What's new

Commands and agents

• Before /plan-prompt saves a plan, it now also asks whether the change holds in every other place the same thing happens, what happens when two mechanisms reach the same person or record, what becomes of whatever points at what is being changed or removed, and whether the change should behave like a flow that already does nearly the same. Each question comes with the answer it recommends. Planning before building
• Before /coordinate marks a goal as reached, it corrects what the goal's own work turned up on its screens and flows, with a plan in the repository that owns the code, instead of asking you about it. Coordinate a goal across repositories

The website

• The features page now lists each feature on its own, with its screenshot, and lets you filter by section; every feature has a page of its own.

Help centre

• The help centre is wider, keeps the "On this page" index of a guide beside its text, has an Up button, and shows the changelog ten days to a page. A guide's back link returns you to the page of this site you came from.
• Several fixes were made to the help centre.

Corrections

• Several fixes were made to the platform.
• Several fixes were made elsewhere in the product.

The platform

• In a project's viewer, a folder with an index.md or a README.md shows it beside its listing, an open document keeps its folder's listing on the left with a row to go up one folder, a link to another document opens it from the folder it was written in, and the images of a folder open in a viewer you step through as one gallery. Your projects and their documents
• Images uploaded to a project's identity/ folder are now kept exactly as they arrived, instead of being scaled down, so your logos and manual stay as you sent them. Your projects and their documents
• Several fixes were made to the platform.

The MCP server

• Your assistant can now upload files such as a PDF manual or a logo to your projects, up to 20 MB each, and list, read and delete them beside your documents. Your projects and their documents
• When your assistant's copy of the toolkit is already current, asking it to install the toolkit answers that it is up to date, with nothing to write. Install the toolkit

Commands and agents

• /coordinate is new: it carries a goal that needs changes in weblabor-base, weblabor-teams and your product to the end, writing the plan each repository needs, watching every queue and bringing each finished change down until the goal works in your product. Coordinate a goal across repositories
• /plan-prompt now closes each plan with a technical plan the build follows as guidance, and a change you will see comes with a drawing of the screen as it is today and as it would look, for you to approve before anything is built. Planning before building
• /test now checks every screen it tests in each language, at phone, tablet and laptop widths, empty and with records, while you wait on an action, after a save is reloaded, with an empty form and on buttons that show only an icon. A defect it finds on a tested screen now fails the plan and goes into its correction, instead of being noted under the verdict. Verify a plan with /test
• A defect on the screen or flow a command is working on is now corrected in the same run, instead of being written down as a pending finding for later. Pending findings
• Implemented plans are no longer deleted a week after their last change: /finish, run from your main branch, deletes the plans and coordination documents whose work reached it, with the drawing saved beside a plan. Run the queue with /next
• /design-catalog now shows its proposals for your product's look as drawings, in rounds where you only choose what you see, keeps the original files of your identity manual, and follows the queue until every plan it wrote is built. Design a product catalogue
• /next keeps the same status table for the whole run, answers with it whenever you ask for the status, and shows a plan rebuilt for a correction as building again, with its attempt number. Run the queue with /next

The standards

• A new Queues and Time Budget standard sets a five-second ceiling on any piece of work and sends heavier work to the queue; /plan-prompt and /develop hold the work they plan and build to it. The analyzer does not check it yet. Read the standards

The analyzer

• The analyzer checks that buttons showing only an icon have a name a screen reader can read, that visible text in a view goes through translation, that plurals come from the translation files, that a grid has a version for small screens, and that a migration whose data work grows with the data runs it on the queue. The analyzer

Corrections

• Several fixes were made to the help centre.
• Several fixes were made elsewhere in the product.

The platform

• The MCP page now also lists, under a heading of their own, the seven tools your assistant uses to read and save your project documents. Get your access token
• Revoking a token now asks in a dialog that says what that kind of token stops doing: the assistants connected with an MCP token, or the pages reading your documents with an API token. Keep your tokens safe
• In a project's viewer, uploading any file whose name already exists in the folder now asks whether to replace it, at the project's root too. Your projects and their documents
• Several fixes were made to the platform.

Commands and agents

• /next builds every plan in its own git worktree, even when the queue holds only one, and reads the queue again after every event, so a plan saved or unblocked while it runs is taken up by the same run. It never stops to ask: a question only you can answer is written into the plan it concerns while the rest of the queue goes on. Its status table shows at the start, after every event and at the close. Run the queue with /next
• When /test or a design review finds a failure, the correction is now written into the plan's own Validations section instead of a new document, and the queue builds it from there within the same three attempts. Verify a plan with /test
• /design-catalog no longer builds your catalogue itself: it reads your product, proposes the components and, once you confirm, writes plans by area for /next, which builds them and has them reviewed against your catalogue and identity manual. Design a product catalogue
• /finish now also keeps the features page of your project's landing true to what was built. What each command does

Help centre

• A new guide explains the statistics dashboard and how to read it. The statistics dashboard
• New guides explain the space your files take and your media library. Your space · Your media library
• /vision and /improve-toolkit have guides of their own. Write the product vision · Report a wrong finding
• Several fixes were made to the help centre.

Corrections

• Several fixes were made to the website.
• Several fixes were made elsewhere in the product.

The platform

• Your projects now keep their documents in the toolkit: the plans, the vision and the pending findings that the commands write are saved in the project's documents instead of in your repository. A run of the analyzer that names a project creates that project if it did not exist yet, and the platform page lists your projects. Your projects and their documents
• In a project's viewer you can now edit a document, replace a document by uploading one with the same name, and drop files on the upload box to add them to the folder you are looking at. Your projects and their documents
• The dashboard's Project filter can now show No project or All projects. Analyse a file
• Another application of yours can now read your project documents and show them on its own pages, with a read-only API token. Show your documents in another project
• Your tokens are now either MCP tokens, for connecting an assistant, or API tokens, for reading your documents from another application. An MCP token no longer works on the documents API. The tokens you already had were converted, so your assistants stay connected. Keep your tokens safe
• Several fixes were made to the platform.

Commands and agents

• Each project can have its own internal commands: write one with /make-command and run it with /call. /cloud-doc reads and keeps saving working documents, such as a research or notes, in the project's documents. Write and run an internal command
• /finish no longer asks what to finish: it covers everything that arrived since the last changelog entry. What each command does

The website

• The features page shows each feature with screenshots of the product, and every guide can be reached from a feature.
• The AI Toolkit now has its own terms and conditions and privacy policy, describing what it stores.
• Every public page has its own address in each language, so a link you share opens in the language it was written in.
• On screens narrower than a laptop, the top bar's links move into the menu.

Corrections

• Several fixes were made elsewhere in the product.

Commands and agents

• /next builds the plans that touch different files in parallel, each in its own git worktree, and verifies them in the browser one batch at a time. A question that comes up is written into the plan instead of stopping the run. A new guide walks through the whole queue. Run the queue with /next

Help centre

• A new guide explains what an earlier run noticed and left for later, and where to find it. Pending findings
• A new guide explains what happens when a defect belongs to the repository your project descends from, and how your work carries on meanwhile. Send a defect to the repository that owns it
• The list of commands now shows which agents each command hands its work to. What each command does
• Several fixes were made to the help centre.

Corrections

• Several fixes were made elsewhere in the product.

Commands and agents

• /design-catalog builds and keeps your product's own catalogue of composed components, so screens of the same product stop drifting apart. What each command does
• A defect in code your product merged in from the repository it descends from is sent to that repository and fixed there, instead of being patched in your copy. Send a defect to the repository that owns it

The MCP server

• Connect Claude Code, Cursor, VS Code or Gemini CLI from the MCP page, each with its own configuration, and the page says which clients install the agents and commands. Install the toolkit

Commands and agents

• /develop writes the documentation a visible change owes, its feature item and its guide when the flow needs steps, in the same run that built the change. How a task travels
• What a command noticed but was not asked to fix is kept in your project's pending findings file, sorted by priority, instead of being lost with the session. Pending findings

Commands and agents

• A request that decides nothing goes straight to the prompt, without the full planning investigation, and a change to how something looks keeps that direct route even when it reaches shared parts. Which route your change takes

The platform

• The help centre and the feature pages are open: guides in English and Spanish on connecting your assistant, analysing a file, the standards and the commands. What the AI Toolkit is

Commands and agents

• A plan can say which other plan it waits for, and /next runs it only once that one is built. A queue no longer stops at a blocker: it writes it down, carries on, and asks its questions at the end. Run the queue with /next
• When /test finds a failure inside a queue, /next builds the correction and verifies it again, up to three attempts in all. Run the queue with /next

Commands and agents

• /vision writes down your project's product decisions. On its first run it also corrects developer documentation that no longer matches the code, and moves an analysis whose feature was built to the status that says so. What each command does
• Every plan /plan-prompt makes is saved, and /next works through the ones marked ready to run. Run the queue with /next
• /plan-prompt notices a decision your project keeps settling the same way and proposes it for the vision. Planning before building
• Before editing existing code, /develop notices when the edit would undo a change somebody committed on purpose, and stops to ask. How a task travels