Changelog
What shipped in Plansmith, in plain language: new features, improvements, and fixes, newest first.
Improvement
The agent asks before it writes your epic
Your first sentence used to turn straight into a full epic, for you to approve. Now the agent asks a few questions first and writes the epic from your answers. Skip them and you still get the epic, and it says what it assumed.
Improvement
Plansmith asks before it writes a spec
Every spec now starts with questions. The agent asks what it cannot tell from your plan, then builds the spec on your answers. Skip them and you still get the spec, marked below Ready, with its assumptions called out.
Improvement
Prototypes follow a checked procedure
Every prototype now runs the same pass: match your linked app, build it, then check it against a fixed list before saving. Revisions keep your pinned comments in place, and owners can wrap the builder with their own rules.
Fix
Images from an imported Jira issue now show up
Pictures that came over with a Jira issue now render in the spec instead of a broken image, and the planning agent can read them, so a screenshot in the ticket is something you can actually talk about.
Improvement
Importing a Jira issue brings its comments
An imported issue now arrives with its comment thread, including the screenshots people posted there. Most of what a bug report knows lives in the comments, and it used to be left behind.
Fix
Selecting text in a document is just selecting text again
Highlighting a sentence raises a small Comment button instead of opening the box on top of you, so you can copy or rewrite what you selected as normal. Escape closes the box and leaves the document open, and scrolling keeps what you typed.
Feature
Push back to Jira without overwriting your team
Plansmith now checks what changed in your tracker before it writes. Nothing changed there? The push goes straight through, exactly as before. Somebody edited an issue since you imported it? The push stops and the agent walks you through it, one item at a time.

- Fast when it's safe. A push with nothing to resolve does one tracker lookup and then behaves exactly as it did before. No card, no conversation, no waiting.
- It stops before it writes, not after. A conflict is caught ahead of the write, so nothing lands half-done and nothing of theirs is lost.
- Resolve it in the conversation. The agent shows you both versions and what changed since you last agreed, and you decide per item. The result is an ordinary spec edit you can review.
- Status changes are not conflicts. Someone moving your issue to In Progress will not stop your push. Only a real edit to the words does.
Feature
Comment on any passage and send it straight to the agent
Highlight a sentence anywhere in a plan and tell the agent what you want changed. Your note arrives with the passage attached, so you never have to explain which part you meant.

- Select and say it. Highlight text and a small box appears right there. Type your note, press Enter, and it goes to the agent as a request scoped to that passage.
- Everywhere you read. Specs, your epic, your context sources, review files, and the agent's own answers in the chat. One gesture, the same everywhere, so there is nothing to remember about which pane you are in.
- Works on the conversation too. Quote a paragraph of the agent's answer back at it. "This contradicts the last spec" lands on the exact sentence instead of the whole reply.
- The agent knows where you meant. Your highlighted text travels with the note, so "this is too vague" lands on the sentence you were reading rather than the whole document.
- Nothing to clean up. The note becomes a message in your chat and the plan updates in place. There are no comment threads to resolve or inboxes to clear.
Feature
Point at a prototype element and say what to change
Click the exact button, card or heading in your prototype and tell the agent what should be different. The instruction lands in your chat box with the element already identified, so you stop describing which one you mean.

- Click it instead of describing it. Turn on element select over your prototype. Everything you hover is outlined, so you can see what you are about to pick, and the agent receives that element, the screen it is on and the prototype it belongs to.
- Review it before it goes. Your instruction arrives in the chat box, focused and ready, added after anything you were already writing. Nothing is sent until you send it.
- Works on every prototype you have. The plan prototype and any prototype file in the drawer both take it.
Improvement
Point the agent at any spec, epic or file with @
Type
@ in the chat to point the agent at exactly what you mean: a spec, your epic, a context file, or a file in a connected repository. It narrows as you type, and what you picked shows under the box, so you see what the agent will read.Improvement
Jira agent tools, one sign-in
Connect Jira once: the agent’s Jira tools now run on that same connection, acting as you, with no second sign-in. Your profile also gains a My connectors page listing every connection you hold.
Feature
Import your Jira work into Plansmith
Pick a Jira epic or issue and bring it into a plan: issues become specs, sub-tasks nest under them, and images come along. Every imported spec stays linked to its Jira issue, so pushing your refined plan updates the originals, never duplicates.

- Choose what comes over. A preview shows the whole tree before anything is written. Untick what you do not want; nothing changes in Plansmith or in Jira until you confirm.
- Keep the round trip. Specs remember their Jira issues. Refine in Plansmith, push back, and the same issues update in place.
- Tell your team. Optionally mark imported issues in Jira with a link to the plan, removed automatically when you push back.
Improvement
One place to say where a spec belongs
“Move under…” is now “Move to…”, listing everywhere a spec can go: the plan root, under another spec, or out into a sub-plan of its own. Moving one out takes everything nested under it, and tells you which handles change before you confirm.
Improvement
Start a plan without saying a word to the agent
Right-click anywhere in the plan tree, including the blank space and a new plan’s empty panel, to add a spec, write the plan’s epic, or start a sub-plan. Writing the epic yourself is new: until now only the agent could create one.
Improvement
Delete anything in your plan tree
Right-click a spec, an epic or a prototype and remove it, or use the Delete in the spec panel itself. Deleting a spec with work nested under it tells you how many items go with it before you confirm.
Fix
Paste markdown and it keeps its formatting
Paste markdown into a spec, a context document or your agent instructions and its headings, lists, checklists and tables now arrive as formatting, not as raw
# and *. Ctrl/Cmd+Shift+V still pastes it raw.Fix
Confirming an Epic always gives you buttons
When the agent writes an Epic it now always ends with a short question you answer with a button, instead of sometimes leaving you to work out what to type.
Improvement
A better message box: Enter to send, and Stop when you change your mind
The message box is now one card that grows with what you write. Press Enter to send and Shift+Enter for a new line, keep typing while the agent is thinking, and press Stop, or Esc, to end a turn you no longer want.
Improvement
Tell the agent what to change, right from its question
When the agent asks "Is this what you had in mind?", picking "Keep chatting" now opens a box where you write what you would change. Your note becomes your answer, so the agent knows what to fix instead of asking again.
Improvement
Confluence: read and write
Plansmith can now create and update Confluence pages and comments, not just read them. Connecting works again too; attempts were bouncing back to the Connectors page with an error before Atlassian’s approval screen ever appeared.
Fix
Panels close when you click away, and only one opens at a time
Open a Spec, click anywhere else, and it closes, the way the Epic and your context documents already did. Opening a second panel now closes the first instead of stacking it underneath. Click away or press Escape, on any panel.
Improvement
Delete files your agent wrote, straight from the tree
Files your agent writes into your context now have Delete in their menu, in settings and in the planning sidebar, with a confirmation naming what you lose. Long files can go too; before, the biggest ones could not be removed anywhere.
Feature
Prototypes now match your app, whatever it's built with
When you ask for a clickable prototype, Plansmith reads the repos you've linked and matches your real product: its routes, its colours, its typography.

- Beyond Next.js. It now recognises Nuxt, Vue, Angular, Blazor, Razor Pages, ASP.NET, SvelteKit, Astro, Remix, Rails, Django and Laravel, so a prototype looks like your app instead of a generic mockup.
- Every repo that has a screen. Link several repos and it profiles each one with a user interface, then tells you which one it matched your plan against.
- Quiet about the rest. Repos without an interface are skipped and said so plainly. No more "could not be determined" notes about your database repo. And if a repo simply could not be read, it says that instead of claiming you have no screens.
Improvement
See the agent working, then see what it did
While the agent works you now see a live timer and the step it is on. When it finishes, the turn folds into one line with how long it worked and what it read and wrote; click it to see every step.
Feature
One project, all your repos
Plansmith now grounds a project in every repository the work touches, not just one, and a single search reaches across all of them.

- Link as many repos as the work spans. Frontend, API and infra can live in three repositories and still be one project. Mix hosts freely: a GitHub repo and a GitLab repo sit side by side.
- A Code screen that shows your repos. Project settings now lists the repositories this project reads, right there: link one, unlink one, or pick which is primary, without leaving the page.
- Honest when a repo cannot be read. If a repository is too large to list in full, Plansmith says so on the row instead of pretending it saw everything. If a token has expired, it names the repository that fell back rather than showing it as empty.
Feature
Adversarial review: three models attack your spec
Run /adversarial-review on a spec and three independent reviewer models each hunt for gaps, vague criteria, and contradictions. Findings merge into your spec’s review file, and every one names the reviewers that raised it, so you can see at a glance what all three agreed on and what only one of them spotted.

- Same review, harder judges. It follows the readiness review you already use, so findings keep their numbering and history across runs.
- You stay in control. Reviewers only read; your planning agent proposes the fixes and you approve them.
Feature
Choose the AI models that plan your work
Plansmith now runs on three model tiers: a main model for your planning conversations, a worker model for background drafting and verification, and a fast model for summaries. Owners can swap any tier, plus the three reviewer models used by adversarial review, from Settings → Models.

- Sensible defaults. Out of the box you get our strongest model for conversations and faster models where speed matters, so plans get smarter without paying more where it doesn’t help.
- Your account, your models. Pick from Anthropic, OpenAI, Google, and more for any slot. Pricing follows the model you choose.
Improvement
You approve the Epic, then spec one job at a time
Saying yes to the Epic now creates exactly that: the Epic. Plansmith shows how the plan breaks down and writes each Spec when you ask. Say "spec the rest" any time, or ask for one small fix and it skips the Epic entirely.
Improvement
Planning conversations, harder to derail
The planning agent treats your sources, linked repos, and fetched pages as reference material to read, not instructions to follow, so text inside a document does not steer your plan. It also re-confirms a revised Epic before building.
Feature
A new documentation site
Plansmith's docs got a full rebuild at plansmith.co/docs: a guided quickstart, a searchable guide to every surface, and a section for everything you can teach the agent.
- Organized around what you are doing. Start here, Plan the work, Give the agent context, Teach it how your team works, Ship the plan, and Run your workspace. Plus a Developers tab for the CLI, the MCP server and the API, and the Changelog you are reading.
- Search everything. Press Cmd+K on any docs page to jump straight to the answer.
- Everything the agent can be taught, written down. Instructions, Skills and Formats each have a page, and one page that tells you which of the three you actually want.
- Light and dark. The docs match the rest of the site, and now carry a dark mode you can switch to.
- Built for AI agents. Every page has a markdown twin: add
.mdto any docs URL. There is a copy-as-markdown button on each page, and a full-site index at /llms-full.txt.
Improvement
Where Plansmith runs, and what Enterprise includes
Two new docs pages. Deployment options covers Plansmith cloud, reading a Git server that only answers inside your network, and running Plansmith in your own environment. Enterprise features covers what an Enterprise plan carries.
Improvement
Say no to a research proposal
When the planning agent proposes research, you can now turn it down: Cancel sits beside Start research, and the note above the message box names both. If the research already started, the card says that instead of claiming it was cancelled.
Improvement
Your sources are named the way you named them
The activity marks under a planning answer used to show filenames Plansmith invented, like www-gtt-es-f1fe2721.md. They now use the titles you gave your sources, and say when a file belongs to another plan. The command is one click away.
Improvement
Delete a whole folder of context at once
A folder could only be deleted once it was already empty, so tidying up meant removing documents one at a time. Delete now clears the folder and everything in it, after telling you what goes. Emptying a folder no longer makes it vanish.
Improvement
You can read what the agent is doing
The activity marks under a planning answer used to read the same cryptic line however different the work was. They now say what happened in plain words: Read the plan, Searched your Context, Wrote context/regulation.md.
Improvement
Long documents keep their endings
Add a document up to 25MB and all of it is kept. No more "the end was left out". The planning agent can also find its way to a passage deep inside a long document by describing what it means, not just by matching its exact words.
Fix
Long research runs keep their work
A long planning run could stop partway through and take its notes with it, while the chat said your research was saved. Runs now save as they go, so what was written is in Context, and a run that stops says so instead of spinning forever.
Improvement
Adding a file you already added
Drop a file or a folder that is already in Context and Plansmith asks what you want: replace what is there, keep both, or skip. One question for the whole drop, instead of quietly making a second copy of everything.
Improvement
Adding a folder of documents is fast now
Drop a stack of files into Context and they land right away, with a running count as they arrive. The tidying and summarising finish in the background, with a note on any document still being read.
Improvement
Documents open in a snap
Opening a document from your Context or your plan no longer re-reads your whole workspace behind the scenes: it fetches just that document. Viewed documents reopen instantly, and the agent reads your context with the same speedup.
Feature
Deep research
Ask a research question while you plan and get a sourced report in your plan's Context, ready to plan against.

- Ask while you plan. Type
/deep-researchin the planning conversation and the agent proposes a brief of research questions drawn from what you have been discussing. Click any question to sharpen it before it runs. - You approve before anything runs. Nothing starts until you click Start research; the agent cannot start a run on its own. It runs in the background while you keep planning.
- The report lands in Context. A few minutes later a sourced research report arrives in your plan's Context, where the agent reads it like any other source. Every claim carries its citations.
- You pay for what actually ran. A run you cancel partway is billed for the work it did, not the full price of a finished report.
Feature
Build with Plansmith: API, CLI, and MCP
Everything your team does in Plansmith is now scriptable.
- One permission model. Install the CLI with
npm install -g plansmith.co, mint an API key in Settings, or point an agent athttps://mcp.plansmith.co. Your credential reaches exactly the plans you can reach in the app, nothing more. - Docs for humans and agents. Guides live under /docs, and every page has a markdown twin your agent can fetch directly.
Feature
Plan sharing
Plans now work like documents, with an owner and a share dialog.
- Private, shared, or open. Keep a plan private to yourself, share it with specific teammates as viewer or editor, or leave it open to the workspace.
- Shared with me. Plans shared with you appear under "Shared with me", and the share dialog shows exactly who has access.
Feature
Slash menu, images, tables, and diagrams
The Spec and Context editor does much more.
- Insert anything with /. Type
/to insert tables, checklists, and code blocks, edit links with Cmd+K, and render Mermaid diagrams right in the document. - Images in your Specs. Paste or upload images; they stay private to the plan they belong to.
Feature
Connect any MCP server
Bring your own tools, with approval in your hands.
- Your server or the directory. Paste the URL of any hosted MCP server, or connect Confluence and Notion from the new directory with your own account.
- Every tool asks first. A tool asks for your approval the first time it runs, and "always allow" remembers your choice per person, not per workspace.
Feature
Prototypes that look like your product
Colors, fonts, and screen patterns come from your connected repository, so a clickable prototype reads like your product instead of a generic mockup.
- Grounded in your design system. The agent reads your app's real tokens and screens before it draws, so stakeholders review something that looks like the product they know.
Feature
Archive plans
Archive a finished or abandoned plan instead of deleting it.
- 30 days to change your mind. An archived plan leaves your lists with its whole sub-tree intact and can be restored for 30 days.
Feature
Numbered acceptance criteria
Acceptance criteria on every Spec are now numbered: AC1, AC2, and so on.
- Stable handles. Point at a criterion in a review or a conversation and its number never changes. A Spec's readiness now reflects how well its criteria cover the work.
Feature
Folders in Context and Specs
Organize both trees with folders.
- Full file management. Create, rename, drag to reorder, and delete from each row's menu. Drop an entire folder from your desktop and its structure arrives intact.
Feature
Self-hosted git on the Team plan
Ground plans in code that lives on your own servers.
- Three hosts. GitHub Enterprise Server, self-managed GitLab, and Bitbucket Data Center connect on the Team plan.
Improvement
Plans have addresses
Every project, plan, and Spec has a stable, readable link: your workspace at
app.plansmith.co/your-team, plans as PLAN-7, Specs as SPEC-12.Improvement
One account menu
Switch workspaces from the account menu, and read your monthly usage at a glance from the ring around your avatar.
Improvement
Rename plans in place
Rename a plan from the plan switcher or its card menu, no detour through settings.
Improvement
Project cards show status
A project card signals its status in the icon and reveals the details on hover.
Improvement
Add usage mid-cycle
Running low before renewal? Add usage from Billing in two pack sizes that mirror the monthly plans.
Fix
Your text stays yours
Typing in the file or Spec drawer while the agent updated the same item could bring your draft back wrong. Drafts now survive.
Fix
Push is honest about prerequisites
The push button no longer shows before a tracker is connected, and a plan with no Specs explains why it cannot push yet.
Fix
Billing refusals name the fix
When a change is refused for billing reasons, the message now says what to do about it instead of a generic "that change didn't stick".
Fix
Notices stay until read
Notices in Context used to vanish on their own. They now stay until you dismiss them.
Fix
Setup steps answer for this project
The getting-started checklist now reflects the project you are looking at, not the whole workspace.