v0.9.6

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

App

• Dropping a card anywhere in a column now works. The drop target used to be only the strip of cards at the top of the column: releasing a card over the empty space below them - including, in an empty column, directly on the "Drop a card here" hint - silently did nothing. The whole column now accepts the drop, matching the highlight shown while dragging.

v0.9.5

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

App

• Cards flag unread discussion. A card with a comment you have not seen yet, posted by someone else, now shows a dot on the board, in lists and in My Tasks; it clears the moment you open the card. New comments light the dot as they arrive, without a reload.
• Notifications arrive live. The bell count updates and a toast appears the moment you are mentioned, assigned, or replied to, instead of on a timer. If the live connection drops it still catches up within the old polling interval.
• A card's whole discussion stays in the loop. A new comment now notifies everyone who has commented on the card, not only its assignee, so a conversation you joined keeps reaching you even when the card was never assigned to you.

v0.9.4

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

App

• The board and your lists keep themselves up to date. A card someone else creates, moves, edits or deletes - or one changed from the CLI or by an agent - now appears on your board or list as it happens, with no reload. Cards slide between columns, tags and comment and attachment counts change in place, and a card that moves off a board leaves it, all without the view flickering or losing your scroll position. Changes you cannot express as a single card - a renamed status, a new column, a filter - refresh the view quietly. If the connection drops, the view catches up the moment it reconnects.

v0.9.3

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

CLI

• Tags and comments are now manageable from the CLI, not just visible on a card. completo tag <card> add|remove|set <names...> changes a card's tags — add and set create a tag the project doesn't have yet, which needs the project owner role — and completo tags lists a project's existing tag names, because tag names are matched case-sensitively and Bug on a project that spells it BUG would otherwise quietly become a second tag. The CLI now says so when it is about to make one. completo comment <card> add|edit|delete manages comments, with --file for a multi-line markdown body; completo comment <card> lists them with the short IDs that edit and delete take. Both commands print the card's resulting list after every change and support --json.

v0.9.2

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

App

• A card you create on a filtered board or list now says so before you save it. Creating a card that doesn't match the view's filters used to save it and then simply not draw it — it was in My Tasks and in search, missing from the board that made it, with nothing said and the header's card count unmoved. The panel now names what the card is short of, in the properties row it is about — "This board's filters need a tag of Website" — and each line clears itself as you set it. Creating the card anyway still works and always will: a card belongs to the project, not to the view you happened to be standing on.

v0.9.1

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

App

• Boards with a lot of cards appear about six times faster. A board of ninety cards spent roughly three quarters of a second drawing itself after its data had arrived, and held the page still while it did. Every card was building the pickers behind its four field controls up front, for menus nobody had opened yet; they are now built the moment you reach for a card, which is the first moment any of them can be used. Nothing on a card looks or behaves differently.

v0.9.0

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

Upgrading

A straight upgrade — the migrations run themselves and nothing needs a decision first. Two notes for anyone driving Completo over the API:

• A board's card display moved from two booleans to one list. showDescription and showTags are replaced by hiddenCardFields, an array naming the fields a board hides, since a card now has eight fields it can show rather than two. Migration 0009 folds the old values in before 0010 drops the columns, so existing boards keep what they were set to.
• DELETE /api/statuses/{id} refuses a status that holds cards. It answers 409 unless the request names moveToStatusId, and it refuses a project's last status outright. It used to delete the cards along with the status.

App

• A new interface. Completo has been rebuilt on one design system. It was shipping two brand colours and two greys at once, which showed in dark mode as a visible seam between the sidebar and the content; colours, type sizes, corner radii and shadows now come from one place, so light and dark stay in step. Every card answers the pointer the same way, every menu behind status, priority, assignee and tags is the same menu, and secondary actions like Cancel and Close now rest a step behind the action they sit beside. Focus marks where you are typing — the field you are in, with the edge it already has — instead of drawing a box around everything you click.
• Every tag, status and priority colour is legible, whichever one you pick. A dark tag colour used to sit on a faint wash of itself and effectively disappear. Completo now keeps the hue you chose and decides the lightness itself, so a label is readable on all sixteen offered colours in both themes. People without a profile picture get a stable colour derived from their name instead of a grey disc, the same one everywhere they appear.
• Cards open in a side panel. Clicking a card slides a panel in from the right instead of covering the board, and the board brings that card's column alongside it — so reading a card and then moving it never means closing anything. The description and comments are visible the moment it opens, status, assignee, priority, due date and tags sit on one line under the title, and a long title wraps instead of running off the edge. On a new card, typing a title and pressing Enter creates it. The card's own page shows the same controls in the same order.
• Walk your cards from inside a card. With a card open, the panel's top corner steps to the next one: on a board, up and down through the column and left and right to the columns either side, with the board sliding across to follow; in a list and on My Tasks, up and down through the order you sorted the rows into. Between the steppers it says where you are — 2/7. Arrow keys do the same without the mouse. My Tasks steps across projects, picking up each card's own statuses, people and tags as you go, so working through your queue is one pass.
• A board card reads as a title. The title comes first and is capped at two lines, tags are pills with a count of any that don't fit, and the ticket ID sits on a quiet footer line. Priority is a thin coloured edge, and only High and Urgent get one, so an urgent card stands out. A card counts its comments as well as its attachments, and shows either only when there is something to count.
• You can set every card field from the board. Tags, priority, due date and assignee are four controls in the same four places on every card, quiet until you reach for one — before, a field only appeared once it already had a value, so the one card that needed a due date was the one card that couldn't be given one without opening it. "Quick add" at the foot of a column takes a card by typing and stays open so you can enter a backlog in one go; the + in the column header opens the full dialog with that column already chosen. Dragging a card lights up the column it will land in.
• Boards choose what their cards show. View settings → Display lists the eight things a card can display — description, tags, ticket ID, comment and attachment counts, assignee, due date, priority — and any of them can be switched off. Hiding a field never takes the field away: it is still there to set when you hover the card.
• Nothing you type is lost. Descriptions and comments in progress are saved locally as you write, so closing a card by accident, reloading or losing the tab no longer costs the text — reopening puts you back in the editor and one click discards it. A draft started in the side panel is waiting on the card's own page. Because nothing can be lost by closing, the panel no longer stops you on the way out to ask.
• Editing a card field is instant, everywhere. Changing a priority, status, assignee, due date or tag used to reload the whole board or list; it now updates immediately and reverts with an explanation if the server refuses. Titles save as you finish typing. The description is the one field with a Save of its own, and it now sits directly beneath the editor rather than at the far end of the panel.
• Press ⌘K (or Ctrl+K) to search. Find a card by title, description or ticket ID across every project you are in, jump to a project, or run a command. Typing TK-42, tk-42 or just 42 goes straight to that card. Results mark the words you typed and preview the selected card beside the list, with the matching part of its description. The sidebar groups projects and admin pages under headings and can be resized.
• A comment thread reads as a conversation. Each comment sits in its own bounded record with the author's face and name above what they said, so four comments no longer run together as eight loose lines. Hovering a timestamp gives the exact date and time behind "2d ago". Code blocks in comments and descriptions have a copy button and say which language they are in, and text no longer changes size mid-sentence around a mention or a snippet.
• A file dropped anywhere on a card is attached to it. Files could only be dropped on the attachments box — drop one on the description or the comments and the browser opened the file itself, taking you off the page. Removing an attachment now asks first, too; it was the only thing on a card that could be destroyed without a question.
• A card with nothing on it looks like one. An empty card used to open on three headings and a full markdown editor sitting open for a comment nobody was writing. It is now three lines that each say what goes there and are each the button that puts it there.
• Deleting a status no longer deletes its cards. A status holding cards took every one of them with it, permanently, on a single click. Completo now asks which status they should move to, says how many are moving, and moves them. An empty status still goes on a second click, and a project's last status stays. Over the API, DELETE /api/statuses/{id} refuses a status with cards in it unless you name where they go.
• Fixed: an API token could revoke itself five seconds after you asked about it. Arming the delete confirmation on a token in Profile → API tokens set a five-second timer that completed the deletion instead of putting the question away — so arming it and then taking a call destroyed a live credential with no second click and nothing said. The timer now disarms.
• A confirmation says what will happen, not "are you sure". Deleting a board says its columns are unlinked and every card and status stays on the project; deleting a list says the same about the cards; removing someone from a project says they are notified and that cards assigned to them keep their name; deleting an AI skill says descriptions it has already written are untouched.
• List views lead with the title, and can count discussion. New lists and My Tasks put Title first and the ticket ID last, and stop printing "Medium" and an em-dash in every unset cell — empty cells stay blank until you hover the row, where the field is still there to click. Comment and attachment counts are available as sortable columns, so "what is being argued about" is a click on a header. Fixed: with enough columns showing, the Title column could collapse to a few pixels and every row rendered blank; the table now scrolls sideways instead.
• My Tasks is a view like any other. It can change a card's status and tags, which it couldn't before — the one screen built for working through your own queue was the one screen where you could not move a card to In Progress. Clicking a card opens it in the side panel instead of navigating away, and its settings open from the same Settings gear a board and a list use.
• View settings is tabbed, and applies as you set it. Columns and filters are separate tabs, each showing a count, instead of one long scroll. Everything applies the moment you set it — there is no Save and nothing asks you to discard on the way out, because the same dialog used to apply half its changes immediately and hold the other half behind Save. The dialog now says whether it is configuring a Board, a List or My Tasks, and whether everyone sees these settings or only you.
• User management is a table. Name, email, role, status and last seen are columns, so finding the one suspended account among eighteen is a glance rather than a rea...

v0.8.0

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

Upgrading

Two things to check before you deploy this one. Everything else is a straight upgrade.

• Serving Completo over plain HTTP from an address other than localhost? Set NUXT_SESSION_COOKIE_SECURE=false, or nobody will be able to sign in: the session cookie is now marked Secure, and browsers silently discard those on an insecure origin. localhost needs no change, and neither does HTTPS terminated at a reverse proxy — which covers most installs.
• Migration 0004 stops the deploy if two accounts differ only in the capitalisation of their email. It lowercases every address and adds a case-insensitive unique index, so a genuine pair — foo@example.com and Foo@example.com — has to be resolved by hand first, since only you can decide which identity keeps its cards and memberships. Check with SELECT lower(email), count(*) FROM users GROUP BY lower(email) HAVING count(*) > 1. No rows means nothing to do.

API clients: mimeType on an attachment is now derived from the file extension rather than echoing the uploader's declared type, and attachment downloads send Content-Disposition: attachment for everything except images, PDF and plain text.

App

• Cards now show who created them. The card detail page has a "Created by" line next to the timestamps, the card modal credits the author next to the ticket ID, and list views (including My Tasks) can show a Creator column, sortable like Assignee. Creator is set once when the card is created and can't be changed. Cards whose creator has since been deleted show "Unknown". Existing cards already show the right person — the creator has been recorded since the first release, just never displayed, so no backfill is needed.

• Fixed: sorting a list by Due Date wouldn't stick. Clicking the Due Date header sorted the rows and then failed with "Invalid sort field" — the column was sortable in the table but the server refused to remember it, so the order reset on reload. Due Date now saves like every other sortable column.

• Fixed: the Done checkbox column couldn't be picked when creating a list. It was offered in a list's settings but missing from the create-view step, so you had to make the list first and add the column afterwards. Both pickers now offer the same set of columns.

• Fixed: pressing Esc in a card editor closed the whole card and threw away what you'd typed. It affected the comment box, a comment you were editing, and the card description — the keystroke was meant to be handled locally (cancel the edit, close the description editor) and it was, but it also travelled on and dismissed the card modal behind it. Esc now stops at the editor it was pressed in: it dismisses the @mention list if one is open, otherwise cancels a comment edit or closes the description editor, and does nothing at all when you're partway through a new comment. Closing the card another way — clicking outside, or the close button — now asks first when an editor is holding text, naming what's at risk, and hands focus back to where you left off if you choose to keep editing.

• Fixed: using an AI skill on a comment, then pressing Cmd+Enter, closed the card and threw the comment away. The shortcut is routed by which editor has focus, and the AI round-trip lost it: the skill menu took focus, then closed onto a button that had meanwhile been replaced by "Stop", leaving focus nowhere — so Cmd+Enter fell through to the card modal, which saved the card, closed, and took the unposted comment with it. Focus now stays inside the editor throughout: on the "Keep" button while you review the suggestion, and back in the text box once you keep or discard it. Cmd+Enter straight from the review state posts the comment as it stands.

• Fixed: the icon list in project settings couldn't be scrolled, and opening it put Save out of reach. Typing in the icon search matches against all ~1,770 Lucide names, so the grid grew to well over a hundred rows. In the edit dialog that pushed the bottom of the form — Key, Slug, Done status, and the Save, Cancel and Delete buttons — past the dialog's edge with nothing to scroll, so the only way out was to close the picker again. The icon grid now scrolls inside a fixed height, and the project form scrolls whenever it outgrows the dialog.

• Fixed: an invited user who signed up on the register page instead of using their invitation link was left stranded. For someone an admin had created an account for, registering returned "Account created. Please check your email" and then did nothing at all — no account, no email, and no way in, since their password had never been set. Registering with that address now re-sends their setup link instead. The response is deliberately unchanged, so it still reveals nothing about whether an address already has an account.

• Fixed: signing in and signing up now ignore capitalisation in email addresses. Someone@example.com and someone@example.com were treated as two different people: registering the second form when the first already existed created a second account, and each could then only be reached by typing that exact capitalisation. Password reset always lowercased, so a mixed-case account could never reset its password either. Addresses are now stored lowercased, matched case-insensitively everywhere, and the database enforces it.

• Fixed: being invited and then signing in with Google, GitHub or Microsoft left the account marked "pending setup" forever. The sign-in worked and the person could use Completo normally, but user management still listed them as never having set up, and password reset stayed blocked for them. Claiming an invitation through a social provider now completes the account.

• The "restricted to approved email domains" message now tells invited people what to do. Someone invited from a domain that isn't on the allowlist hits the same wall as an uninvited stranger, with no hint that their invitation email is the way in. The message now says so — to everyone who sees it, since wording it based on whether that address has an invitation would let anyone probe who has been invited.

• Fixed: password recovery couldn't be completed in the browser. "Forgot password?" and every emailed reset link redirected to the sign-in page — and the link's token was discarded on the way — so anyone locked out stayed locked out. Both pages were missing from the list of routes that load without a session. They also stay reachable while signed in, which is the only way back for someone who is still logged in somewhere but has forgotten the password they'd need to change it.

• Security: an emailed link can no longer be redeemed for something it wasn't issued for. Verification, password-reset and account-setup links all drew on one pool of tokens, and each page accepted any of them — so a "confirm your email address" link could be used to set a new password on that account, and the reply signed the sender straight in. Anyone who saw one such message (a forwarded mail, a shared or team inbox) could take the account over without knowing its password. Each link is now valid only for the flow that sent it. Setup links additionally refuse accounts that have already been set up, where they could previously overwrite a live user's password and display name.

• Security: an uploaded file can no longer run as part of Completo. Downloads echoed back the content type the uploading browser claimed, and showed the file inline, so a project member could upload an HTML file — or an SVG, or any name ending in .md while declaring itself HTML — and anyone who opened it ran the uploader's script inside the app, with that person's access to every project they can see. Downloads now determine the type from the file's extension, and only show images, PDFs and plain text inline; anything else downloads. Image previews on cards are unaffected.

• Security: the session cookie is now marked Secure. It was explicitly configured not to be, overriding a safe framework default, so every HTTPS install sent its session cookie over any plain-HTTP request that reached the same host — a stray http:// link, a misconfigured proxy or a redirect was enough to leak it. If you reach Completo over plain HTTP from anything other than localhost — a home server at http://192.168.1.50:3000, say — set NUXT_SESSION_COOKIE_SECURE=false, or signing in will appear to work and drop you back at the login page. localhost needs no change (browsers treat it as a secure origin), and neither does TLS terminated at a reverse proxy. Existing sessions stay valid.

• Fixed: adding a member by email failed on capitalisation. Typing Someone@Example.com for an existing user created a pending invitation instead of adding them — and that invitation could never be accepted, because invitations are only picked up while registering. The address is now matched regardless of case or surrounding spaces.

Dev

• Migration 0004 lowercases existing email addresses and adds a case-insensitive unique index (users_email_lower_unique). users.email was UNIQUE without COLLATE NOCASE, so normalising in code alone would leave the next endpoint free to reintroduce the split. The UPDATE runs against the old case-sensitive constraint on purpose: on an install that ...

v0.7.0

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

App

• Comments on cards. Every card now has a comment thread below the description, on both the card detail page and the board's card modal. Comments support markdown and @mentions using the same editor as descriptions. Mentioned project members get an in-app notification, and the card's assignee is notified of new comments — never for their own. You can edit and delete your own comments; edited ones are marked as such. Editing only notifies people newly mentioned, so fixing a typo doesn't ping everyone again.
• AI assistance for comments. The comment editor now offers its own AI skills instead of the card-description ones, which made no sense there. Two ship by default: Fix Spelling & Grammar, which corrects mistakes without rephrasing anything, and Improve Clarity, which rewrites for readability using the card and the previous comments as context. Both preserve @mentions exactly, so improving a comment never breaks a notification. Admins can add their own comment skills under AI Skills. Requires AI_PROVIDER to be configured; with it unset the AI button stays hidden as before.
• Cmd+Enter now follows what you're editing. It used to always save the card. Pressed inside the comment box it posts the comment, inside a comment you're editing it saves that comment, and anywhere else on the card it still saves the card.
• Fixed: @mentions could notify the wrong person. Mentions were matched by display name, which isn't unique, so with two similarly named members an arbitrary one was picked — and renaming a user broke their mentions. Mentions now carry the user's ID. Mentions written before this change are no longer linked and show as plain text; re-add them to restore the link.
• Security: dependency updates clearing all critical and high advisories (3 critical / 19 high → 0). Nuxt 4.4.6 → 4.5.1 (server-side template-injection RCE in island props, cross-user SSR payload-cache disclosure, route-rule middleware bypass), Nodemailer 8 → 9 (a raw message option bypassed disableFileAccess/disableUrlAccess), @nuxt/devtools → 3.4.0 (unauthenticated RPC allowing command execution on the developer's host), @nuxt/ui 4.7.1 → 4.10.0, DOMPurify 3.4.5 → 3.4.12, plus transitive bumps to tar, shell-quote, ws, vite, postcss, js-yaml, svgo, and brace-expansion. Vite moved 7 → 8 as part of the Nuxt upgrade. No pnpm.overrides were needed — the parent version ranges already allowed the patched releases.
• Security: env.sample no longer ships a working NUXT_SESSION_PASSWORD. It contained a real 44-character key, so any install that copied the file verbatim ran on a cookie-signing key published in the repo. It is now a deliberately too-short placeholder (replace-me), which fails fast with "Password string too short (min 32 characters required)" instead of silently working. Existing installs that copied that value should rotate it (openssl rand -base64 32); doing so invalidates current sessions.

Dev

• Fixed: default AI skills could never reach an existing installation. db-seed.ts guarded its whole default-skills block with "does this table have any rows", so once an install had any skill, newly added defaults were skipped forever — even on a re-run. The guard is now per skill, and product defaults ship via a migration (0003) rather than the seed, since db:migrate is the only step that runs exactly once on every install. Without this, adding the comment skills would also have made a fresh pnpm setup silently skip the two card skills.
• Integration tests can now run inside the dev container. The test harness freed its port with lsof -ti:PORT | xargs kill -9, which works on macOS but not on Alpine, where lsof is a BusyBox symlink that ignores -t/-i and prints every open file — so xargs kill -9 killed PID 1 and the test runner, and the suite died with SIGKILL before running a single test. It now resolves the previous server via /proc and signals only that process.
• New tests/unit/dependency-singletons.test.ts asserts that Vue's runtime packages and vue-router each resolve to exactly one version in the lockfile. Two copies of Vue break the app at runtime while every API test still passes — which is exactly what the Nuxt 4.5.1 upgrade did, undetected by 481 green tests.
• Removed the unused @nuxt/test-utils devDependency. Nothing imported it (the e2e harness is hand-rolled), and it declares vue as a hard dependency rather than a peer, which is what pinned the duplicate Vue copy.
• The dev environment no longer uses a .env file. Dev secrets (NUXT_SESSION_PASSWORD, the GitHub/Google/Microsoft OAuth client IDs and secrets, AI keys) now come from a 1Password Environment, attached to the app service with op-env: in .zdev/config.yaml. Only the Environment ID is committed — it isn't secret, and values are fetched by the op CLI when a container is created. Requires the beta 1Password CLI (brew install 1password-cli@beta) with the desktop-app integration enabled.
  • After rotating or adding variables in 1Password, run zdev update --refresh-secrets. A plain zdev restart or zdev update will not pick them up, since the env is baked in at container creation.
  • Non-secret dev values (SMTP_HOST, APP_URL, DATABASE_URL, the seeded dev logins) stay as explicit environment: entries, which always win over injected variables.
  • Host-side pnpm dev / pnpm setup no longer receive these vars automatically — use op run or export them.
• env.sample still applies to Docker and manual installs — only the zdev container stopped using a .env file.

v0.6.7

Quick start

docker run -p 3000:3000 \
  -e NUXT_SESSION_PASSWORD=$(openssl rand -base64 32) \
  -e ADMIN_USER_EMAIL=admin@yourdomain.com \
  -e ADMIN_USER_PASSWORD=change-this-password \
  -v completo-data:/data \
  ghcr.io/scalecommerce-dev/completo:latest

Open localhost:3000 and log in with the admin email/password you set above. Demo data included.

CLI for AI agents

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/scalecommerce-dev/completo/main/install.sh | sh

Windows: Download completo-windows-amd64.exe from the assets below.

---

App

• Docker: zpinit now supervises the server instead of exec'ing it. The image sets no CMD, which puts zpinit in supervise mode — it stays PID 1, so it reaps zombies for the container's whole lifetime and restarts the Nuxt server with capped-exponential backoff (1s→30s) if it crashes. Previously zpinit exec'd the server and exited, leaving no reaper and no in-container restarts. A readiness probe backs zpctl ready, and zpctl status/restart/tail now work (they were inert before).
  • Note: after 5 consecutive crashes the service is marked FATAL while the container still reports running. Use zpctl ready as a healthcheck if you need the container to fail loudly.
  • Passing a command to docker run still overrides this and runs that command once, so the ad-hoc CLI script invocations are unchanged.
• Docker: zpinit is pinned to 0.5.5 instead of latest for reproducible builds, and its config is validated at build time (zpinit --check-config) so a malformed service file fails the build rather than the deploy.
• Fix: pnpm setup failed on a fresh clone. scripts/init-admin.sh hardcoded scripts/node_modules/.bin/tsx, which only exists in the Docker image (cd scripts && npm install) and is gitignored otherwise — so the documented first-time bootstrap died at db:init-admin with MODULE_NOT_FOUND. It now runs node scripts/user-create.ts directly, like every other db:* / user:* script.
• Node 24 LTS and pnpm 11.17.0 are now pinned consistently across package.json, the prod image and the dev image. The prod runtime base moves from alpine:3.23 to alpine:3.24 (same Node 24.18.x, same native ABI).
• tsx is gone — scripts run as node scripts/foo.ts. Node strips TypeScript natively (default since 22.18), so the same command now works in dev, in prod, and on the host. Previously the prod entrypoint used a vendored tsx binary while dev used pnpm db:*, so the two paths differed for no good reason. tsx is dropped from scripts/package.json, shrinking the prod image by ~16MB (188MB → 172MB), and the runtime image now contains no package manager and no TS loader at all.
  • New engines.node >= 22.18 in package.json. scripts/*.ts must stay within erasable syntax (no enum, namespace, parameter properties or decorators) and use no extension-less relative imports.
  • Ad-hoc invocations change accordingly: docker exec <container> node scripts/user-create.ts … (was node ./scripts/node_modules/.bin/tsx scripts/user-create.ts …).
• README corrections. The documented SMTP_PORT default was wrong (587 — the code falls back to 1025). The quickstart claimed Node alone was enough, but a fresh Node has no pnpm on PATH, so corepack enable is now an explicit step alongside the Node 22.18 floor. "All commands work with both npm run and pnpm" was true of the individual scripts but not setup, which chains via pnpm. db:cleanup was described as removing "expired sessions and soft-deleted data" — it touches neither; it drops orphaned rows and expired invites/tokens, prunes unused uploads, and VACUUMs.
• README: new "Development environment (zdev)" section documenting the containerised dev setup and its seeded logins.

Dev

• Local dev environment migrated from .scdev/ to .zdev/ (the tool's current name and config format). Use zdev start; run zdev update — not zdev restart — after editing .zdev/config.yaml.
• The dev container now boots via zpinit too, from its own .zdev/Dockerfile. On every start it runs install → migrate → seed, then supervises the Nuxt dev server.
• The dev container is built not to die, so there's always something to debug. zpinit runs in supervise mode (stays PID 1, restarts the dev server with 1s→30s backoff) and entrypoint_on_failure = "continue", so neither a crashed dev server nor a failed pnpm install takes the container down — the error stays in zdev logs and zdev exec app sh keeps working. zpctl is available inside the container (zpctl status, zpctl restart app, zpctl tail -f app).
• Two fixed dev logins are seeded on every boot and shown by zdev info: admin@completo.local / admin1234 (admin) and demo@completo.local / demo1234. Both are created auto-verified. Dev-only — production still provisions from ADMIN_USER_*.
• Dev env wires the shared Mailpit (SMTP_HOST: mail), so invitation and verification mails are catchable via zdev mail. Because isEmailEnabled() keys off SMTP_HOST, dev logins require a verified email. APP_URL points at the routed HTTPS domain so links in those mails resolve.
• The dev SQLite DB moved to /app/data/sqlite.db in a named volume, out of the file sync (WAL files over Mutagen risk corruption). It is now the only dev database — previously, with no DATABASE_URL set, the container fell back to the relative 'sqlite.db' default and silently shared the host's file. *.db* is also in mutagen.ignore so a host-side DB can't sync in and shadow it. Dev data is disposable by design: zdev down -v -f destroys the volume and the next zdev start reseeds from scratch.
• Docs: drizzle-kit push is now documented as prohibited, not a dev shortcut. Every schema change needs a committed migration. Two verified failure modes are recorded in CLAUDE.md: a push-built DB can never be migrated (db:migrate restarts at 0000 and dies on table already exists, which also breaks pnpm setup), and pushing then generating the matching migration still breaks the next migrate with duplicate column name.
• Boot applies committed migrations (node scripts/db-migrate.ts) rather than drizzle-kit push, which can prompt on destructive changes and would hang in the TTY-less boot.
• New zdev migrate command (.zdev/commands/migrate.just): zdev migrate applies pending migrations, plus generate (new SQL from schema changes), seed and cleanup. Each is a thin alias for the same command the container runs at boot. There is deliberately no push alias — see below.
• The dev image no longer installs anything with npm (the global tsx install is gone) — pnpm is the only package manager in it.
• .zdev/local/ is gitignored for per-developer overrides (deep-merged onto the committed config).