Release notes
What changed in each release, newest first.
- 0.1.155 — Verify on legacy app now records Proof for the legacy desktop app (its screens, SQL, web service calls and data changes), shown beside the new app's. Storybook projects can follow a design system kept in its own repository, and a document regenerated after its template changed now follows the new template.
- 0.1.154 — Each scenario now has a Proof button that shows why it works — the screens, the data it changed and everything the app did along the way — so you can accept a feature without reading its code. Prototypes are checked against the legacy app's screens, and the Domain diagram stays readable on large models.
- 0.1.153 — Agent questions and requests now reach you in Slack, testers can report issues straight from your staging site, and every bug report gets a ticket in your issue tracker, kept current both ways. You can install Claude Code plugins for agents and developers, and run the legacy app yourself in the browser and in the Repave IDE.
- 0.1.152 — 0.1.150 and 0.1.151 were not published. Upgrading from 0.1.149 to 0.1.152 brings everything in both: read their notes for what changed.
- 0.1.151 — Larger test runs can be submitted: the upload limit is now 5 GiB by default and set per project. Lines of code now include Delphi/Pascal and SQL scripts saved as UTF-16.
- 0.1.150
- 0.1.149 — The top-bar search now finds modules and individual settings, and you can correct a detected unit test suite's commands in Settings. Agents that change your code also run the static checks your CI runs.
- 0.1.148 — Stopping a module's implementation now keeps the features it finished, and code review also folds new code that repeats what your codebase already has. Repave IDE sessions on a module ask which feature you mean, and desktop notifications reach only the people a job concerns.
- 0.1.147 — Every open branch now takes new commits from main, even while an agent is working in it. When you say Yes to an agent's scenario change, it is written straight away, and agents can now ask to remove a scenario. Modules can be implemented again from the dialog, and you can check a page for accessibility problems from Repave IDE.
- 0.1.146 — A feature, module or worktree can now build on work that has not merged yet, and Repave reads every check on its pull requests and fixes what fails. Repave learning gets its own switch and a Learning page, the home page shows how your projects connect and what a change would affect, and the agent log shows time and can be filtered. Custom workflows and AWS AI-DLC arrive, both off until you turn them on. The test step after you try a feature is now called Write Tests.
- 0.1.145 — Verify on legacy app now lets the agent choose which scenarios to drive, takes your instructions, and keeps the legacy app's screens per feature. A feature or module with open warnings has a Warnings tab and can Continue fixing. Implementing a module brings every feature's work onto its branch, and the module page shows what its branch and features need. Every enabled validation rule is now checked, with no Applies when condition.
- 0.1.144 — An agent now writes the title and description of every pull request you open, and Repave can read SonarCloud's quality gate on its pull requests and fix what fails. A feature whose checks still found something when their review rounds ran out says so until someone resolves it. Daily test runs test what is merged in your repository, and each fix works on a branch you can pick up. Agents drive the legacy app without stopping for approvals, and a project's keys and tokens no longer leave Project Settings, in the browser or in a backup.
- 0.1.143 — Features can now be linked to Jira tickets, open feature branches keep up with your base branch on their own, and the Repave IDE can run your project's own test stack the way agents do. The automatic accessibility checks are removed; the preview's Accessibility button stays.
- 0.1.142 — Staging replaces Bundles: stage features and modules together on one branch and run them side by side before anything is delivered. Discovery now tells your organisation's own systems apart from third parties' and links them to the Repave projects that own their code, and every implementation ends with a code review.
- 0.1.141 — Repave can now run your legacy Windows desktop app and have its agents check scenarios against it. Maintenance projects start from your repository's own feature files and follow its own agent instructions, and a change can wait for a design doc you approve before any code is written.
- 0.1.140 — Repave now maintains an existing app or builds a new one from your description, as well as modernizing a legacy one. It also answers reviewers' comments on its pull requests, connects to Gitea and Forgejo, and its project assistant can do most of what you do in the portal.
- 0.1.139 — Repave now learns from the changes your developers make to delivered features, connects to GitLab, creates projects on a page of their own, and can run your whole test suite every day with an agent fixing what fails.
- 0.1.138 — A project assistant you can chat with about the whole project, a record of what every agent run was given and changed, project documents written in your organization's own template, and a daily email digest of what is waiting on you.
- 0.1.137 — Every job now shows who started it, Repave IDE sessions stop once a day at a time you choose, and Skills have their own page that every project member can read.
- 0.1.136 — Each project can now read, edit, add and version the skills its agents and Repave IDE sessions follow, from a new Skills tab in Project Settings.
- 0.1.135 — Prototypes can now be built in Storybook from your own components, and agents can ask you questions, wait for your approval and read your messages while they run. Label keys can be described for agents, and new projects' Speed tier runs Claude Haiku 5.5.
- 0.1.134 — A prototype change is now committed by the agent that made it, so files that are not part of the feature's work no longer end up on the feature branch.
- 0.1.133 — Agent costs now include failed and cancelled runs and the work of subagents, projects that merge through pull requests no longer have Repave write configuration to the base branch, Repave IDE can push its own branch, and the Repave CLI stays signed in.
- 0.1.132 — Agent costs for resumed runs are corrected, merged features show where their work really is, project settings and the agent log each lose a control that repeated another, and GitHub sync results are easier to read.
- 0.1.131 — A fix for builds that failed and lost their work while an implementation preview was already running, and a quieter Implementation pane.
- 0.1.130 — Budgets now cover every agent, not only implementation, and preflight asks a feature's questions once. Repave IDE gains worktrees for changes that are not a feature, and the MCP server can say who spent what and how many rounds a feature took.
- 0.1.129 — A feature's target-stack prototype is now the start of its own branch, and a new project builds its application shell before its features. Administrators can cap what an implementation run spends, and you can see the memory each Repave IDE session and agent run is using.
- 0.1.128 — You can now organise features with tags and labels, bring every open feature branch up to date with its base in one step, and see what your project's agents remember on a new Memory page.
- 0.1.127 — Repave now has an admin portal, where an administrator manages users, organization-wide API keys, email and the license. You can add test data to a running implementation or Target Stack preview, the To-Be suggests scenarios it may have missed, and agents interrupted by a busy provider pick up where they stopped instead of starting over.
- 0.1.126 — You can now run, scan, adjust and revise a feature's implementation from its Preview tab, the same way you already review its prototype. Agent runs and Repave IDE sessions now share one runtime image, so there is one airgap bundle to download and no toolchain setting to get right. Repave can also be connected to Claude and other MCP clients, and agents can use Jev for small judgments.
- 0.1.125 — You can now scan any screen of the target-stack preview for accessibility problems, pick the ones worth fixing, and send them straight to the agent. Web IDE sessions are also handed the feature's prototype as a reference, and new projects start their Elite tier on Claude Opus 5.5.
- 0.1.124 — The target-stack preview no longer shows your app through a rewriting proxy. It now runs a real browser on the server, beside the app, and streams the picture to you — so the screens you review are the screens the app actually serves.
- 0.1.123 — Agent runs work again. On 0.1.122 every agent run failed immediately with a model error, and this release is the fix.