Cambios
Qué hay de nuevo.
Releases, funciones y actualizaciones de seguridad. Suscríbete vía RSS.
- Release #
v1.8.1 — the app edits a secret the way a password manager does
Adding a secret starts with a plain title — "GitHub (personal)" — and ClauLock derives the identifier Claude will use from it as you type, shown as a small chip you can edit with one click. The credential sits below as rows in the order you fill them: username, password, website, then "Add more" for a 2FA secret, recovery codes or a custom field. The strong-value generator lives inside the password row; the value is still created in the vault's own process and never shown in the window. Tags and notes stay visible; the developer details (purpose, provider, environment, client, project, rotation, view-protection) fold into "More options". Enter saves.
The secret's own page uses the same rows, so what you see is what you created: fields with Copy on every row and Reveal one click further, the username above the password, the access controls (view protection, agent approval, allowed folders, live grants), classification, inventory and projects all laid out the same way, and editing happens in place.
Across the app: the Settings tabs can be walked with the arrow keys; dialogs hand focus back to the button that opened them; the filter popover closes with Escape; every icon-only control has a name for screen readers; rotating the vault passphrase asks before it acts; copying the generated passphrase during setup confirms it and clears the clipboard after twenty seconds; small text in the light theme is darker to meet the contrast guideline; and the activity log's colours follow the theme.
Fixed: the secret page could fail to open when the usage statistics were incomplete.
- Release #
v1.8.0 — a clearer credential prompt, and a lock on who may ask for secrets
When Claude needs a credential it has never had, ClauLock pops a dialog for you to type it into — the model never sees the value. That dialog used to open with a paragraph: the secret's internal name, whatever the agent had written about why it needed it, a reassurance about encryption, and only at the very end the thing it actually wanted. It now leads with the ask — "Enter your Password", "Ingrese su Usuario / correo" — followed by the secret's name, one line of purpose and, once, the note that the value goes straight to the vault. Spanish is used when your system language is Spanish (or set CLSEC_LANG=es for the MCP server). Agents can now attach a one-line hint per field saying where to find the value, shown under the input, instead of burying it in the description.
Also fixed in that dialog: a username, email or URL was typed into a password box, so you could not see what you had written. Only secret values are masked now.
Security, opt-in: the background service already noted in its audit log when a program other than ClauLock's own binaries connected to it. Starting the service with --peer-baseline-mode enforce turns that note into a refusal: such a program keeps only the read-only surface Claude has and cannot unlock the vault, add, delete, resolve or reveal anything. The default is unchanged (audit only); the architecture and threat-model documents describe when to turn it on and how to recover after rebuilding your binaries (run clsec install again).
The output scrubber that redacts secrets from what Claude reads now also recognises a value whose letters were upper- or lower-cased by a tool, and a long value folded across lines at 64, 76 or 80 columns.
In the app: the Approvals, Devices and Browser screens no longer show a spinner forever if their first load fails — they say why and offer Retry. Confirmation dialogs for destructive actions focus Cancel, so a reflexive Enter no longer deletes. While the vault is locked, Tab can no longer reach the page behind the lock screen. Onboarding no longer offers to generate a new passphrase to someone who already has a vault. The app also stops re-rendering every two seconds and pauses its queue polling while locked.
- Seguridad #
v1.7.2 — ClauLock now actually works on Windows
We owe Windows users an apology. Until this release, ClauLock installed on Windows but never worked: the background service that holds your vault turned away every program that talked to it, so the app, the command line and Claude Code could not reach it. Our automated checks only confirmed that the Windows version built, never that it ran. We now test every release on a real Windows 11 machine, as a standard user without administrator rights, before it ships.
On many home PCs the setup also stopped at its first step, because Windows refused the background task ClauLock registers at sign-in. It now registers reliably for any account, and if Windows still refuses it, ClauLock starts the service for the current session instead of leaving you with nothing.
Claude Code's ClauLock protections need Node.js, which a clean Windows machine does not have and Claude Code does not install. The installer now includes it, so they work out of the box. Claude Code on Windows also needs Git for Windows to run commands; ClauLock now tells you when it is missing.
When the vault locks itself after a period of inactivity, the app now offers a one-click Unlock on Windows. Before, it asked for a passphrase you were never shown.
Security: on Windows, an AI agent could have made the command-line tool print a secret by hiding the signs that it was running inside an agent session. That path is now closed, as it already was on macOS and Linux. We have no indication it was used.
Also fixed: the installer no longer overwrites Claude Code's configuration files if it cannot read them, and running the setup twice no longer adds a duplicate entry to them. And while ClauLock had written a configuration file with your secrets for a running program, Claude was refused ordinary commands that merely contained a slash — a text substitution, a code comment. Those run again; commands that search the whole disk are still stopped.
The Windows installer is still not code-signed, so Windows shows a one-time "Windows protected your PC" notice: choose More info, then Run anyway.
- Release #
v1.7.1 — a real Windows installer
Installing on Windows used to leave you with command-line tools and nothing on screen: no app, no Start menu entry, and a list of commands to type before anything worked. The download button now gives you an installer that puts the ClauLock app on your computer, adds it to the Start menu and the desktop, and opens it when it finishes.
The first time it opens, ClauLock does the rest by itself: it starts its background service, creates your vault, keeps the vault's passphrase protected by Windows for your account only, and connects Claude Code. You never open a terminal, and the installer does not ask for administrator rights.
Updating keeps everything in place: the installer stops the background service while it replaces it and starts it again afterwards, so Claude keeps working. Uninstalling removes the app and its background task but never touches your vault, so reinstalling brings your secrets straight back.
The installer is not code-signed yet, so Windows shows a one-time "Windows protected your PC" notice: choose More info, then Run anyway. The MSI package remains available for IT teams that deploy through management tools; it installs the command-line tools only.
- Release #
v1.7.0 — sign in to websites with ClauLock
ClauLock now signs you in to websites. A small key appears in username, password and verification-code fields; pick the login and confirm with Touch ID, and the username, password and live one-time code are filled together. Only logins saved for that exact website are offered on the page, so a look-alike address gets nothing. After you sign in somewhere new, ClauLock offers to save it.
An agent driving your browser can now ask to sign you in. You see the request on the page, approve it with Touch ID, and the agent carries on — it never receives the password, and while the password could still be read from that page, ClauLock blocks the agent's tools from looking at it.
The browser connection is locked to ClauLock itself: only the signed helper inside the ClauLock app may ask for a login, only when Chrome (or another supported browser) starts it, and only for the ClauLock extension. A script running on your Mac cannot pretend to be it. Every fill is recorded in Activity.
Setup lives in Settings → Browser: one button connects your browsers, and a short guide loads the extension. This first version is for macOS with Chrome and Chromium-based browsers.
- Fix #
v1.6.1 — a sidebar that shows everything
With a long Space name in the list, the whole sidebar grew wider than its column: the number of secrets in each Space fell off the edge, long names were cut without an ellipsis and the button to create a Space sat under the border. Every row now fits, long names end in an ellipsis and show in full when you hover them, and counts always line up on the right.
The sidebar is easier to live in with many Spaces. The main navigation stays put while the list below it scrolls, the first eight Spaces show with a link to the rest (and a filter once there are many), each section folds away and remembers it, and a Space's actions sit behind one menu instead of a row of icons over its name. Drag the sidebar's edge to make it wider or narrower; double-click the edge to reset it.
Page headers now match across the app, the item header gives its description the full width instead of a narrow column, item names wrap only between words, and the list toolbar's controls share one height.
After unlocking, ClauLock could keep macOS in Secure Keyboard Entry, which quietly broke other apps' keyboard shortcuts and text expansion until ClauLock quit. That is fixed.
- Release #
v1.6.0 — find it faster, keep it healthier
Press Command-Shift-Space anywhere (or Command-Option-Space if another password manager already owns the first) and start typing: Quick Access finds a secret by name, project or tag and copies it after Touch ID, without opening the main window. Command-K inside the app does the same for navigation and actions, and the menu-bar icon now searches too. Every copy still asks for your fingerprint and still clears itself from the clipboard after twenty seconds. A one-time code can stay live for a short window after one Touch ID, so it refreshes on screen the way you expect.
The new Health page reads your vault and tells you what wants attention: credentials stored in a shape that hides a missing username, rotations that are overdue, production secrets anyone at the keyboard could reveal, logins without two-factor codes, weak or reused passwords, credentials about to expire, secrets nobody has used and items that belong to no project. Each problem comes with a proposed repair, and nothing changes until you confirm it. Strength and reuse are judged inside the background service; the page itself never sees a value.
You can now bring your vault with you. ClauLock imports 1Password exports, Bitwarden exports and browser CSV files, from the app or the command line, and shows you what it found — names and types, never values — before anything is written.
Optionally, ClauLock can check your passwords against Have I Been Pwned. It is off until you turn it on, and even then only the first five characters of each password's fingerprint leave your Mac; the comparison happens locally, and neither the password nor the name of the item is ever sent. A password that turns up in a known breach is flagged on the Health page and on the item itself.
Items grew up. Creating one is a single step that either fully succeeds or leaves nothing behind, with tags, Space, environment, client and project set right there. Fields can be an email, a phone number or a date; a credential can carry an expiry date separate from its rotation schedule; favourites live in the vault instead of one window's memory; the last five previous values of a secret are kept encrypted and can be restored; and items can be duplicated or moved in bulk. Spaces and contacts are edited in place, projects get their own place in the sidebar, and each item shows who and what may use it — with Touch ID required only when you loosen access, never when you tighten it.
This release changes how the app and the background service talk to each other. Update both together — the desktop app and the command-line tools from the same version. An older background service will refuse the new app and say so on screen until it is updated, and a vault written by this version is not readable by earlier ones, so take the app's backup first if you might need to go back.
- Fix #
v1.5.3 — Windows: install, connect, use
A Windows user reported, step by step, everything that stood between the install command and a working ClauLock. Three of those steps blocked the install outright: the installer only ran under PowerShell 7 (it used an API the built-in Windows PowerShell does not have), it tripped over the redirect GitHub uses for the latest release, and with a user folder containing a space it declared a perfectly valid signature invalid — the worst failure possible for a check meant to build trust. Each is fixed; the installer now works in both PowerShells and never splits a path.
The bigger defect was underneath: the command line and the background service used two different names for the same connection, so nothing could talk to it — status, unlock, `clsec run` and secrets inside Claude Code all reported the service as unreachable while it was running fine. Both sides now derive that name from one place, and the check for “is the service running” no longer looks for a file that a Windows connection never creates.
Setup is one command. `clsec install` on Windows now creates the vault, stores the passphrase with Windows data protection, registers the service as a per-user logon task (no administrator rights, no 72-hour cut-off, no console window left on the taskbar), brings it up unlocked at every sign-in, and wires Claude Code's hooks and the ClauLock server — the same thing it has always done on macOS and Linux. The installer runs it for you; `clsec doctor` checks every piece.
- Fix #
v1.5.2 — the health warning now really clears
v1.5.1 added a one-line way to refresh the list of program files ClauLock expects to see talking to its background service. It wrote the new list correctly — but the background service only read that list when it started up, so nothing changed until the next restart.
Worse, it looked like it had worked: the health check hides entries older than the list itself, so the warning disappeared while the service was still going by the old list. The next program to connect would be flagged all the same.
The service now notices when the list is rewritten and picks it up straight away, without a restart. A list written while the service is already running arms immediately too. And if a program is still not covered by the new list, it gets reported again rather than staying quietly hidden.
- Fix #
v1.5.1 — a health warning you could not clear
ClauLock keeps a list of the program files it expects to see talking to the background service, and the health check warns when something outside that list connects — usually a copy left behind by an older version. Clearing the warning meant re-running the full installer, which redoes your keychain entry, the background service and the assistant wiring just to rewrite one small list.
`clsec install --baseline-only` now does only that. It refreshes the list from the copy of ClauLock actually on your path, and touches nothing else.
The check behind it also had a fault worth naming. On a Mac with the desktop app installed, pointing it at the wrong folder still found the app and reported success — while recording none of the command-line programs that actually connect. It now checks for those specifically, and refuses, loudly, rather than leaving you with a list that looks complete and is not.
- Release #
v1.5.0 — it keeps itself up to date, and waits its turn
Until now, staying current meant remembering to upgrade. The app could tell you a new version existed, but nothing installed it. ClauLock now checks once a day and installs new versions on its own, through whichever channel you originally installed from.
The interesting part is when it declines to act. The command line tool, the background service and the desktop app have to move together, so installing an update means restarting the background service — and restarting it locks your vault. If your vault is unlocked, something is using it right now, and an update at that moment would interrupt whatever that is. So updates apply only while the vault is already locked. If it is in use, the update simply waits for the next check.
The gentler-sounding alternative was to install the new files and leave the restart for later. That is the one arrangement that actually breaks: the new tools and the running service would disagree about how to talk to each other, and every stored value would stop resolving until the next restart. It is not offered.
You can see the state at any time with `clsec autoupdate status`, apply an update on the spot, or turn the whole thing off. Automatic updates are switched on when you install, and can be declined then or disabled later.
Separately, releases from this version onward carry a build attestation: a signed record of which source and which build produced each file, published to a public transparency log. Anyone can check it with `slsa-verifier`, without an account and without access to the source. Earlier releases do not have one and cannot be given one retroactively — an attestation is produced by the build that made the files. The existing signatures and checksums are unchanged and still cover everything.
- Release #
v1.4.0 — assistants save the whole login, not just the password
An account is a username and a password. Until now, when an assistant saved one for you it asked for a single value — so the password was stored properly and the email it already knew was written into the free-text description, or saved as a separate entry entirely. The result was a vault full of items that looked like passwords with no account attached, and pairs of entries that were really two halves of one credential.
Assistants now collect the identity and the password in the same prompt — one form, instead of one window per field. If an assistant says an item is an account, the username is collected whether or not it thought to ask for it, and an item saved as a plain password that turns out to have a username is filed as a login automatically.
For everything already stored, `clsec doctor --secrets` reports what it finds: an account whose email is stranded in the description, two entries that are really one credential, an item filed under the wrong type, or a login missing a field it needs. Adding `--fix` repairs the ones it is certain about, after copying your vault aside first. Merging keeps the original entry in place, so anything referring to it by its old name keeps working.
An email in a description is not just untidy: descriptions are readable by the assistant, and values in fields are not. Moving an identity into its own field takes it out of the assistant's view — so this is a privacy improvement as much as a tidiness one. Where the identity is already stored properly but a copy is still in the description, the check reports it and leaves the decision to you, because it cannot tell a leftover copy from a different person you wrote down on purpose.
Importing from 1Password used to keep one value per item and discard the rest, so every login arrived without its username, its website or its two-factor code. It now reads the whole item and keeps all of it, including one-time-password seeds, which then work as live codes.
Working with a credential's individual fields no longer needs the app: `clsec field` lists, sets, renames and removes them from the terminal, and `clsec merge` folds two entries into one. Renaming or retyping a field never moves its value.
Projects and workspaces became real things instead of strings you retyped. A project used to be a label repeated on every secret, so renaming it meant editing each one and a typo made a second project nobody noticed. Now it is an entity: rename it once and every secret follows, and a credential can serve several projects instead of being duplicated. `clsec project backfill` converts what you already have, splitting the labels that named two projects at once.
Duplicates can be merged — projects and accounts alike — without breaking anything that already points at them. The sidebar is editable, so a workspace can follow an account, a project, a filter, or whatever you decide, instead of showing whatever a field happened to say.
Search learned operators for vaults that outgrew scrolling: `project:acme -tag:legacy is:unused`. And a health panel says what the vault itself is neglecting — what is unused, unassigned, untagged, or overdue for rotation.
The app and the command-line tools update separately, and the app needs the updated tools. Installing the `.pkg` now restarts the background service for you; installing only the app does not, so if you take that route, restart the service afterwards from Settings › Security. The app will remind you.
One-way upgrade: opening your vault with this version updates its format, and earlier versions will not be able to open it afterwards. There is no way back, so if you keep an older copy of ClauLock around, upgrade it too.
- Release #
v1.3.2 — set up and connect everything from the app
On a clean machine the app used to open and go nowhere: it needs a background service running before it can create your vault, and installing ClauLock did not start one. It now starts the service itself the first time it needs it, so the setup wizard just works.
Connecting the tools that will use your secrets is now a button per tool. Cursor, Codex, Windsurf and Gemini used to require a terminal command; Claude Code needed one too. Connecting registers ClauLock for that tool and leaves the rest of its configuration untouched — restart the tool afterwards so it picks up the change.
A new Health section in Settings shows what ClauLock checks about your machine — is the service running, is your vault reachable, are your tools connected — and offers a one-click fix for the common problems. Where a repair needs a decision from you, it says what to do rather than pretending a button can decide it.
Assistants now know to ask before they are blocked. If you have put a secret behind approval, an assistant that needs it requests access with a reason and waits for you, instead of failing and — worse — asking you to paste the value into the chat.
Same invariant, as always: your assistant can use your secrets, and still cannot see them.
- Release #
v1.3.1 — Homebrew upgrades work again
On a Mac or Linux box where ClauLock came from Homebrew, running setup could replace brew's links with real files. Nothing looked wrong: brew kept reporting the new version, the formula stayed current, and the app kept working. But from that point on, upgrades stopped reaching the programs that actually run — you could upgrade for months and stay on the version you started with.
Setup now recognises a Homebrew-managed install and leaves those files alone, doing only the parts that are still needed: your vault, your keychain entry, the background service, and the assistant wiring. Upgrades go through brew, as they should.
If an earlier setup already broke the link on your machine, `clsec doctor` now reports it and prints the one command that repairs it. Worth running once if you installed with Homebrew before today.
No change to how secrets are stored or used, and nothing to redo.
- Release #
v1.3.0 — secrets with real structure, and 2FA that works
Until now a secret was one value with a name. In 1.3 it can hold the fields that actually belong together — a username next to its password, an API key next to its region and account id, a card next to its expiry. You pick the type when you create it and the app offers the fields that type usually carries, so you are filling in a form rather than inventing field names.
Every field carries a human label and a position you choose, so an entry reads like something a person wrote instead of a wall of slugs. Copy any one field in a single click; revealing still takes a deliberate second step behind Touch ID.
Two-factor codes are now first class. Paste an authenticator setup code once and ClauLock generates the live 6-digit code on your machine, with the countdown shown so you know whether to wait for the next one. Your AI can use that code in a command without the code — or the seed behind it — ever entering the conversation.
Your AI can now ask before it is blocked. Secrets you have gated behind approval used to fail first and explain later, which read like a broken tool. Now the assistant can request access up front, telling you which secret and why, and you approve it once with a time limit. It can also see what it currently holds and hand it back when the task is done.
Editing got easier: change a stored value from the app, reorder fields, and fix a description without recreating the entry.
Under the hood this release also closes a long list of hardening items from a full audit — tighter redaction of encoded values, plaintext cleared from memory on every path, reveals revoked the moment the vault locks, and stricter rules about which components may ask for what.
Same invariant, end to end: your AI can use your secrets — it still can't see them.
- Seguridad #
v1.2.1 — security hardening
1.2.1 is a security-hardening release following a full audit of ClauLock. The core guarantee held across every layer — the model never sees a plaintext secret — and this release tightens the defenses around it.
Redaction fails closed. If the output scrubber can't run for any reason, output is now withheld rather than shown, and even very short secrets are always redacted. A revealed secret is released only once it's recorded in the tamper-evident audit log.
Sync gets rollback protection. A compromised or misbehaving relay can no longer serve an older snapshot to quietly undo a credential rotation or bring back a secret you deleted — out-of-date snapshots are refused.
A more resilient app. The desktop app no longer blanks out on an unexpected error — you get a clear recovery screen instead — and anything copied to your clipboard is cleared automatically even if you close the window first.
Stronger supply chain. The installer verifies the release signature by default, and every shipped binary is signed, so a tampered download is caught before it runs.
Same invariant, end to end: your AI can use your secrets — it still can't see them.
- Release #
v1.2.0 — a calmer, 1Password-inspired workspace
ClauLock 1.2 reworks the desktop app around clarity. One title bar (no more doubled wordmark) with global search built in, a single collapsible sidebar, list rows trimmed to the essentials, and a read-first detail view that groups a secret's fields into one card — editing is now an explicit step.
The detail panel stays hidden until you pick a secret, so a fresh launch or a filtered list feels calm instead of cluttered.
Window dragging is back: you can move and reposition the window across displays again (a regression from the new title bar, now fixed).
Same invariant, end to end: your AI can use your secrets — it still can't see them.
- Función #
v1.1.0 — accounts & contacts for your secrets
ClauLock 1.1 adds a lightweight CRM over your vault. Create accounts (a client company or an individual) and contacts, then link them to a secret by role — account, owner, billing, technical, or security. Now "the Stripe key for Acme Corp, security contact Jane" is something you and your agent can actually find.
The privacy line holds and gets stronger: contact details (email, phone) are encrypted at rest and shown only in the app. Your AI agent can route by an account or contact's name and role, but the daemon never hands it an email or phone number — the same way it never hands over a secret value.
Everything stays local and end-to-end: the model can use your secrets and now understands who they're for, without ever seeing the secrets or anyone's personal details.
- Release #
v1.0.0 — new identity + structured secrets
ClauLock turns 1.0 with a complete visual rebrand. The new mark is the placeholder you already use — curly braces holding a keyhole — because that's exactly what ClauLock does: your agent references a secret by name, and the value stays locked. The app and the site move to a warm Vault palette (espresso, bone, champagne gold) that sets ClauLock apart from the sea of cold-blue security tools.
Structured secrets: a single entry — a login, a database, a cloud account — can now hold many fields (username, password, host, one-time-code seed, recovery codes), each revealed only with Touch ID and never shown to your agent.
Faceted Secrets workspace: organize by client, provider, project, or environment, with smart views, favorites, a deployment map, bulk actions, and per-field reveal.
Same invariant, end to end: the model can use your secrets — it still can't see them.
- Release #
v0.6.0 — DMG installer + first-run auto-detection
ClauLock now ships a real .dmg installer next to the .app.zip and the brew tap. Drag-to-Applications, eject, launch from Spotlight. The .dmg and the .app inside are both signed with our Apple Developer ID and notarized, so Gatekeeper accepts the install offline.
First-run auto-detection: the UI now opens the onboarding wizard automatically the first time you launch it without an existing vault. No flag, no terminal — drag-to-Applications, double-click, you're in the wizard. The previous CLI-driven onboarding still works exactly the same.
Tray + Dock pattern carried over from v0.5.2: Dock icon visible at launch, menu-bar tray icon also visible, close-to-tray instead of quit. If you prefer a pure-menu-bar boot, set CLAULOCK_BACKGROUND=1.
- Fix #
v0.5.2 — UX hotfix: Dock icon + window on launch + tray click
v0.5.0 ran as a pure menu-bar app (no Dock icon, window hidden at boot). On a fresh launch users saw nothing happen — no Dock bounce, no window. v0.5.2 reverts to the standard Mac layout: Dock icon visible, window opens at launch, menu-bar tray icon also present. Closing the window goes to the tray; quit comes from the tray menu.
Tray click no longer fails on the first try. Earlier versions had a small race where the very first click could miss its anchor and swallow itself; v0.5.2 always opens the popover.
Tray icon now shows "ClauLock" next to it on macOS so it's findable in cluttered menu bars and notch laptops.
- Release #
v0.5.0 — Menu-bar app + popover + branded icons
Left-click the menu-bar icon for a 320×420 popover that shows vault status, secret count, anything overdue, and one-click actions for Lock now, Add secret, and Open the full window. Right-click for a native menu (Lock now, Open ClauLock, Settings, Quit).
Closing the main window now keeps the process alive in the menu bar — your audit log, status, and quick actions stay one click away. Quit from the tray menu when you actually want to exit.
New branded icons: a cyan lock on a deep-ink background, designed at full 1024×1024 so the macOS .icns and Windows .ico render crisp at every size, plus a monochrome menu-bar template that recolors for light/dark/highlighted states automatically.
Note: v0.5.0 originally launched as a pure menu-bar app (no Dock icon). v0.5.2 reverts to the standard Mac layout — see that entry.
- Seguridad #
v0.4.0 — Apple Developer ID notarization + Touch ID native
Every macOS binary in the release is signed with our Apple Developer ID and submitted to Apple for notarization before shipping. Gatekeeper accepts ClauLock as "Notarized Developer ID" on every install — no "unidentified developer" warning, no right-click-to-open dance.
Touch ID now drives the vault unlock natively, end-to-end. Earlier macOS releases occasionally fell back to the device passcode when our entitlement didn't match what newer macOS expected; v0.4.0 ships the right entitlement so Touch ID is the only thing you ever see.
- Función #
v0.3.1 — Onboarding wizard in the desktop app
Same five steps as the terminal version (welcome → passphrase → unlock method → first secret → done) but in the ClauLock desktop window. Your passphrase only ever lives in the typing field — it's wiped from memory the moment the vault commits.
If you install via `clsec install`, the GUI opens by itself when the install finishes; if it can't reach the desktop app it falls back to the terminal wizard with a one-line pointer. Set CLAULOCK_NO_UI=1 to skip the GUI launch entirely.
- Función #
v0.3.0 — `clsec setup` first-run wizard
Run `clsec setup` and you walk through five screens: welcome (with the never-reveal model in plain English), passphrase + confirmation, unlock method (Touch ID / polkit / passphrase-only), an optional first secret, and a final health checklist. The passphrase is zeroed from memory the moment it's committed.
The wizard works in any terminal — SSH, container, headless CI runner — and the desktop GUI version of the same flow landed in v0.3.1.
- Fix #
v0.2.1 — Install robustness on macOS Tahoe
`clsec install` now writes binaries atomically — every step (copy, sign, verify, rename) is checked, and the destination only receives a fully-valid file or stays untouched. Filesystem races that occasionally produced zero-byte binaries on APFS are eliminated.
When macOS won't grant the Keychain biometric entitlement we asked for, `clsec install` now degrades cleanly to device-only or passphrase-only mode instead of aborting. The doctor check tells you which mode you're in. v0.4.0's Developer ID notarization removes the underlying issue entirely.
Launching the background daemon now has a foreground fallback: if launchd refuses to bootstrap the agent, `clsec install` keeps running with the daemon in the foreground and tells you what to look at. No more silent "I/O error" dead-ends.
The placeholder-substitution and output-scrubber hooks now ship inside the release tarball at a stable path, and `clsec install` finds them automatically. Without these, Claude's Bash tool calls would lose the never-reveal guarantee — keeping them in the tarball makes the invariant a default, not a configuration step.
The doctor check now distinguishes "no Keychain entry by design" (you chose `--no-keychain`) from "Keychain entry missing" (something broke). And the cosmetic "SHA256SUMS not found" warning on signed builds is silenced.
- Nota #
Site: Founders wall, OG cards, /compare polish, Windows .msi (beta), perf
The marketing site now opens with the leak, not the architecture: "You pasted a GitHub token into chat last week, didn't you?" Five new home sections walk through how a real paste plays out — the popup, the comparison vs. cloud password managers, common Claude workflows, an audit-log preview, and a public Founders wall. A new blog post tells the story end to end (EN + ES).
Performance + privacy foundation: scroll-reveal animations honor `prefers-reduced-motion`, fonts are self-hosted instead of a third-party CDN (faster paint, no font-loading privacy leak), and the install picker now follows the visitor's OS for both the primary and the secondary download paths.
Windows .msi installer is in beta — drag-to-run on Windows 10+, supply-chain story identical to every other artifact (minisign + cosign). The Authenticode signing certificate isn't live yet, so SmartScreen will show a one-time warning on first launch.
- Release #
v0.2.0 — cross-device sync (E2EE) + Pro/ENT relay
ClauLock now syncs secrets across your machines without ever giving the relay anything but encrypted blobs. The sync key is derived locally from your passphrase, separately from your local and backup keys; the relay can store and serve, but it can't read. The MCP server can't call sync either — only your CLI and UI can — so an exfiltration attempt from a compromised model can't reach the network path.
Pairing a new device uses SPAKE2 with a six-word code (or a QR you can scan from your phone). Vault format moves to v2 with per-record timestamps for clean last-writer-wins on simultaneous edits; older vaults open transparently and upgrade in place on the next write.
Pro tier (hosted): a relay we run for you, Stripe-billed, with the same never-reveal guarantee as the local install. Pricing and the Founder offer live on the pricing page.
Enterprise tier (self-hosted): the same relay binary, packaged with a Postgres + object-store bundle and an offline-verified license. License expiry flips the relay to read-only — it never bricks, you can always migrate off. Custom pricing.
On the desktop side, the ClauLock UI gains a Pair page and a sync status card that counts what was pulled, pushed, kept, and rejected.
- Release #
v0.1.2 — Windows port
ClauLock is now a tri-platform tool. The Windows port adds native binaries with a Windows-native keystore (DPAPI), a Windows-native IPC transport (Named Pipes) with peer authentication, and locked memory for the encryption key — symmetric with the macOS Keychain and Linux secret-service paths that have shipped since v0.1.0.
Install on Windows: `irm https://claulock.com/install.ps1 | iex`. The PowerShell installer mirrors the macOS / Linux flow — verify, extract to your user-local app data, add to PATH. macOS and Linux paths are unchanged.
Outstanding for upcoming releases: a fully Rust-native installer flow for Windows, a Microsoft Store / winget submission, and a desktop UI Windows bundle.
- Release #
v0.1.1 — supply chain audit fixes
Three CRITICAL fixes that unblocked the public install paths:
1) install.sh shipped with a literal `__MINISIGN_PUBKEY_PLACEHOLDER__` — the build never substituted it, so curl-pipe-sh fell through to a runtime fetch from a path that 404'd. Fix: embed the real pubkey (id 82B01474F764C7B2) in the proper two-line minisign format and correct the fallback URL to where the mirror actually stores it.
2) v0.1.0 release tarballs were missing the `clsec-scrub` binary, which `clsec install` requires. Fix: add `clsec-scrub` to release.yml's CLAULOCK_BINARIES and the tap formula's File.exist? install loop. v0.1.1+ tarballs ship all five binaries.
3) .deb / .rpm / .pkg builders hardcoded LICENSE-MIT in their file lists, but LICENSE-MIT was removed in the v0.1.0 license restructure. Fix: handle both the legacy v0.1.0 set (LICENSE-MIT + LICENSE-APACHE) and the v0.1.1+ set (LICENSE-APACHE + LICENSE-BSL + NOTICE) gracefully.
Plus a sweep of audit findings: README badges, SECURITY.md verification URLs, packaging/keys/README.md, CLA.md, threat-model cosign example, brand-page key-link 404s — all corrected to point at github.com/Mackint0uch/claulock-releases (the public mirror) instead of the still-private source repo.
- Release #
v0.1.0 — first public release
ClauLock is now public. Install with `curl -fsSL https://claulock.com/install.sh | sh` on macOS 13+ or a modern Linux.
What's in the box: encrypted vault (XChaCha20-Poly1305 + Argon2id), local daemon, 16 MCP tools for Claude Code, Tauri UI for unlock/reveal/approve, importers for .env / 1Password / Doppler, rotation queue with overdue notifications, per-secret usage timeline, audit log with deep-link filtering.
Release artifacts are signed with minisign + Sigstore cosign and include SLSA provenance. Verify by hand following packaging/RELEASE.md.
- Función #
Fase 2 — rotation + timeline + deep-links
The Tauri UI now surfaces rotation due-dates prominently: a red banner on the Secrets page counts every overdue secret and deep-links to a filtered rotation queue. The queue itself is sortable and shows 'days overdue' at a glance.
Each secret has a new usage timeline panel showing the last 30 days of unlocks, reveals, and exec uses, with one click to open the audit log pre-filtered to that secret.
Metadata (tags, rotation policy, description) is editable inline from the Secrets detail drawer — no modal, no round-trip.
- Función #
Output scrubber + unified error formatter
The scrubber watches every byte leaving the wrapped command and replaces any substring matching a vault secret with the original {{NAME}} placeholder before it reaches Claude. This catches the common leak path where a tool prints the value it was given (auth failures, debug output, badly scoped set -x).
Error messages across CLI + UI now flow through a single formatter so 'vault is locked' reads identically whether it's a Bash tool call or a /secrets page load.
Las entradas del changelog son bilingües en términos técnicos. Para contexto narrativo en español, visita el blog.