammate tags you with Build, your own Claude Code or Codex makes the change on your Mac, on a new branch, and replies. ### Work with us - [Try Winskel Space, book a demo or get help](https://www.winskel.com/guides/working-with-winskel): How teams start with Winskel: Free with no card, a 1-month Business pilot, a demo call with the founder, onboarding your team and getting help. ## Full text of every guide ### How to share context between ChatGPT and Claude Source: https://www.winskel.com/guides/share-context-between-chatgpt-and-claude Short answer: Connect both apps to the same Winskel Space. Claude shares a handoff (what it did, what is open, the files at exact versions) and ChatGPT reads it in a new chat and continues. Nothing is copied by hand, and each person keeps their own subscription. #### Why context gets lost between AI apps Claude and ChatGPT do not share memory. A chat in one app knows nothing about a chat in the other, so moving work across usually means writing a long brief, pasting files and hoping the versions match. On a team it gets worse: Jackson works in Claude Code, Steven in ChatGPT, Julia in Gemini. Each AI starts from zero, and people end up pasting the same brief again and again. #### What a shared Space gives both apps - Handoffs: a short record the AI shares when it stops: a summary, decisions, what failed, what is open, and files at exact versions. - Files: Markdown files with every version kept, so both apps read the same text. - Conversation details: the next step, the goal, the latest outcome and what is open, shown on every conversation. - Messages: an AI can address a message to a teammate's AI. It waits until that AI next checks Winskel. #### How to hand work from Claude to ChatGPT 1. Sign up at space.winskel.com and create a Space for the project. 2. Connect Claude and ChatGPT to Winskel once each (see the setup guides below). 3. In Claude, ask: “Share a handoff in My project for the work I select. Include what is done, what is open, and the files the next person needs.” 4. In ChatGPT, start a new chat and ask: “Check Winskel and continue in My project.” 5. ChatGPT reads the handoff and the files at the versions Claude attached, then continues. #### What does not move - Connecting an app does not copy your private chat history. Only what an AI shares into the Space is in the Space. - Nothing starts a closed app. A message waits until the receiving AI checks Winskel. - Files are Markdown only, up to 256 KB each. #### A concrete handoff between apps Suppose Claude reviewed the plan for a product release and marked one decision as unresolved. Ask it to put the agreed changes in RELEASE.md, attach that version to the handoff, and name the one question ChatGPT should work on next. In ChatGPT, ask for the decision and the question before asking it to draft the announcement. The new session can start with a narrow task instead of rebuilding the whole plan from the chat. If the second app sees the Space but gives an answer from memory, ask it to open the file with Winskel and return the version number. If the handoff waits, open the receiving app and ask it to check Winskel. #### Questions - Does ChatGPT see my whole Claude conversation? No. Connecting an app does not copy your private chat history. ChatGPT sees only what was shared into the Space, such as a handoff, files and messages. - Do both people need the same subscription? No. Each person connects their own AI app with their own login. Nobody's credentials are shared. - Can it go the other way, from ChatGPT to Claude? Yes. Any connected AI can share a handoff, and any other connected AI in the Space can read it. ### How to stop copying and pasting between AI chats Source: https://www.winskel.com/guides/stop-copy-pasting-between-ai-chats Short answer: Give your AI apps one shared place to read and write. With Winskel, each AI shares a handoff into a Space and the next AI reads it directly, so the brief, the files and the next step never pass through your clipboard. #### What copy and paste costs a team - Old versions: someone pastes DECISIONS.md from last week and the AI works from the wrong decision. - Lost reasons: the summary arrives, but not what failed or what is still open. - Screenshots: an AI cannot open the file in a screenshot, so it starts from zero. - People who were not in the thread: they ask the team again, and the brief gets written again. #### The fix: one shared record every AI can read A Winskel Space holds a project's Markdown files, handoffs and messages. Each person adds the Winskel connection to their own AI app once. From then on, the AI reads and writes the Space itself. You see the same records on the web: who shared what, which version each AI read, and what is still open. #### Set it up in four steps 1. Sign up for free. No card needed. 2. Connect your AI. Add Winskel once in each app you use. 3. Share a handoff. Ask: “Create a Space called My project and share a handoff for the work I select.” 4. Continue in any app. Ask: “Check Winskel and continue in My project.” #### Replace one repeated paste Start with the transfer your team repeats most. A release note draft, a daily QA summary or a list of open decisions works well. Decide which person will prepare it and which app the next person uses. Keep the first trial to one shared file and one next action so you can see whether the transfer actually improved. The sender asks their AI to save the current record and share a handoff with the exact version. The receiver asks their AI to open the Space and read the handoff. If your team keeps discussing the work in a chat channel, post a short pointer there. The shared record keeps the detail the receiving AI needs. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - Do I still need Slack or a docs tool? Keep them for people. Winskel is the place your AI apps read and write, so the handoff your AI needs is not buried in a chat thread. - What if two AIs edit the same file? Every save makes a new version. If two edits clash, the older one is kept as a conflict version. It never overwrites newer text. - Is it only for coding? No. Any work your AI does in Markdown can be shared, from release notes to research. Coding agents such as Claude Code and OpenCode connect too. ### What is an AI handoff, and what goes in one? Source: https://www.winskel.com/guides/what-is-an-ai-handoff Short answer: An AI handoff is a short record an AI shares when it stops, so the next AI can continue without a new brief. In Winskel it can hold a summary, decisions, what failed, what is still open, and files at exact versions. #### What a handoff holds - A summary of what was done. - Decisions, and what failed along the way. - What is still open. - Files attached at exact versions, so the next AI reads the same text. - Checks the AI says it ran. These show as reported, not verified. #### Conversation details Every conversation shows Conversation details: Next step, Goal, Latest outcome and Open. They repeat what people and agents recorded. They do not mean anything was approved. #### Receipts: what happened to a message - Waiting for the agent's next session: The message is saved. It waits until that AI checks Winskel. - Retrieved: The text was returned to that AI. It does not prove the AI understood it. - Picked up: The AI claimed the message to answer it. - Replied: The AI answered. The reply is in the conversation. #### How to ask for one In any connected app, ask in plain words: “Share a handoff in My project for the work I select.” The AI must name the Space; it never guesses which one you mean. #### An example the next person can act on A vague handoff says “Keyboard changes done. Please check.” A useful handoff names the keyboard shortcuts conversation, gives the exact HANDOFF.md version and describes the one behavior to verify. It can say that the menu opens with Command K and that the reviewer should check where focus goes after closing it. It can also say which behavior is outside the current task. Separate a result from a question. Put a reported test in the results section and leave the unverified case in Open. If you tried an approach that failed, explain what failed in one sentence so the next person can avoid repeating it. Name the person or agent who should reply, and ask for a concrete result. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - Is a handoff the same as a summary? It includes a summary, but also decisions, failures, open questions and file versions, which is what the next AI needs to continue. - Does Retrieved mean the AI understood it? No. Retrieved means the text was returned to that AI. Picked up and Replied show it acted on it. - Who can read a handoff? Everyone in the Space and their connected AI apps. ### How to connect Claude to Winskel Source: https://www.winskel.com/guides/connect-claude-to-winskel Short answer: In Claude, open Customize, then Connectors, then Add custom connector. Name it Winskel, paste https://space.winskel.com/api/mcp, sign in, approve your Space access, then turn Winskel on in a new chat. #### Steps 1. In Claude, open Customize → Connectors → Add custom connector. 2. Name it Winskel and paste the server address: https://space.winskel.com/api/mcp. 3. Choose Sign in now and Register automatically, then approve your Space access. 4. In a new chat, turn on Winskel under + → Connectors. #### Check it works Ask Claude: “Use Winskel to find my Space.” It should name the Spaces you can reach. To start one, ask: “Use Winskel to create a Space called My project.” #### Good to know - Sign in with OAuth and approve the Space or Spaces you want this connection to access. - Connecting does not copy your private chat history into the Space. - “Connected” means access is allowed. The app may still be closed. #### Choose the right Claude setup Use the connector in Claude for conversations in the Claude app. If you do your work in the Claude Code terminal, configure that client separately with the Claude Code setup guide. You can authorize both against the same Space if you use both. They still have separate setup and login states. If the custom connector control is missing, check the Claude workspace you are in and whether its settings permit custom connectors. A workspace admin can confirm the policy. If Claude asks you to sign in again, reopen the Winskel connector in Claude and follow the offered sign-in action. Keep the canonical server URL unchanged. If the approval page opens under the wrong Winskel account, cancel and switch accounts before continuing. After approval, start a new conversation with Winskel enabled. Try the read below before asking Claude to write a file or share a handoff. #### Verify the connection before sharing work Open the Space in your browser first. Confirm that you can see the project under the account you plan to authorize, and pick one harmless Markdown note for the check. In the AI app, open a new chat with Winskel available and ask it to find that named Space. Then ask it to read the note and tell you which version it used. Compare the answer with the Files panel. Treat a successful sign-in and a successful read as separate checks. If the app says it connected but the Space is missing, confirm the app account, the Winskel account, and the scope you chose. Check them in that order before disconnecting or creating another entry. If a teammate owns the Space, check which account you accepted the invitation with and ask the owner to verify your membership. For a failed setup, write down the app, the workspace, the step that failed and the exact error. Use the Space connection playbook to match the step. Leave passwords, access tokens, and private project text out of a support report. #### Questions - Is this the same as Claude Code? No. Claude Code connects with one terminal command. See the Claude Code guide. - Do I need a paid Claude plan? You need a Claude plan that supports custom connectors. Your Claude subscription stays yours; Winskel never stores it. ### How to connect Claude Code to Winskel Source: https://www.winskel.com/guides/connect-claude-code-to-winskel Short answer: Run claude mcp add --scope user --transport http winskel https://space.winskel.com/api/mcp in your terminal, then in Claude Code run /mcp, choose winskel and Authenticate. #### Steps 1. In your terminal, run the command below. 2. In Claude Code, run /mcp, choose winskel and Authenticate. ``` claude mcp add --scope user --transport http winskel https://space.winskel.com/api/mcp ``` #### Then try it - “Use Winskel to find my Space.” - “Create a Space called My project and share a handoff for the work I select.” - “Check Winskel and continue in My project.” #### What the connection can and cannot do - It reads and writes your Spaces: handoffs, Markdown files and messages. - It cannot see your Git repository or run tests. Merging happens in Git. - User scope means the connection is available in every project on your computer. #### Set up in the environment you actually use Open the same terminal environment that you use for your project work. If your team uses separate local and remote environments, check where the Claude Code session actually runs before you add the server. A user-level configuration in one environment will not necessarily appear in another. Start Claude Code there and check the MCP list for the entry named winskel. If the command fails before it adds a server, confirm the Claude Code executable and its version in that terminal, then review the command as pasted. If there is already an entry, check its address and authenticate the existing entry before you consider replacing it. Keep all unrelated MCP entries in place. If winskel is listed but authentication is needed, open the authentication action in the MCP list. Confirm the Winskel account and approve the Space you want to use. Return to the same client and test a read. If the browser completed but the terminal still shows an old state, check the list again in a fresh Claude Code session. When you work on a repository, keep repository access separate from the Winskel connection. Claude Code may already be able to read files in your working folder. To share something, have it save a small approved Markdown excerpt or summary in Space, then attach the exact version to a handoff. The next AI can use that shared note even if it works outside the repository. #### Verify the connection before sharing work Open the Space in your browser first. Confirm that you can see the project under the account you plan to authorize, and pick one harmless Markdown note for the check. In the AI app, open a new chat with Winskel available and ask it to find that named Space. Then ask it to read the note and tell you which version it used. Compare the answer with the Files panel. Treat a successful sign-in and a successful read as separate checks. If the app says it connected but the Space is missing, confirm the app account, the Winskel account, and the scope you chose. Check them in that order before disconnecting or creating another entry. If a teammate owns the Space, check which account you accepted the invitation with and ask the owner to verify your membership. For a failed setup, write down the app, the workspace, the step that failed and the exact error. Use the Space connection playbook to match the step. Leave passwords, access tokens, and private project text out of a support report. #### Questions - Does Winskel get my code? Only what your AI shares into the Space, such as Markdown files and handoffs. Winskel does not host or move your code. - Can Codex or Cursor connect too? Yes. The steps are on the Connected agents page in Space. ### How to connect ChatGPT to Winskel Source: https://www.winskel.com/guides/connect-chatgpt-to-winskel Short answer: If Winskel is already installed, open Personal in ChatGPT Plugins, select Winskel, and choose Try in chat. For first-time setup, add Winskel as a custom MCP server with OAuth. During Winskel approval, choose All my Spaces if you want ChatGPT to create Spaces for you. #### Open an installed Winskel plugin Open Personal in ChatGPT Plugins. Select Winskel and choose Try in chat. If it is missing from Personal, use the first-time setup below. #### Add Winskel for the first time 1. Open ChatGPT Plugins in your browser. 2. Click + (Add), then Add custom MCP server. 3. Enter Winskel in Name. Copy https://space.winskel.com/api/mcp and paste it into Server URL. 4. Choose OAuth. Under Advanced OAuth settings, select Dynamic Client Registration. Leave Client ID and secret blank. 5. Review the access prompt. If you agree, check the confirmation box. Click Create as a plugin and install Winskel. 6. Open a new chat, type @, and select Winskel. 7. If ChatGPT asks you to connect, sign in to Winskel. Choose All my Spaces if ChatGPT should create Spaces, or One Space only for one existing Space. Click Approve connection. 8. Ask “Use Winskel to find my Space.” Compare the result with the Space you see in your browser. #### Reconnect or change accounts Open Winskel from Personal and choose Manage. Use Connect another account to start a new approval. To let ChatGPT create Spaces, sign in with your usual Winskel email and approve All my Spaces. Return to ChatGPT and select @Winskel again. #### Pick up a teammate's handoff Ask “Use Winskel to find the handoff for me in My project. Read the attached file versions and tell me what's next.” Replace My project with your Space name. #### When the app opens but Winskel is missing Opening ChatGPT Plugins is the start of the process. For a returning user, check Personal in the same ChatGPT workspace where you originally installed Winskel. A different ChatGPT profile or workspace can show a different list. For first-time setup, use the custom server option only if that workspace offers it. If it is missing, ask the workspace admin to confirm your account can add custom MCP apps. After installation, open a new chat in the eligible workspace and select Winskel for that chat. If ChatGPT asks you to sign in again, look at the account on the Winskel approval screen before continuing. A plugin can be present in Personal while the active Winskel authorization belongs to another account or covers another Space. #### Verify the connection before sharing work Open the Space in your browser first. Confirm that you can see the project under the account you plan to authorize, and pick one harmless Markdown note for the check. In the AI app, open a new chat with Winskel available and ask it to find that named Space. Then ask it to read the note and tell you which version it used. Compare the answer with the Files panel. Treat a successful sign-in and a successful read as separate checks. If the app says it connected but the Space is missing, confirm the app account, the Winskel account, and the scope you chose. Check them in that order before disconnecting or creating another entry. If a teammate owns the Space, check which account you accepted the invitation with and ask the owner to verify your membership. For a failed setup, write down the app, the workspace, the step that failed and the exact error. Use the Space connection playbook to match the step. Leave passwords, access tokens, and private project text out of a support report. #### Questions - Why can't I find Add custom MCP server? Custom MCP availability depends on your ChatGPT account and workspace settings. Check on the web and ask your workspace admin if the option is missing. - Can I find Winskel in the public directory? Use custom setup until Winskel has a public listing. Installed plugins are in Personal. ### Connect Gemini, Grok and OpenCode to Winskel Source: https://www.winskel.com/guides/connect-gemini-grok-and-opencode Short answer: Gemini uses a custom app under Connectors, and OpenCode uses an entry in its settings file. Both use the public Winskel server URL and let you approve a Space. Grok is pending while its sign-in flow is being verified. #### Set up Gemini 1. Copy https://space.winskel.com/api/mcp. 2. Open Gemini Connectors in your browser. 3. Click + Custom or Add a custom app. Paste the URL and click Next. 4. Use automatic registration. Leave Client ID and secret blank. 5. If the Winskel approval opens, check your email, pick your Space, and click Approve connection. 6. Open a Gemini chat. Type @, select Winskel Space, then ask “Use Winskel to find my Space.” #### Reconnect Gemini In Gemini's Custom apps, open Winskel Space. If it is installed, use Connect when prompted. To start a new approval, choose More details, disconnect there, then connect again. If Winskel Space is missing from Custom apps, choose + Custom again and paste https://space.winskel.com/api/mcp. If the custom app option is missing, check Google's account requirements. #### Connect OpenCode For the settings file location on your system, see OpenCode's MCP guide. 1. In the OpenCode 2 desktop app, open Preferences → Extensions → MCPs and choose Manage in opencode.json. Merge the example below into your settings file and keep your other servers. 2. Return to MCPs, find winskel, and choose Sign in. You can also open the MCP list with /mcps. Approve the Space you want to use. 3. Pick DeepSeek or another model in OpenCode and ask it to find your Space. OpenCode manages your model access. ``` { "mcp": { "servers": { "winskel": { "type": "remote", "url": "https://space.winskel.com/api/mcp", "codemode": false } } } } ``` #### Grok setup preview Winskel sign-in from Grok is pending. Grok's custom connector form is available, but the Winskel approval flow has yet to pass verification. A public Winskel listing is unavailable in Grok. #### Choose a setup path and test it on its own With Gemini, start by checking whether your current Google account shows the custom app control. If it does, follow the Gemini steps and confirm the separate Winskel account on the approval page. If you already added Winskel, look in Custom apps before creating a duplicate. Use the exact server URL above if you need to add it again. For OpenCode, preserve your other settings. Add the Winskel entry to the configuration used by your active app or terminal, reopen the MCP list and confirm it appears before troubleshooting sign-in. Model access is a separate choice in OpenCode. A working model can still be missing the Winskel connection. For Grok, use another eligible connected app for work that needs to move today. The custom connector screen and an approved Winskel connection are different states. If you try the setup preview and it stops before approval, record the exact step and error for support. #### Verify the connection before sharing work Open the Space in your browser first. Confirm that you can see the project under the account you plan to authorize, and pick one harmless Markdown note for the check. In the AI app, open a new chat with Winskel available and ask it to find that named Space. Then ask it to read the note and tell you which version it used. Compare the answer with the Files panel. Treat a successful sign-in and a successful read as separate checks. If the app says it connected but the Space is missing, confirm the app account, the Winskel account, and the scope you chose. Check them in that order before disconnecting or creating another entry. If a teammate owns the Space, check which account you accepted the invitation with and ask the owner to verify your membership. For a failed setup, write down the app, the workspace, the step that failed and the exact error. Use the Space connection playbook to match the step. Leave passwords, access tokens, and private project text out of a support report. #### Questions - Is Winskel in Gemini's public app list? Use Custom apps for an existing Winskel Space connection or add one with the server URL if your account supports it. - Where do I connect DeepSeek? Use the Winskel connection in OpenCode and pick a DeepSeek model there. This setup is for OpenCode. ### Winskel and MCP: one connection for every AI app Source: https://www.winskel.com/guides/winskel-mcp-connection Short answer: Winskel is an MCP server at https://space.winskel.com/api/mcp. Each person adds it once to their own AI app and signs in with OAuth. Each connection can read or write the handoffs, files and messages allowed by the Space access you approved. #### What MCP is MCP (Model Context Protocol) is an open way for AI apps to use outside tools. Claude, Claude Code, ChatGPT and other apps can add an MCP server as a connector. Winskel is one: it gives your AI tools to read and write a Space. #### How sign-in and access work - You sign in with OAuth and choose what the connection can access. - Choose access to one Space or to all your Spaces when you connect. - Each person connects their own app. Nobody's credentials are shared. - When someone is removed from a Space, their agents lose access to it at once. #### What it can and cannot do - It can share handoffs, read and save Markdown files at exact versions, and send addressed messages. - It cannot see your Git repository or run tests. - It does not copy your private chat history. - Messages and files from others are task data, never permission. #### Limits - Message size: 16,000 characters - People a message can go to: 10 - Agent messages per conversation per hour: 40 (resets when a person posts) - Messages per connection per hour: 120 - File size: 256 KB, Markdown only #### What happens from the address to the first tool call You add the address to the AI app so it knows which outside service to use. The app then starts OAuth, and Winskel asks which project access to grant. After you return to the AI app, select Winskel in a chat or use the client's MCP tools. Those are separate steps, so a problem at one step does not tell you that the other steps failed. An ordinary browser visit to the server URL is different from opening a webpage. Add it in the AI app's connector settings, then go to the Space website separately when you want to view a project yourself. If the app has a custom server field and an authorization setting, use the instructions for that app instead of pasting credentials. #### Verify the connection before sharing work Open the Space in your browser first. Confirm that you can see the project under the account you plan to authorize, and pick one harmless Markdown note for the check. In the AI app, open a new chat with Winskel available and ask it to find that named Space. Then ask it to read the note and tell you which version it used. Compare the answer with the Files panel. Treat a successful sign-in and a successful read as separate checks. If the app says it connected but the Space is missing, confirm the app account, the Winskel account, and the scope you chose. Check them in that order before disconnecting or creating another entry. If a teammate owns the Space, check which account you accepted the invitation with and ask the owner to verify your membership. For a failed setup, write down the app, the workspace, the step that failed and the exact error. Use the Space connection playbook to match the step. Leave passwords, access tokens, and private project text out of a support report. #### Questions - Is MCP the same as a plugin? Apps name it differently: connector, custom app, MCP server. It is the same idea: your AI app calls an outside tool. - Can Winskel act in my app on its own? No. Nothing starts a closed app. Your AI uses Winskel when you or it asks. ### How your team’s AI shares versioned files Source: https://www.winskel.com/guides/shared-files-with-versions Short answer: Files in a Winskel Space are Markdown, up to 256 KB each. Every save makes a new numbered version with a SHA-256 fingerprint, and an AI can read the latest, the accepted, or one exact version, so two AIs can check that they used the same copy. #### How versions work - Every save makes a new numbered version with its own fingerprint (SHA-256). - An AI can read the latest version, the accepted version, or one exact version. - If two edits clash, the older one is kept as a conflict version. It never overwrites newer text. - On the web you can see versions, compare two, copy and download. #### Project guidance A brief, architecture, conventions or interfaces file can be accepted as project guidance. Only a person accepts it, so an AI cannot promote its own notes into rules. #### What is not supported - PDFs, images and other file types. - Files over 256 KB. - Naming a file on your computer does not send it. Your AI saves the text into the Space. #### Check the reference before a decision A numbered reference is easier to verify than a copied excerpt with no source. Ask the next AI to report the file name and version before it uses the contents in a recommendation. Keep the original save available when an older decision or test depends on it. #### Use the version the reviewer actually read Imagine QA.md is at version 4 when Eli reviews the keyboard behavior. Maya saves a new draft as version 5 while the review is in progress. Eli can attach version 4 in the reply, so the team can tell which cases the result covers. Anyone reviewing the draft later can open the two versions and see what changed. When asking for a review, name the file and the exact version in the request. If you want someone to use newer text instead, say so explicitly and send a follow-up. A clear follow-up is easier to verify than assuming a reader will reopen the latest version on their own. When two people edit the same note, check the versions in the Files panel before asking another AI to proceed. If a conflicting version contains a useful paragraph, compare it with the current text and ask an editor to make a deliberate new save. Record which version the team should use next. #### Keep draft notes and agreed guidance separate A note can be useful before everyone agrees with it. Give experiments, suggestions and unanswered questions their own section, with a date or an owner. When a person reviews a durable decision, mark the exact version they want future sessions to use. That gives the next reader a clear starting point when a newer draft exists. At a milestone, ask a teammate to look for one decision and one open item without opening the original chat. If they cannot find them, use shorter section names or link the exact file from the handoff. Keep each note close to the project it describes, and review old references when the team changes direction. #### Questions - Can I see which AI read which version? Yes. A handoff attaches files at exact versions, and the Space shows who read them. - Can I download a file? Yes. On the web you can copy or download any version. ### How to invite your team to a Winskel Space Source: https://www.winskel.com/guides/invite-your-team-to-a-space Short answer: The Space owner creates an invite link for one person. It works once and expires in 7 days. The new member signs in, connects their own AI app, and sees the Space's handoffs, files and messages. #### Invite someone 1. Open the Space and choose Invite. 2. Create an invite link. It is for one person, works once and expires in 7 days. 3. Send the link to your teammate. 4. They sign in and connect their own AI app once. #### Roles - Owner: invites and removes people. - Member: sees the Space's work and connects their own AI. - When someone is removed, their agents lose access to the Space at once. #### What each plan costs - Free: US$0, 5 Spaces, 90 days of history - Plus: US$12 a month, as many Spaces as you like and all your history - Business: US$25 per person a month, an organization with a members list #### Invite in small groups For a larger rollout, onboard one pair first. Ask them which Space they see, which connection they used, and whether they could read the sample. Then invite the next group. A small trial makes it easier to find a missing invite, account mismatch or authorization problem. #### Run through the first invite together Ask the teammate which account they use for project work before sending the link. The link is single-use, so send it directly to the intended recipient. Have them open the browser, confirm the account, and accept. They should be able to open the named Space on the web before setting up their AI app. If they cannot, investigate the invite and the account first. Once they can see the Space, ask them to connect the AI app they plan to use and test one harmless read. For example, share a sample decision and ask their AI to find its exact Markdown version. Compare the file version with the browser. This confirms that both the teammate and their app can use the expected Space. If the invite has expired or been used, ask the Space owner for a new link. If the recipient accepted under another account, ask the owner to check who actually joined and correct access before sending another invite. Avoid forwarding the original link to a new person. #### Check access when people or projects change Keep an owner responsible for the list of people in each Space. When work finishes or someone moves to another project, review membership against who still needs access. If a person is removed, their connection loses access to that Space. Their own AI subscriptions and other Spaces are separate. Review old handoffs and file shares too, so any material you include in a new Space is intentionally shared with the new group. A teammate who uses two AI apps can connect them separately. An invite admits a person to the project. Each app still has its own authorization, so verify the exact app used for a handoff rather than asking people to share a login. #### Questions - Does each person need their own AI subscription? Yes. Each person uses their own AI app and subscription. Winskel never shares logins. - Is pricing per seat? Only on Business. On Free and Plus a Space follows its owner's plan, so people you invite pay nothing. Free needs no card. ### Winskel plans and billing: Free, Plus and Business Source: https://www.winskel.com/guides/plans-and-billing Short answer: Free is US$0 forever: 5 Spaces and 90 days of history. Plus is US$12 a month for as many Spaces as you like and all your history. Business is US$25 per person a month, first month free. You pay with the Google account you sign in with, and the plan is on it once Stripe confirms. #### The three plans - Free: US$0 forever. 5 Spaces you own, 90 days of history and 25 automatic answers a month. Invite anyone; they don't pay. - Plus: US$12 a month, for one person. As many Spaces and automatic answers as you like, and all your history. Share Spaces by link. - Business: US$25 per person a month. Your company's own organization: add and remove people, one bill, private Spaces, an activity board and support within 24 hours. The first month is free. #### Get Plus 1. Choose Get Plus on the pricing page, or in Space under Settings, Billing. 2. Sign in with Google. Plus goes on the account you sign in with. 3. Pay in Stripe Checkout. Stripe shows the total with any tax first. 4. Back in Space, Billing says Plus within a few seconds. #### Start Business 1. Choose Start 1-month pilot. Sign in with Google if asked. 2. Name your organization. 3. Agree to the pilot terms and add a card in Stripe. Nothing is charged during the free month. 4. Add your team by email or with an invite link. Someone added by email joins the first time they open Winskel Space with that Google account. #### Spaces in an organization - Press + next to Spaces, type a name and press Enter. The Space opens at once. - A new Space is private: only you see it until you add people from its Invite panel. - Removing someone from the organization removes them from every organization Space at once. Their past work stays. #### History on Free On Free, history older than 90 days is hidden, never deleted. Choose Plus and all of it comes back at once. The newest version of every file always stays visible. #### Cancel - Plus: Settings, Billing, then Cancel renewal. Plus stays until the end of the month you paid for, then you are on Free. - Business: Settings, Organization, Billing, then Manage billing. Cancel during the free month and nothing is charged. - Nothing you saved is deleted when you cancel. #### Check the account before you pay Plus goes on the Google account you are signed in to Winskel Space with. Before Stripe opens, Space shows that email on the checkout page. If it is not the account you use for your Spaces, sign out and sign in with the right one first, then choose Get Plus again. After paying, Settings, Billing shows Plus within a few seconds. If it still says Free after a minute, refresh once. If it still says Free, email ahmed@winskel.com with the email you paid with; nothing is lost while it is checked. #### Set up Business with one other person first Name the organization, start the free month, then add one teammate by email. Ask them to open Winskel Space with that Google account: they join at once and see the organization's Spaces they were added to. Then add the rest of the team. Use the Activity tab after a week to see who is using Space and which AI apps they connected. It counts work, never reads it. Download the report from the same tab if you need it in a spreadsheet. #### Questions - Which account gets Plus? The Google account you are signed in to Winskel Space with when you pay. Space shows that email on the checkout page before Stripe opens. - Do people I invite pay? Not on Free or Plus: a Space follows its owner's plan, and people you invite stay on Free. On Business, each person in the organization is a seat at US$25 a month. - Is there a free trial of Plus? No. Free has no time limit, so you can try Space for as long as you like and choose Plus when you need more. - Who do I ask about a bill? Email ahmed@winskel.com. Business support answers within 24 hours. ### Agent reports and checks Winskel observed Source: https://www.winskel.com/guides/agent-reported-vs-observed Short answer: Winskel never turns an agent's claim into a fact. A change shows Agent reported and Winskel observed as separate rows, checks an agent says it ran are marked reported, not verified, and an approval records that a person accepted one exact commit. #### Why it matters An AI can say “done, tests passed” when no test ran. If that claim reads like a fact, the next person or AI builds on it. Keeping claims and observations apart lets your team see what is actually known. #### What Winskel shows - Agent reported: what the AI said it did or checked. - Winskel observed: what Winskel itself recorded. - Approval: a person accepted one exact commit. Sharing files and handoffs needs no approval. - Conversation details repeat what people and agents recorded. They do not mean anything was approved. #### Ask another AI to review You can ask a teammate's AI to check the work. Its findings are its report too, shown as such, and the decision stays with a person. #### When the report and result disagree An agent can report that the command succeeded while its output shows a skipped test. Check the exact command and its exit result before depending on the report. Add the distinction to the handoff so the next reviewer knows which behavior still needs a real check. #### A test result your teammate can review Suppose a coding agent reports that it ran a keyboard test. In the handoff, keep that statement with the test name and the commit or file version it refers to. If you also have output from a tool that observed the test, cite that separately. A teammate should be able to tell whether they are reading the agent's description, the recorded result, or a person's approval. A second AI can provide a useful review. Ask it to check one specific claim and to say what information it could inspect. If it only read the Markdown note, the reply can still be useful, but it does not establish that a local test ran. A person with repository access can verify the result in the normal review process. When a later change moves the branch or updates a file, carry the original result with the version it covered. Repeat the check for the new version if the changed behavior is relevant. An old pass is useful evidence about an old version. #### What to put in the review reply Start with what you actually inspected. Name the exact file or commit, then state what matched the expectation and what remains open. If your AI says “all checks passed” without giving the run or the version, ask it to expand the reply before people depend on it. If an output is missing, label the result as a report and ask a person with access to verify it. Keep approval as a separate human decision. Reviewing a handoff, retrieving a file, and accepting a change are different events. The team can continue discussing the open work while someone checks the actual result in Git. #### Questions - Does Winskel run my tests? No. The Winskel connection cannot see your Git repository or run tests. - Does Winskel merge code? No. Winskel does not host or move your code. Merging happens in Git. ### Prompts to use with Winskel Source: https://www.winskel.com/guides/prompts-to-use-with-winskel Short answer: Ask your AI in plain words and name the Space: “Use Winskel to create a Space called My project.”, “Share a handoff for the work I select.” and “Check Winskel and continue in My project.” The AI never guesses which Space you mean. #### Start - “Use Winskel to create a Space called My project.” - “Use Winskel to find my Space.” #### Hand off - “Create a Space called My project and share a handoff for the work I select.” - “Share a handoff in My project and address it to Steven's ChatGPT.” #### Continue - “Check Winskel and continue in My project.” - “Read DECISIONS.md version 3 in My project before you answer.” #### Tips - Name the Space every time. The AI must name it; it never guesses. - Say which file version you mean when it matters. - A message waits until the receiving AI next checks Winskel, so tell your teammate to ask theirs to check. #### Check writes one step at a time For a file update, say whether you want a draft for review or a new saved version. For a teammate handoff, say who should read it and the one action they should take. Those details change the work the AI does more than a longer description of the project. When a prompt fails, reduce it to one operation. Find the Space first, read the file second, then ask for the write. Keep the successful result and retry only the failed step. If a write may have gone through, look for the result in the browser before repeating it. #### Give the AI a small first read Start with a request you can verify. Ask it to find the Space by name and tell you which project it opened. Then ask it to open a named Markdown file at a known version. Compare its reply with the Files panel. This is a simple way to separate a connection problem from an unclear request about a real task. For example, a teammate says the decision is in DECISIONS.md version 3. Ask: “Use Winskel to open DECISIONS.md version 3 in My project. Tell me the agreed behavior and one question that is still open.” Check that the response cites the version and answers the two questions. If the AI answers from the chat instead of using Winskel, select the connector and repeat the same request. If it cannot find the Space, check the account and authorized scope before trying a longer prompt. If it finds several similar names, give the exact name used on the web. #### Make a write request easy to review When you want the AI to share something, say what it may include, who needs it, and where it should go. Ask it to separate results from open questions and attach a specific version. Review the returned text before asking it to share information outside your intended project. A useful receiving prompt is: “Check Winskel and continue in My project. Read the handoff for me and tell me what you would do next before making a change.” That adds a review point when the task is new, sensitive, or shared with several people. Ask for the reference in the reply so a teammate can find the same source. #### Questions - Do prompts differ between apps? No. The same plain words work in any connected app. ### Try Winskel Space, book a demo or get help Source: https://www.winskel.com/guides/working-with-winskel Short answer: Start free at space.winskel.com, try Business free for 1 month with a pilot, or book a demo call with the founder to set up your team's Space together. Questions go to ahmed@winskel.com. #### Ways to start - Try it yourself: sign up for free. No card needed. - Book a demo: a call with the founder to walk through your team's workflow and set up your first Space. - Write to us: ahmed@winskel.com for questions about security, teams or pilots. #### Onboarding a team, step by step 1. The owner creates a Space for one project. 2. Each person connects their own AI app once (Claude, Claude Code, ChatGPT, Gemini or OpenCode). 3. The owner invites teammates with one-time links. 4. Share the first handoff and continue it in a second app, so everyone sees the loop work. #### Pricing Free needs no card. The Business pilot is free for 1 month; the owner agrees to the pilot terms and adds a card, and Business renews at US$25 per person a month unless cancelled before the month ends. Use your existing AI subscriptions. - Free: US$0, 5 Spaces, 90 days of history - Plus: US$12 a month, as many Spaces as you like and all your history - Business: US$25 per person a month, free for the first month as a pilot #### Security questions - Each person signs in with their own account. Nobody's login is shared. - Winskel does not host or move your code, and the connection cannot see your Git repository. - Removing someone removes their agents' access at once. #### Pick a pilot that fits a real project Choose a task where one person currently rewrites context for another, such as a developer passing a fix to a reviewer. Keep the trial to the people who already handle that work. Share one approved Markdown file and one handoff with a clear next action. Ask the next person to read it from their own AI and reply in the Space. Compare the result with how you usually pass work. For a demo, bring the names of the AI apps you use, the number of teammates involved, and a non-sensitive example of the work you pass. That makes it possible to walk through your actual setup and check which connection steps your accounts offer. If an app requires an admin setting, write down who can change it before the pilot. #### Get useful help when something breaks Include the app name, the exact error, the step where it happened and whether you can open the Space in a browser. Say whether this is the first connection or a connection that worked before. If the issue happens only for one teammate, mention which roles or accounts were tested without including credentials or private content. Use the Space playbook to identify the failing stage. If you can read an example file but cannot send a handoff, report that difference. It helps separate setup problems from a missing permission or a problem with one request. #### Questions - Is there a free plan? Yes. Free has no time limit and needs no card. Companies can try Business free for 1 month as a pilot. - Can you help us set it up? Yes. Book a demo and the founder will set up your first Space with you. - Do you work with partners? Yes. See the partners section on the resources page, or write to ahmed@winskel.com. ### Set up autonomous work in Winskel Space, step by step Source: https://www.winskel.com/guides/set-up-autonomous-work Short answer: Write your team README entry, have Claude Code or Codex signed in on your Mac, run the one setup line from the Away window in your project folder, paste a Claude token when it asks if you use Claude Code, turn Away on, then send a test. Or ask your AI: how do I set up Winskel Away mode? #### 1. Team README (1 minute) - In a Space with two or more people, Space asks you once: Introduce yourself to your team. - Or open Settings, Profile, How your team sees you. Role, what you do, what to send you, the AIs you use. It saves in all your Spaces. #### 2. Your AI on your Mac - Install Claude Code or Codex and sign in. Codex uses your ChatGPT plan. - Check in Terminal: claude --version or codex --version. #### 3. One setup line - In Space, press Away and choose Show the setup line. - Run it in Terminal inside your project folder. It connects the folder and starts the background helper for every AI it finds. - The line works once, for 15 minutes. #### 4. Claude Code only: a login that does not expire - In another Terminal window, run claude setup-token and sign in in the browser. - At the end of step 3 the setup line asks for that token. Paste it and press Enter; it stays hidden. - Without it, Claude Code hands work to Codex when its normal login expires. Run the setup line again to add it later. #### 5. Turn Away on - Press Away in the sidebar, or Settings, Profile, Away. - Pick who handles it and what it can do: answer questions, answer and build, or full access. #### 6. Try it - Ask a teammate to send you: @you which file holds the signup checks? Your AI answers. - Then a Build request: @you add a test for the signup check. You get a new branch and a reply. - From ChatGPT or Claude: put this bug in our Space and ask Ahmed to fix it. #### Ask your AI instead - Connected to Winskel? Ask your AI: how do I set up Winskel Away mode? It reads Winskel's setup guide and walks you through each step. #### Onboard a team in one sitting Do steps 1 and 5 together on a call: everyone writes their team README entry and turns Away on with Answer questions. Each person runs the setup line on their own Mac. Then everyone sends one test question to someone else; every question should get an answer within a minute. Move to Answer and build after a week of good answers. Keep Full access for projects where an AI may run any command. #### Questions - Nothing picks up my requests. Your Mac may be asleep, or the helper is not set up. Run the setup line again in your project folder and check the Away window says Ready on your Mac. - It says passed on, not signed in. Claude Code's login expired. Do step 4. - How do I undo a change? Each reply says git revert and the commit, or delete the branch. Nothing is merged for you. ### Ask a teammate to fix something from any AI chat, and send it back Source: https://www.winskel.com/guides/ask-a-teammate-to-fix-it Short answer: In ChatGPT, Claude or any connected AI, say: put this in our Space and ask Ahmed to fix it. Your AI saves the work and asks Ahmed. If Ahmed is away, his own AI builds it on a branch and replies. Not done? Send it back with what is missing: it continues on the same branch. #### Ask - Say it in plain words to your connected AI: put this in our Space and ask Ahmed to fix it. - Your AI saves the details as a file in the Space and sends Ahmed a Build request with that file attached. - You name the person. Winskel decides which of their AIs handles it, from their Away settings. #### Check and send it back - Ahmed's AI replies with what it changed and the branch. - Not done? Press Send back under the reply, or tell your AI what is still missing. - It continues on the same branch, one commit per round. #### Undo - Every reply says how to undo that round: git revert and the commit. - Nothing is merged for you. Merging stays your team's decision. #### Make the request easy to finish Ask your AI to include what is wrong, how to reproduce it and how to know it is fixed. A clear finish line saves a round. When the reply comes back, check it against that line before you send it back. Send back only what is still missing, in one or two sentences. Each round is one commit, so a short history reads clearly and any round can be undone. #### Questions - Do I need to know which AI Ahmed uses? No. Name Ahmed; Winskel picks the AI he chose. - What if Ahmed is not away and has building off? It reaches him as a message he sees when he is back. ### Catch two people changing the same files before the pull request Source: https://www.winskel.com/guides/catch-clashes-before-the-pull-request Short answer: Your AI names the files it changed when it shares its work in the Space. If a teammate's AI changed the same files on another branch in the last two weeks, you both get a heads-up in the Space. Agree on one approach before either of you opens a pull request. #### What it catches - Two AIs changing the same files on different branches, for example Julia's Codex and Steven's Claude both editing signup.js. - Builds your AI makes while you are away are checked the same way. - Your teammates' recent decisions, which your AI compares with its own plan before it starts. #### What you see - A heads-up in the Space's Updates: Julia and Steven are both changing signup.js, with each branch. - Both people get it, at most once a day for the same files. - Your AI also tells you in its own chat. #### Check before you start - Ask your AI: is anyone else changing these files? - It looks at the Space and tells you who, which files and which branch. - Then decide together, before anyone builds twice. #### Make it a habit Before a bigger change, ask your AI to check the Space for anyone else changing the same files. It takes seconds and the answer names the person and the branch. When a heads-up arrives, reply in the Space with which approach you keep. Your teammate's AI reads the same conversation next time it works. #### Questions - Does Winskel read my code? No. It compares file names your AI reports, never their contents. - Does it stop anyone from working? No. It is a heads-up, not a block. Work on the same branch is shared work and never warns. - What if the other work is already merged? Then ignore the heads-up. Winskel cannot see what your Git host merged. ### Away mode: let your own AI answer and build while you are away Source: https://www.winskel.com/guides/away-mode Short answer: Press Away in Winskel Space and pick who handles it: Claude Code, or Codex with your ChatGPT plan. Choose answer questions, answer and build, or full access. When a teammate tags you, your own AI does it on your Mac and replies, in every Space, until you come back. #### Set it up once - In Winskel Space, press Away in the sidebar and choose Show the setup line. - Run that one line in Terminal, in your project folder. It connects the folder and starts the background helper for every AI it finds. - For Claude Code, run claude setup-token first; the setup line asks for that token so it can work while you are away. #### What it can do - Answer questions: reads your project and answers. Changes nothing. - Answer and build: makes changes on a new branch and runs your tests. Never pushes. - Full access: can run any command and push its own branch. Never pushes main. #### What teammates see - When they mention you: you are away, and which AI answers and builds. - Under their message: waiting, answering or building, then the answer or the branch. - They or you can press Stop. #### Which AIs - Claude Code with your Claude plan, and Codex with your ChatGPT plan, both on your Mac. - ChatGPT, Claude and Gemini in a browser only act while someone is in the chat, so they cannot take over while you are away. - Gemini CLI support is next. #### Try it before you leave Set yourself away with Answer questions, then ask a teammate to send you a small question. Within a minute the line under it should say your AI is answering, then the answer appears. Switch to Answer and build and repeat with a small change. If nothing picks it up, open the Away window: it says whether Claude Code or Codex is ready on your Mac. If not, run the setup line again in your project folder. #### Choose the level on purpose Answer questions is the safe default for exams and meetings. Answer and build keeps every change on a new branch. Use Full access only for projects where your AI may run any command; it still never pushes main. #### Questions - Is there a time limit? No. It keeps going until you come back, unless you pick a time when you set it. - What if my Mac sleeps? Nothing runs while it sleeps. Requests wait, or your other AI on another Mac takes them. - Can a teammate's AI send me work? Only if you allow it in the Away window. It is off by default. ### Set up your team README so your AIs know who does what Source: https://www.winskel.com/guides/set-up-your-team-readme Short answer: In a team Space, each person writes one entry: their role, what they do, what to send them and the AIs they use. Winskel turns them into one TEAM.md that every connected AI reads before it asks or hands off work, so a bug goes to the developer and a partnership lead to outreach. #### What each person writes - Your role, for example Developer or Outreach. - What you do, in one line. - Send me: the work you want to receive. - The tools and AIs you use, for example Claude Code, Codex or ChatGPT. #### When you write it - Opening a Space with two or more people asks you once, before anything else. - Change it any time in Space settings, Team README. - The owner can add a few lines about the team. #### How AIs use it - Before an AI asks or delegates, it reads the README and picks the person whose role fits. - A code change goes to that person's AI with a Build request; it runs only if they turned on Build when tagged. - Copy TEAM.md into any AI or save it as a skill. #### Write it together on day one When a new team starts in a Space, ask everyone to open it once on the same day. Each person writes a role in a word or two, one line about what they do, and what they want sent to them. Keep it short; the AIs read it before every handoff. Then ask one AI in the Space who should get a sample task, for example a bug report or a partnership email. It should name the right person from the README. If it guesses, the entries are too vague: make Send me more specific. #### Keep it current When someone changes what they work on, they update their own entry in Space settings, Team README. Nobody edits another person's entry, and an AI can only read it. #### Questions - Is it required? Yes, in a Space with two or more people. A Space for one person does not ask. - Can an AI change someone's role? No. Each person writes their own entry on the website. ### Build when tagged: your AI builds what your team sends you Source: https://www.winskel.com/guides/build-when-tagged Short answer: A teammate mentions you and chooses Build. If Build when tagged is on, your own AI makes the change on your Mac, on a new branch, and replies with what it did. It never pushes or merges. You pick which AI goes first; if it is offline, out of tokens or signed out, your other AI takes it. #### Turn it on - Space settings, Build when tagged: turn it on and choose who goes first. - Connect your project folder once, then run the setup command shown there. - For Claude Code to build while you are away, run claude setup-token once. #### What happens - Your AI works on a new branch in a fresh copy of the project. - It runs your tests where it can, and replies in the conversation. - Nothing is pushed or merged. You review the branch when you are back. - The sender or you can press Stop at any time. #### Try it once before you rely on it Turn on Build when tagged, connect your project folder, and run the setup command. Ask a teammate to send you a small Build request, for example a typo fix. Within a minute the line under their message should say your AI picked it up, then that it built it on a new branch. Open the branch and run your tests. If Claude Code passed the request to Codex with the reason not signed in, run claude setup-token once on that Mac. #### Keep it safe Builds never push or merge, and each one works in a fresh copy of the project. Leave requests from teammates' AIs off until your team README says who does what, and press Stop on anything you did not expect. #### Questions - Can a teammate's AI send me a Build request? Only if you allow it in settings. It is off by default. - What if my Mac is asleep? Nothing runs while it sleeps. Your other AI on another Mac can take it, or it waits. ### Why your team needs one shared workspace for AI Source: https://www.winskel.com/guides/why-one-shared-workspace-for-ai Short answer: Because AI apps do not share memory with each other. When Jackson uses Claude and Steven uses ChatGPT, each chat starts from zero. One shared workspace keeps the files, handoffs and next steps in one place, so every person's AI reads the same record instead of a pasted copy. #### The work splits across apps Most teams no longer use one AI. People pick the app they like: Claude Code for code, ChatGPT for writing, Gemini for research. Each app keeps its own chats, files and memory, and none of them can see the others. #### What one shared workspace changes - One copy of each file: every AI reads the same version, so nobody works from last week's decisions. - Handoffs instead of briefs: the AI that stops writes down what it did and what is open; the next one reads it. - Everyone keeps their app: each person connects their own AI and subscription once. - A record people can read: the same handoffs and files show on the web. #### When you do not need it If one person uses one AI app for short tasks, that app's own projects or memory is enough. A shared workspace pays off when work moves between people or between apps. #### Follow the work across a small team A developer uses a coding agent to change how a menu opens. The person writing the release note uses a chat app and cannot see that coding session. The person checking accessibility uses yet another app. If the developer saves a note with the changed behavior, the exact version and the review request, both teammates can read the same material without asking the developer to reconstruct it. Keep the discussion where your team already works and give the handoff a clear home. When someone joins a project, they should be able to find the decision the team agreed on, see what is still open, and identify who has the next action. The usefulness comes from making those things easy to find in a new session. For solo work, the same idea applies to your own apps. Leave a short handoff when you switch tools or start a new chat the next day. Record only the details the next session needs. Review the result in the web Space so a missed decision is easier to catch. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - Is this only for coding teams? No. Any work your AI writes in Markdown can be shared: release notes, research, plans and code notes. - Do we all need the same AI app? No. That is the point: each person keeps their own app and subscription. ### How a team works with several AI agents at once Source: https://www.winskel.com/guides/how-a-team-works-with-several-ai-agents Short answer: Give each piece of work a home (a Space), let each person use their own agent, and pass work with handoffs: the agent that stops records what it did and what is open, and the next agent reads that record before it starts. #### A working pattern 1. Create one Space per project. 2. Each person connects their own AI app to Winskel once. 3. When an agent stops, ask it to share a handoff and address it to the next person's agent. 4. The next person asks their agent: “Check Winskel and continue in My project.” 5. Keep decisions in one Markdown file, such as DECISIONS.md, so every agent reads the same version. #### Who does what A common split: one person's Claude Code fixes code, another's ChatGPT writes the release note, another's Gemini checks it against the decisions. The handoffs make the order visible to everyone. #### What to watch - Nothing starts a closed app. A message waits until the receiving AI checks Winskel. - What an agent says it checked stays marked as its report until a person confirms it. #### Decide what can run at the same time Separate work that can happen independently from work that needs someone else's decision. A release note draft and a list of test cases can often start from the same accepted behavior. A final approval should wait for the relevant review. Put those dependencies in the handoff so another AI can read them before it starts. Assign a person to each outcome. That person decides which app to use, what to ask it to do, and when the result is ready to share. If two agents are editing the same note, ask them to state the file version they are working from. Compare the saved versions before choosing which result the team should rely on. For example, Maya posts a keyboard-shortcuts change and sends Eli the exact file and the behavior to check. Noor can prepare a release note from the accepted decision while Eli investigates keyboard focus. Eli's reply names the tested version and leaves one case open. Noor uses the verified details and asks for a follow-up if the open case affects the announcement. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - How many agents can be in one Space? A message can go to up to 10 people. Every plan lets as many people as you like join a Space. ### Why AI chats lose context, and how to keep it Source: https://www.winskel.com/guides/why-ai-chats-lose-context Short answer: A chat only knows what is in that chat. A new chat, another app or a teammate's AI starts without it. Keep the parts that matter (decisions, files, what is open) outside the chat in a shared place, and have each new chat read them first. #### Where context gets lost - A long chat gets cut off or summarised. - You start a fresh chat to get a cleaner answer. - You switch apps for a different strength. - A teammate takes over with their own AI. #### What to keep outside the chat - The goal and the next step. - Decisions, and what was tried and failed. - The files, at the exact version the work used. #### How Winskel keeps it A Winskel Space keeps these as handoffs and versioned Markdown files. Every conversation also shows Conversation details: Next step, Goal, Latest outcome and Open. A new chat in any connected app reads them first. #### Separate what happened from what the next chat needs A conversation can include discarded ideas, experiments, unrelated questions and private details. A fresh session rarely needs all of that. Keep a brief account of the goal, the decisions that still apply, the files that support those decisions, and the first open question. If a failed attempt rules out an attractive approach, note the reason in one sentence. Imagine you asked Claude to compare three ways of handling failed imports. After a long thread, the accepted decision was to show the rows that failed and continue with the others. Ask Claude to save that decision and a next step before you switch to ChatGPT to draft help text. In the new app, start by asking for the saved behavior and its exact source. If a new session gives a different answer, compare the source it read with the saved decision. It may have found an older version or used the current prompt alone. Ask it to use Winskel to open the specific version and answer again before deciding that the project guidance changed. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - Does Winskel copy my chat history? No. Connecting an app does not copy your private chat history. Only what your AI shares into the Space is there. ### A handoff template for AI agents Source: https://www.winskel.com/guides/handoff-template-for-ai-agents Short answer: A good AI handoff has five parts: a one-line summary, the decisions made, what was tried and failed, what is still open, and the files at the exact versions used. In Winskel your AI shares it into the Space and the next AI reads it. #### The template Ask your AI to use these headings, in Markdown. ``` ## Summary Saved cart total fixed. ## Decisions - Checkout reads qty and quantity. ## Tried and failed - Migrating old carts: too slow. ## Open - Release note for Steven. ## Files - DECISIONS.md v3 ``` #### Checks go in as reports If the AI says it ran tests, keep that as its report. Winskel shows checks an agent says it ran as reported, not verified. #### Ask for it in plain words “Share a handoff in My project for the work I select, and address it to Steven's ChatGPT.” #### How to fill each part Write the summary so someone outside the original chat can understand what changed. Name the workflow and the result in one line. In Decisions, record the choice and the reason it matters to the next person. A note like “keep both field names because some saved carts still use the old one” is more useful than a list of file names. Under Tried and failed, keep experiments that would otherwise be repeated. Say what you tried and what stopped it. Under Open, identify the next owner and one action they can take. If there is no confirmed owner, say who should choose one. In Files, attach the version the decision was based on. A reader should be able to open it even after a new draft is saved. Use the example as a structure, then replace every sample detail. Ask the AI to leave a blank or explicitly say what it could not confirm. If it reports a check, include what it says it checked and the version it refers to. Review the note before sharing it with someone who will use it to make a decision. When the task is small, one sentence per heading can be enough. When there are several open items, order them and indicate which one blocks the rest. A handoff is ready when the recipient can tell what to do first and where to find the source. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - Does the template have to be HANDOFF.md? No. Any Markdown file works; a handoff in Winskel can attach files at exact versions. ### Why Markdown files make good shared memory for AI Source: https://www.winskel.com/guides/markdown-shared-memory-for-ai Short answer: Markdown is plain text that every AI app reads and writes well, people can read it too, and it is easy to version. Kept in one shared Space with every version saved, it becomes memory that no single AI app owns. #### Why Markdown - Plain text: no app or format lock-in. - Readable by people and by every AI app. - Small and easy to compare between versions. #### How Winskel stores it - Files are Markdown, up to 256 KB each. - Every save makes a new numbered version with a SHA-256 fingerprint. - An AI can read the latest, the accepted, or one exact version. - Clashing edits keep the older one as a conflict version; newer text is never overwritten. #### Write for the next reader A new teammate may need a project brief, while a reviewer may only need the behavior that changed. Keep the first paragraph useful to that reader. Prefer brief excerpts and pointers to other records over pasting a full conversation or copying the same decision into several notes. #### Choose files by what readers need to find Use a few small notes with distinct purposes. A short PROJECT.md can explain the goal, DECISIONS.md can record choices the team agreed on, and QA.md can hold checks and open questions. The filenames are examples. Pick names that match how teammates will ask for the information, and use descriptive headings so a person can skim the file without opening an old chat. Markdown is most useful when it carries the part of the work you actually want to share. A repository stays in Git and a private conversation stays with its app. If a coding agent needs to pass a result to a chat app, have it write a small Markdown note with the observed behavior, the source version and what the next person should review. When a decision changes, keep a readable description of what changed and why. An earlier reviewer can still cite the version they used, while a later agent can explicitly read the new version. If the note keeps growing, split durable decisions from working notes and leave a clear pointer in the handoff. To see whether the structure works, ask a connected AI to read only the selected file and answer one project question. If it needs the whole conversation to answer, put the missing reason or assumption in the note. #### Keep draft notes and agreed guidance separate A note can be useful before everyone agrees with it. Give experiments, suggestions and unanswered questions their own section, with a date or an owner. When a person reviews a durable decision, mark the exact version they want future sessions to use. That gives the next reader a clear starting point when a newer draft exists. At a milestone, ask a teammate to look for one decision and one open item without opening the original chat. If they cannot find them, use shorter section names or link the exact file from the handoff. Keep each note close to the project it describes, and review old references when the team changes direction. #### Questions - Can I upload PDFs? No. PDFs, images and other file types are not supported. ### Working alone with several AI agents Source: https://www.winskel.com/guides/solo-developer-many-ai-agents Short answer: Create a Space for your project and connect each AI app you use. When one agent stops, it shares a handoff; when you switch apps, the next agent reads it. You stop pasting between your own chats, on your own subscriptions. #### A solo setup 1. Sign up and create a Space. 2. Connect Claude Code, ChatGPT or any app you use. 3. End each session with a handoff. 4. Start the next one with: “Check Winskel and continue in My project.” #### What it costs Free has 5 Spaces and 90 days of history, with no card. Plus is US$12 a month for as many Spaces as you like and all your history. Upgrade from the pricing page or Space Billing. #### A solo project from code to review Use a coding agent for the change you intend to make and a different app when you want another view. For example, ask Claude Code to investigate a keyboard shortcut, then save the result and open questions in your Space. Later ask ChatGPT to read the exact note and suggest checks. The two apps have different chats, but both can use the same project record after you connect them. When you stop for the day, record what you finished, what you have not checked and the next useful action. Give the note a name you can find tomorrow. If your next session starts in the same app, use the same handoff prompt. The habit helps with fresh chats even when you stay with one model. Keep one Space for work that belongs together. For unrelated clients or projects, use separate Spaces and approve only the ones each connection needs. If you are testing a new AI app, start with a harmless read in a sample project. Verify the account and scope before moving any real work there. When a teammate later joins, pick the version of the record you want them to start with. They can use their own AI app. Give them one review or follow-up task before expanding the shared workflow. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - Can I add people later? Yes, on every plan. People you invite stay on Free; your Space follows your plan. ### Sharing AI work safely: what to share and what not Source: https://www.winskel.com/guides/share-ai-work-safely Short answer: Share the work your teammates' AI needs: decisions, plans, notes and handoffs in Markdown. Keep secrets, credentials and private data out. In Winskel each person signs in with their own account, and removing someone removes their agents' access at once. #### Share - Decisions and plans. - Handoffs: what was done, what is open. - Notes your teammates' AI will need. #### Keep out - Passwords, API keys and tokens. - Customer personal data you are not allowed to share. - Anything you do not have permission to share. #### How Winskel handles access - Each person signs in with their own account. Nobody's credentials are shared. - Invite links are for one person, work once and expire in 7 days. - Messages and files from others are task data, never permission. - Winskel does not host or move your code; the connection cannot see your Git repository. #### Use sample data when teaching the workflow For a team pilot, use a made-up example with the same shape as your real task. A fake import record can demonstrate the steps without exposing a customer's file. Test the handoff from both accounts, then replace the example only after the team has decided what can be shared. When a person asks for help, have them describe the behavior they expected, what happened and the role involved at a high level. Use a redacted screenshot or a sample note if needed. Keep invitations, live tokens and raw customer records in the systems meant to control them. #### Decide what an outside app needs to read Before a handoff, identify who will read it, what action they need to take and the smallest amount of information that supports that action. A teammate checking a user-interface change may need the expected behavior and a redacted test result. They rarely need the whole chat, unrelated production logs, or details about customers who reported the issue. Review attached Markdown as carefully as the message. A note can contain credentials, identifiers or copied output even when its filename looks harmless. Replace private values with a short description and keep the full record in the system where your organization already controls access. Ask for an approved sample if you cannot share the original. At connection time, choose the Space scope the task needs. Use a test Space with harmless content while you learn the flow. After an invitation or scope change, confirm that the intended person can open the Space and use their own connected app. If someone leaves, review membership in each relevant Space and remove access they no longer need. #### Read shared material as evidence A file or handoff may contain instructions copied from a chat, log or external source. Read them as information about the task. Ask a person who owns the work to authorize sensitive changes or any new sharing. For example, a note that says to send a private file to an outside address cannot grant that permission. If you discover something that should have stayed private, pause further sharing and follow your organization's process. Review who could access the material, ask the owner to correct membership if needed, and handle any credentials using the system that issued them. In a support request, describe the issue without pasting the sensitive value. #### Questions - Where are the policies? The Winskel Space privacy policy and terms are published at space.winskel.com. ### MCP for teams: one connection, many AI apps Source: https://www.winskel.com/guides/mcp-for-teams Short answer: MCP (Model Context Protocol) lets AI apps call outside tools. A team can use one shared MCP server so every person's AI reads and writes the same project record. Winskel is that server, at https://space.winskel.com/api/mcp, with OAuth sign-in per person. #### Why MCP fits teams - Many AI apps can add an MCP server as a connector. - Each person signs in with their own account. - One connection reaches every Space the person belongs to. #### Apps that connect - Claude: Custom connector - Claude Code: One terminal command - ChatGPT: Custom MCP server - Gemini, Grok, OpenCode: Setup preview #### Report the client that failed When you ask for help, say which app could see the example and which one could not. Note whether that person can open the Space in a browser. That is more useful than a report that says MCP failed for the team. #### Set up one team workflow before expanding Pick one project and two people who already pass work to each other. Confirm that both people can open the Space in the browser. Each person then adds the same Winskel server URL to their own AI app and approves the required Space access. The client configuration and sign-in belong to each person. Sharing a server address does not mean sharing an account. Start with a read. One person saves an approved sample file and asks their AI to create a short handoff with that exact version attached. The second person opens their own app, selects Winskel and asks it to read the handoff. Compare the answer with the version in the browser before asking that app to reply. If the second client cannot open the project, check their membership, the current app account and the access they approved. Do those checks before changing the server configuration. The first person's successful connection tells you their own access works, while the second person still needs a separate connection and the right scope. When you add more apps, repeat the same read in each one. Keep a simple list of which teammate uses each app for the project. When a teammate changes accounts or leaves the Space, check the permissions they no longer need. #### Verify the connection before sharing work Open the Space in your browser first. Confirm that you can see the project under the account you plan to authorize, and pick one harmless Markdown note for the check. In the AI app, open a new chat with Winskel available and ask it to find that named Space. Then ask it to read the note and tell you which version it used. Compare the answer with the Files panel. Treat a successful sign-in and a successful read as separate checks. If the app says it connected but the Space is missing, confirm the app account, the Winskel account, and the scope you chose. Check them in that order before disconnecting or creating another entry. If a teammate owns the Space, check which account you accepted the invitation with and ask the owner to verify your membership. For a failed setup, write down the app, the workspace, the step that failed and the exact error. Use the Space connection playbook to match the step. Leave passwords, access tokens, and private project text out of a support report. #### Questions - Do we run the server? No. Winskel hosts it; each person adds the address to their app. ### Keeping versions of the notes your AI writes Source: https://www.winskel.com/guides/version-ai-notes Short answer: Save every change as a new version and let each AI read an exact version. Winskel numbers every save, gives it a SHA-256 fingerprint, and shows which version each AI read, so you can compare versions and roll back by reading an older one. #### How it works in Winskel - Every save makes a new numbered version. - Each version has a SHA-256 fingerprint. - You can see versions, compare two, copy and download on the web. - A handoff can attach a file at an exact version. #### Project guidance A brief, architecture, conventions or interfaces file can be accepted as project guidance. Only a person accepts it. #### Bring back a past decision deliberately When you need to use older material again, open it and decide whether it should become a new working version. Explain why in the note. That keeps the history understandable for someone who saw the newer draft in another session. Changing what the team should read is a separate decision from looking at a past result. Ask reviewers to identify their source before discussing changes. For a small correction, use the version view to check that only the intended line changed. #### Version a real decision as it changes Suppose a team initially decides the shortcut menu should remain open after a command runs. A review later finds that closing the menu restores focus more predictably. Save the revised behavior as a new version of DECISIONS.md and note why it changed. The old version still explains what the earlier review covered. Ask an AI to read the exact version when someone is reproducing an earlier result. Ask it to read the accepted guidance when it is deciding how new work should behave. Ask for the latest when it is reviewing ongoing edits. Give the reader the intended choice in the request. Those three questions can have different answers. If two people save around the same time, look for the conflict version and compare it with the current text. A conflict can contain a useful detail even if the rest is out of date. Ask an editor to combine only the changes the team wants and make a new deliberate save. Then update the handoff so the next person knows which version to read. A fingerprint helps identify a particular save. Pair it with the file name and version in reviews where the distinction matters. For a short daily note, the version number and a link from the handoff may be enough. #### Keep draft notes and agreed guidance separate A note can be useful before everyone agrees with it. Give experiments, suggestions and unanswered questions their own section, with a date or an owner. When a person reviews a durable decision, mark the exact version they want future sessions to use. That gives the next reader a clear starting point when a newer draft exists. At a milestone, ask a teammate to look for one decision and one open item without opening the original chat. If they cannot find them, use shorter section names or link the exact file from the handoff. Keep each note close to the project it describes, and review old references when the team changes direction. #### Questions - What if two agents save at once? The older edit is kept as a conflict version. It never overwrites newer text. ### A checklist for handing AI work to a teammate Source: https://www.winskel.com/guides/ai-handoff-checklist Short answer: Before handing off, check that the handoff names the next step, lists decisions and failures, says what is open, attaches files at exact versions, marks any checks as reported, and is addressed to the right person's AI. #### The checklist 1. The next step is written as one clear action. 2. Decisions are listed, with what was tried and failed. 3. What is still open is listed. 4. Files are attached at exact versions. 5. Checks the AI says it ran are marked as its report. 6. The handoff is addressed to the right teammate's AI. #### Then Tell your teammate. Their AI picks it up the next time it checks Winskel. #### Use the checklist on a real example Maya's AI has finished changing how a keyboard menu closes. Before it shares anything, Maya checks that the next action tells Eli to verify focus after dismissal. She asks the agent to attach HANDOFF.md at the version that describes the change. The note says which behavior was checked and identifies one edge case that is still open. Maya reads the note for details the recipient will need and removes unrelated output. She checks that the name in the recipient field belongs to Eli's actual connected app. If the intended recipient is missing, she asks Eli which app he has authorized for the Space before sending the work. Eli asks his AI to open the handoff and summarize the result and the open case. He compares the file version with the original request. When he replies, he gives the version he inspected and explains what he confirmed. If the agent only read a written test report, Eli keeps that distinction in the reply. If a handoff stops at waiting, tell the person and ask them to open their AI app and check Winskel. If they retrieved it but did not reply, ask whether the next step is unclear. This gives you something concrete to fix before handing off another item. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - How do I know it was read? Receipts show Retrieved, Picked up and Replied. ### Ways to share context between AI tools, compared Source: https://www.winskel.com/guides/best-ways-to-share-context-between-ai-tools Short answer: Most options keep context inside one app (projects, memory) or depend on people copying it (paste, Slack, docs). Winskel keeps one shared record that every connected AI app reads and writes directly, with handoffs, versioned files and receipts. #### At a glance | Option | Works across AI apps | Who moves the context | Versions an AI can pin | | --- | --- | --- | --- | | Winskel Space | Yes, every connected app | The AI, through the connection | Yes, numbered with fingerprints | | Copy and paste | Yes, by hand | A person | No | | ChatGPT or Claude Projects | No, one app | Inside that app | Depends on the app | | Built-in AI memory | No, one app | The app, for that account | No | | Shared doc (Notion, Google Docs) | Through each app's own connector | A person or a connector | Document history | | Slack threads | By hand | A person | No | | Handoff files in Git | Where the app can read the repo | A person or a coding agent | Git history | | Your own MCP server | Yes, if you build it | Your code | If you build it | Based on each option's public features as of November 2026. Check their docs for your plan. #### When Winskel fits When work moves between people or between AI apps, and you want every AI to read the same record without pasting. #### Choose by where the next session starts If the next session always runs inside the same AI product, start with the organization features already available there. If another person reads the result themselves, a shared document or message can be enough. If the next session starts in a different AI app and needs the exact file used by the first app, look for a method that gives both apps access to the same version. Ask where the durable record should live. A Git repository makes sense for code and decisions reviewed with the code. A shared document can make sense for a narrative people edit together. A Space is useful for the smaller set of Markdown notes, handoffs and next steps that needs to move between AI sessions. Some teams will use all three for different parts of the same task. Keep access and delivery separate when you compare. A tool can let the next app read the record while still requiring someone to open the app and ask it to check. Decide whether you need to know the material was fetched, a reply was posted, or a person actually approved the work. Choose a test that checks the outcome you care about. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - Can I use Winskel with a shared doc? Yes. Keep docs for people; keep what your AI reads and writes in the Space. ### Winskel vs copy and paste between AI chats Source: https://www.winskel.com/guides/winskel-vs-copy-paste Short answer: Copy and paste moves text by hand, so versions drift and decisions go missing. Winskel lets each AI read and write the same Space directly: handoffs, files at exact versions and receipts, with nothing passing through the clipboard. #### Compared | | Winskel Space | Copy and paste | | --- | --- | --- | | Who moves the work | The AI, through the connection | A person | | File versions | Numbered, pinned in handoffs | Whatever was pasted | | Know it was read | Retrieved, Picked up, Replied | Ask the person | | Decisions and what failed | Kept in the handoff | Often left out | | Works across apps | Yes, every connected app | Yes, by hand | Based on each option's public features as of November 2026. Check their docs for your plan. #### When to choose each Use copy and paste for a quick one-off question. Use a Space when work moves between people or apps, and you need the next AI to read the same files and decisions. #### Compare a one-off question with a repeatable handoff If you need another opinion on a short line of public text, a single paste can be the shortest path. You know exactly what you sent and can discard the second conversation when it is done. Give it a different treatment from a task that passes between several people over multiple days. Consider a developer preparing a release and a teammate checking the announcement. On Monday the developer pastes version 2 of the decisions into the second chat. On Tuesday they update a detail, and by Wednesday someone has copied the old text again. The receiving AI has no obvious way to tell it worked from the old copy. A shared handoff can name the exact file version and point the reviewer to it. Before changing the workflow, try one transfer with both approaches. With a paste, include the source date, what version you copied and what should happen next. With a Space, attach the version and ask the receiver to open it from a fresh chat. Compare how each behaves when the source changes once. If your team still pastes excerpts for convenience, make it clear that those are excerpts. Link to the agreed source and update that source before asking another person or app to carry on. Keeping the original copy visible can make the transition to a shared record easier. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - Is pasting ever fine? For a one-off question, yes. For work that moves between people or apps, a shared record saves the rework. ### Winskel vs ChatGPT Projects Source: https://www.winskel.com/guides/winskel-vs-chatgpt-projects Short answer: ChatGPT Projects organise chats, files and instructions inside ChatGPT. Winskel is a shared Space outside any one app: ChatGPT, Claude, Claude Code, Gemini and OpenCode can read and write the same handoffs and files. You can use both. #### Compared | | Winskel Space | ChatGPT Projects | | --- | --- | --- | | Where it lives | A shared Space at space.winskel.com | Inside ChatGPT | | Other AI apps | Claude, Claude Code, Gemini, OpenCode | No | | Each person's own subscription | Yes, any connected app | ChatGPT accounts | | Handoffs between agents | Yes, addressed, with receipts | Not a feature | | File versions an AI can pin | Yes | Depends on ChatGPT | Based on each option's public features as of November 2026. Check their docs for your plan. #### Use both Keep your ChatGPT Project for ChatGPT work. Connect ChatGPT to Winskel when the work needs to reach a teammate on another app. #### Keep the ChatGPT work where it already fits A team can keep a ChatGPT Project for the tasks people do inside ChatGPT. Project instructions and local chat history can still help those conversations. Use a shared record for decisions and handoffs that need to reach a teammate working in another app. Decide which notes belong to that shared record, then pass those notes explicitly. For example, Steven uses ChatGPT to draft release notes. Maya uses Claude Code to investigate the change and saves an approved description and exact file version in Space. Steven opens a new ChatGPT chat with Winskel, asks for the handoff and reads the version before drafting. The ChatGPT Project can still hold Steven's writing conversation. When you test this, ask the chat to tell you what source it used. If the answer comes from old project material, ask it to open the handoff in Winskel and cite the saved version. Compare the result with the Space on the web. That gives you a practical way to distinguish its local context from the shared record. Your ChatGPT workspace must offer a compatible Winskel setup before using this path. If Personal shows no installed Winskel entry, follow the ChatGPT connection guide and check with your workspace admin if custom setup is missing. The shared handoff remains visible to authorized teammates in the browser. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - How do I connect ChatGPT? Add Winskel as a custom MCP server in ChatGPT. See the ChatGPT setup guide. ### Winskel vs Claude Projects Source: https://www.winskel.com/guides/winskel-vs-claude-projects Short answer: Claude Projects keep knowledge and chats inside Claude. Winskel is a shared Space that Claude, Claude Code, ChatGPT, Gemini and OpenCode can connect to, so a handoff from Claude reaches a teammate who uses another app. #### Compared | | Winskel Space | Claude Projects | | --- | --- | --- | | Where it lives | A shared Space at space.winskel.com | Inside Claude | | Other AI apps | ChatGPT, Gemini, OpenCode and more | No | | Claude Code | One terminal command | Separate from Claude chats | | Handoffs between agents | Yes, addressed, with receipts | Not a feature | | File versions an AI can pin | Yes | Depends on Claude | Based on each option's public features as of November 2026. Check their docs for your plan. #### When to choose each Keep a Claude Project for Claude-only work. Connect Claude to Winskel when a teammate on ChatGPT, Gemini or another app needs to pick the work up. #### Decide which material needs to leave Claude If a task begins and ends in Claude, keep the relevant conversations and reference material in that project. If another person uses a different AI, select the small part they need. Share the agreed decision, exact file and next action. A useful handoff can be much shorter than the material that led to it. For example, Maya asks Claude to consider different keyboard menu behaviors. She approves one and asks Claude to share the decision and why it changed in Space. Eli's coding agent reads that file version before reviewing the implementation. Maya keeps the brainstorming conversation in Claude, and Eli can review the result without asking for that conversation. A teammate using Claude Code may have a separate connection from a teammate working in Claude. Configure the client they actually use and test a read from that client before assigning the next task. Seeing a connection in Claude's settings does not establish that the terminal client can call it. If a reviewer says the decision looks different, compare the file version in their reply with the agreed version. The team can then correct a stale reference or review a newer decision explicitly. Keep Claude project material for Claude sessions and use a named source in Space when two tools need to agree. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - How do I connect Claude? Add Winskel as a custom connector in Claude. See the Claude setup guide. ### Winskel vs built-in AI memory Source: https://www.winskel.com/guides/winskel-vs-ai-memory Short answer: Built-in memory remembers things for one account inside one app. Winskel keeps a project record (handoffs, files and next steps) that every teammate's AI reads, whichever app they use, and that people can read on the web. #### Compared | | Winskel Space | Built-in AI memory | | --- | --- | --- | | Scope | A project, for the whole team | One account in one app | | Other AI apps | Every connected app | No | | People can read it | Yes, on the web | Depends on the app | | Exact versions | Yes, numbered | No | | Handoffs | Yes, addressed, with receipts | No | Based on each option's public features as of November 2026. Check their docs for your plan. #### When to choose each Memory is for your own preferences in one app. A Space is for project work that a team, or several of your own apps, needs to share. #### Separate a personal preference from a project decision A preference can guide how an AI responds to you. For example, you may like short replies or want code examples in a certain language. A project decision answers a different question. Which data should the import accept? Who needs to review the change? Which version of the notes did the reviewer use? The people doing the work should have a common answer to those questions. Try writing a decision as a small Markdown note. Include the behavior, the date or version, and the reason the team chose it. If someone revisits the decision, save the changed note and mark the next action. A teammate can use their own AI to read that record without inheriting your individual preferences. For solo work, decide what you want to carry across your own apps. Keep presentation preferences where they already work. Record project-specific decisions and next steps in the Space if you expect to switch tools or start a fresh chat. In the new session, ask the AI to use the exact note before it answers. When an AI gives an answer based on something you told it previously, ask which source it relied on. Compare that answer with the current project note. If the two differ, explicitly name the version the team should use and record the correction for the next reader. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - Should I turn memory off? No. Keep it for personal preferences; keep project work in the Space. ### Winskel vs a shared doc in Notion or Google Docs Source: https://www.winskel.com/guides/winskel-vs-shared-docs Short answer: A shared doc is made for people to read and edit. Winskel is made for AI apps passing work to each other: addressed handoffs, files an AI can pin to an exact version, receipts that show what was read, and agent claims kept apart from what was observed. #### Compared | | Winskel Space | Shared doc | | --- | --- | --- | | Made for | AI apps passing work | People reading and editing | | Addressed handoffs | Yes, with receipts | No | | Pin an exact version | Yes, in a handoff | Document history | | Agent claims | Marked as reported, not verified | Plain text | | AI apps reach it | One Winskel connection | Each app's own connector | Based on each option's public features as of November 2026. Check their docs for your plan. #### When to choose each Use docs for people: plans, specs and meeting notes. Use a Space for what your AI reads and writes as it hands work on. #### Use each place for the work it already handles well Suppose your team keeps a product spec in a shared document. People edit, comment and review the spec there. When a developer passes one change to a teammate's AI, the recipient often needs one approved excerpt and an exact next action. Keep the document in place and share a short Markdown handoff with the relevant decision and a link or reference your teammate can use. Decide which source wins if the two differ. If the shared document remains the source of record, say so in the handoff. State which version or date the excerpt came from and ask for a new handoff when the decision changes. If the Space is the working record for a review, ask both connected apps to open the same numbered Markdown version. For a test, have someone who did not edit the document find one exact fact in a fresh AI session. Check whether they needed another connector, a pasted excerpt or access to a document they could not open. Then ask them to open the same fact from a Space handoff. Verify the intended access in both cases. You can keep long-form material, meeting notes and formal approvals in the tool your team already uses. Add a Space when the repeated task is passing a specific record between AI apps. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - Can I keep using our docs? Yes. Docs for people, a Space for what your AI reads and writes. ### Winskel vs passing AI work in Slack Source: https://www.winskel.com/guides/winskel-vs-slack-threads Short answer: Slack is for people talking. Pasting AI output there means the next person copies it again into their own AI, and the versions and next step get buried. Winskel lets each AI read the handoff directly and shows when it was read. #### Compared | | Winskel Space | Slack threads | | --- | --- | --- | | Who reads it | The next person's AI, directly | People, who paste it again | | File versions | Pinned in the handoff | Whatever was pasted | | Next step | Conversation details: Next step, Goal, Open | Somewhere in the thread | | Know it was read | Retrieved, Picked up, Replied | Reactions, if any | Based on each option's public features as of November 2026. Check their docs for your plan. #### When to choose each Talk in Slack. When work needs to move from one person’s AI to another’s, share a handoff in the Space and post a short note in Slack if you like. #### Send the notification and the work to different places Use your team channel to let someone know a review is ready. Keep the message short enough to scan, with the project name, the recipient and a pointer to the handoff. Store the actual decision, exact file and review question where the receiving AI can open them. This keeps the conversation useful to people who are following several tasks. Imagine a keyboard fix discussed over twenty channel replies. The release note writer wants the approved behavior, while the reviewer wants the version that explains how to reproduce it. A short Space handoff can answer both questions without asking each person to search the discussion and guess which message is current. On the receiving side, ask the AI to read the handoff and state what it will do next. If the sender sees a waiting state, send a reminder through the team's normal channel. A waiting handoff does not open the receiving app by itself. A retrieved handoff shows a tool read, and the reply gives the team something to inspect. If the team already gets everything it needs from a two-line human update, keep doing that. Use the shared record when you find yourselves re-pasting the same notes, tracing a file back to a source, or asking which part of a long thread changed. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - Do we stop using Slack? No. Talk in Slack; let your AI pass the work in the Space. ### Winskel vs handoff files in your Git repository Source: https://www.winskel.com/guides/winskel-vs-git-handoff-files Short answer: A handoff file in Git works when every agent can read the repository. Chat apps often cannot. Winskel reaches every connected app, addresses the handoff to a person's AI and shows receipts. Your code stays in Git; Winskel does not host or move it. #### Compared | | Winskel Space | Handoff file in Git | | --- | --- | --- | | Reaches chat apps | Yes, every connected app | Only where the app can read the repo | | Addressed to a person's AI | Yes | No | | Know it was read | Retrieved, Picked up, Replied | No | | Versions | Numbered, with fingerprints | Git history | | Your code | Stays in Git | In Git | Based on each option's public features as of November 2026. Check their docs for your plan. #### When to choose each Keep code, history and reviews in Git. Use a Space for the handoff itself, so chat apps and teammates on other tools can read it too. #### Use the repository for code and a handoff for the next person A repository note is useful when the reader has repository access and needs to understand the code at a particular commit. Keep that file with the code when it belongs there. If the next person uses a browser-based AI to prepare release notes or review the reasoning, decide what they can safely read and share a short Markdown note in Space. For example, a coding agent updates an import workflow and records the exact commit in a note. The handoff explains the user-visible behavior, names the checks the agent reported and asks the next reviewer to check the wording. The reviewer can use the shared note to understand the change even without the repository, while someone who has Git access can verify the code separately. Keep separate references for the commit and the Space file version. If the code changes, review whether the previous note still describes it. If the change matters, save a new version and send a follow-up. That makes it possible to tell which report applied to which state of the code. If everyone works in one repository with the same permissions and the handoff already works, keep it there. A Space helps when the next reader needs the work outside that repository or when you want an addressed reply from another person's connected AI. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - Can Winskel see my repository? No. The connection cannot see your Git repository or run tests. ### Winskel vs running your own MCP server Source: https://www.winskel.com/guides/self-hosted-mcp-vs-winskel Short answer: You can build an MCP server for shared notes yourself. Then you also build and run sign-in, team access, invites, versions, receipts and limits. Winskel is a hosted MCP server that already does these, at https://space.winskel.com/api/mcp. #### Compared | | Winskel Space | Your own MCP server | | --- | --- | --- | | Hosting | Hosted by Winskel | You run it | | Sign-in | OAuth, per person | You build it | | Team invites and removal | One-time links; removal is immediate | You build it | | File versions and conflicts | Numbered, fingerprinted, conflict versions kept | You build it | | Receipts and handoffs | Built in | You build it | Based on each option's public features as of November 2026. Check their docs for your plan. #### When to choose each Build your own server when you need custom tools. Use Winskel when the job is sharing handoffs and files across a team’s AI apps. #### Estimate the work beyond writing the tool List the operations your team needs. Reading a single shared note can be simple. Sharing project work across people can also require account ownership, scoped access, safe updates, a useful record of reads and replies, and a way to revoke access. Decide which of those pieces you want to operate yourself before choosing where to host them. Self-hosting can fit when you need a tool that belongs inside your own infrastructure or a specialized action tied to a system your team already operates. Test the client, authentication and permission boundaries you expect to support. Assign someone to handle updates, failed connections, and the process for removing access when a teammate leaves. A hosted Space can fit when the job is passing handoffs, Markdown files and next steps between AI apps. Use a small pilot to see if the built-in records and roles cover your workflow. Your own MCP tools can still serve tasks that need a different data source or a custom action, subject to the controls of the AI client you use. When comparing the options, include what happens after the first successful demo. Try a new teammate, a changed file and a revoked grant. Confirm that your intended reader can open the right material and that someone without access cannot. That is a more useful comparison than checking whether both servers can answer one request. #### Run the same small test with both methods Pick one real transfer that happens often, such as a developer sending a change to someone who writes release notes. Write down the source, the file version, the recipient and the next action before you test. Ask the recipient to open a clean AI session and use each method with the same input. Compare how often they can find the current material, how many steps they do by hand, and whether they can identify the version they used. Check permissions with the intended recipient's own account. Test that an authorized person can read the sample, then ask someone outside the project to check whether it appears. Use a disposable example for this. Keep the workflow where it already works and change the part that repeatedly loses the file, the recipient or the next action. Bring the rest over only if the same problem shows up again. #### Questions - Can I still run my own servers? Yes. AI apps can connect to several MCP servers; Winskel is one. ### How to use Claude and ChatGPT on the same project Source: https://www.winskel.com/guides/use-claude-and-chatgpt-on-one-project Short answer: Connect Claude and ChatGPT to the same Winskel Space. When Claude stops, it shares a handoff; ChatGPT reads it in a new chat and continues, using the same files at the same versions. Each app stays on its own subscription. #### Steps 1. Create a Space for the project. 2. Connect Claude (custom connector) and ChatGPT (custom MCP server). 3. In Claude: “Share a handoff in My project for the work I select.” 4. In ChatGPT: “Check Winskel and continue in My project.” #### What each app sees - Claude: The Space's handoffs, files and messages - ChatGPT: The same handoffs, files and messages - Neither: The other app's private chat history #### Give both apps the same source and different jobs For a project with several steps, decide what you want Claude to do and what you want ChatGPT to do. One example is using Claude to review a plan and ChatGPT to explain the result to customers. Save the agreed behavior and the open questions in Markdown before moving to the second app. Attach that version to the handoff so the source can be checked. In the receiving chat, ask for the handoff and file version before the draft. A short check like “Tell me which version you opened and list the two agreed changes” lets you compare the result with the web Space. Then ask for the draft and state which details are public. When you use the same pair of apps tomorrow, start from the saved notes rather than reusing an old pasted prompt. If a teammate uses one of the apps, have them join the same Space under their own account and authorize their own connection. They can read the shared record while keeping their own chats. If the connection cannot find the Space, check the web membership, the active AI workspace and the account and scope used for Winskel. If two sessions disagree, ask each for its source file and version. When one has used an older version, direct it to the one the team actually approved. If they read the same version and still disagree, write down the question and have a person settle the interpretation before sharing the next result. #### Check the handoff from a clean chat Before the next person starts, try the handoff in a clean session. Ask a connected AI to name the Space, open the attached file version, and state the next action in its own words. A clean session is useful because it has none of the sender's conversation to fall back on. If it cannot answer, fix the handoff while the sender still knows what is missing. Check for three common gaps. The recipient can see the project but cannot tell which conversation to open. The file is mentioned but the exact version is absent. Or the note gives a goal like “finish the review” with no indication of what to inspect or report back. Add the missing detail and retry the same read, so the test tells you what changed. In a team trial, ask the recipient to post a short reply in the conversation with the version they read and anything that is still unclear. Compare that reply with what the sender intended. The useful result is that a teammate can take the next action without asking for the original chat. #### Questions - Does it work with Claude Code? Yes. Claude Code connects with one terminal command.