A l'arrivee sur l'ecran de connexion, /ping est interroge : son 404 JSON
exact ne peut venir, avant connexion, que de la 2e couche. Dans ce cas les
onglets disparaissent et seul Login reste. Tout autre echec laisse
Register visible — le masquer a tort empecherait un vrai client de creer
son compte. Le test porte sur le corps JSON, pas sur le seul statut : le
verrou du jeton repond aussi 404, sans corps.
Cosmetique : /register etait deja refuse cote serveur. On cesse seulement
de proposer un formulaire qui n'aboutit jamais.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Avant toute session, le 404 generique ne peut venir que de la 2e couche :
les routes appelees a ce stade existent toutes, et leurs vraies erreurs
portent un autre texte. api() vide son message, toast() ignore un message
vide — qui a rapiece la 1re couche ne voit rien.
Apres connexion rien ne change : pieces jointes et dossiers renvoient un
404 'Not found' legitime qu'il faut montrer. Prix assume : une route
perdue par un build casse echouerait aussi en silence avant connexion.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
LICENCE-TESTS.md : cinq series. B eprouve ce qui doit tomber (chaine
quelconque, licence tronquee, charge utile trafiquee, autre poste, autre
paire de cles), C le risque symetrique et plus couteux — refuser un
client qui a paye (espaces colles, retour a la ligne, redemarrage
machine). D est dediee a la stabilite de l'empreinte, le defaut deja
rencontre. E delimite la portee de l'activation.
Le plan citait cmd://licence/clear sans moyen de l'appeler : un bouton
"Deactivate (debug)" au bas de Settings -> Licence, revele seulement
quand config.txt existe. Il donne au passage son unique appelant a
ClearLicence, qui etait du code mort.
Le document dit aussi ce qu'aucun essai ne prouve : rapiecer l'exe reste
possible, il n'y a pas de revocation, et le brouillage n'est pas du
chiffrement.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Porteur, courriel, date d'emission et Installation ID, lus dans la charge
utile deja verifiee. Rempli une fois au demarrage sur les deux chemins
(deja licencie, activation reussie) : rien n'y change pendant une
session. Section masquee hors navigateur hote.
Les champs sont echappes avant interpolation dans ExecuteJavaScript bien
qu'ils viennent d'une charge utile signee — signee ne veut pas dire
inoffensive.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
C'etait la decision prise, et je l'avais mal lue : "pour le moment 1"
repondait a la question 1, pas "une seule licence".
PM.MachineId melange l'UUID de la carte mere (WMI) et la serie du volume
systeme, puis n'en publie que 16 hexa : l'empreinte se recopie dans un
courriel, l'UUID brut n'a pas a circuler. Un UUID nul ou tout a FF est
rejete — VM et images clonees l'annoncent, et deux postes tomberaient
alors sur la meme empreinte.
Le champ m est compare APRES la signature. m absent = licence flottante,
acceptee : seule la cle privee peut en emettre, et le champ ne peut pas
etre retire sans casser la signature. Cote script -Machine est
obligatoire sauf -Floating explicite.
L'ecran d'activation affiche l'Installation ID : une licence liee ne peut
plus etre emise avant l'installation.
Verifie sur ce poste : liee a lui acceptee, liee a un autre refusee,
flottante acceptee.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Le cyan de .auth-hint dit "voici l'etat des choses" (session expiree,
coffre verrouille). Un refus n'est pas un etat, c'est un echec : il doit
se distinguer a la couleur avant d'etre lu. Variante .is-error posee sur
le refus, retiree a chaque nouvel essai.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ensureLicence() en tete d'init(), avant tout le reste : un ecran
d'authentification qui s'affiche puis se fait recouvrir donnerait
l'impression que le coffre etait accessible.
L'ecran reprend les classes de l'ecran d'authentification — zero style
nouveau hormis un z-index. Le statut est demande deux fois avant de le
montrer : un demarrage a froid lent manquerait la fenetre de reponse, et
le silence se lirait comme un refus.
Une cle refusee ne dit pas pourquoi, et le journal note le verdict sans
jamais la cle.
ATTENTION : tant que LicencePublicKeyBlob est a zero, l'ecran est
infranchissable. Poser la cle publique avant de deployer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
localStorage meurt avec l'origine a chaque lancement, donc un autofill
desactive revenait sur ON. Miroite en pref DPAPI comme unlockMode et
rememberedUsername ; la copie DPAPI fait autorite, relue dans enterApp
juste avant la poussee des combos vers Delphi.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Synchroniser est une ACTION, pas un reglage. Elle n'existait qu'au fond
du panneau des reglages, a cote de l'URL et du jeton : on l'y configure
une fois, on la declenche tous les jours.
Sans URL, sans "Enable sync" ou sans mot de passe de sync, l'appui ouvre
les reglages ET DESCEND jusqu'a la section - un bouton qui ne peut
qu'echouer ne vaut pas mieux que pas de bouton, et repondre "Sync not
configured" laisse chercher ou reparer. Le panneau n'a pas d'onglets,
c'est une colonne longue : le battement dit lequel des blocs on est venu
chercher.
L'etat est lu SANS les secrets - hasEncPwd dit que le mot de passe existe
sans le lire, ce que loadSyncConfig documente : les tirer les planterait
dans le renderer pour la vie du processus.
L'entree disparait sans le pont, comme la section : le frontend PHP n'a
pas de sortie reseau.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Le meme defaut, par une autre porte : restaurer une sauvegarde par-dessus
des entrees existantes passe aussi par encryptImportEntry puis PUT, donc
re-chiffre un mot de passe inchange et fait croire au serveur qu'il a
change. La sync etait la cause de ce qui se voyait, pas la seule.
La regle est extraite en une fonction, keepCiphertextIfSamePassword,
appelee par les deux. Les re-publications partielles - tag, dossier,
migration des metadonnees - sont hors de cause : elles reutilisent deja
le chiffre stocke.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
L'instantane transporte le mot de passe EN CLAIR, et encryptImportEntry
le re-chiffre avec un IV neuf : le chiffre differe donc toujours, meme
quand le mot de passe n'a pas bouge. Or le serveur decide d'ecrire une
version dans l'historique en comparant les chiffres
(`encrypted_password <> :newenc`) - il voyait une difference la ou il n'y
en avait aucune, et chaque synchronisation qui redescendait une entree y
deposait un doublon date du jour. L'historique montrait la meme valeur
deux fois, a des jours d'ecart.
Le serveur ne peut pas trancher, il ne voit que du chiffre. Seul le
client connait les deux clairs : il compare, et rend le chiffre d'origine
quand ils sont egaux - ce que le commentaire du handler appelle deja
"JS reuses originalEncrypted bit-for-bit".
Trouve en deroulant la serie H sur le telephone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The desktop keeps twenty previous versions per entry and never sent any. It
now sends three.
Three, not twenty, because the trade is not symmetrical. On the desktop they
sit encrypted column by column and never move. In the snapshot everything
travels in the clear inside the container, and comes down into a phone cache
sealed only by the device key - not by the master password. An old password
stays sensitive: it is often still valid somewhere else. The real use of the
list is short - "I changed it recently, what was it?" - and three cover it.
The other seventeen would be copies with no need facing them.
The phone reads them ON DEMAND from the raw snapshot, never into TVault:
loading them with the vault would put every entry's old passwords in the clear
in memory from the moment it opens, for something one looks at now and then.
Copying goes by RANK, not by value. The page displays the text but never sends
it back - that is the bridge's rule everywhere else, and an old password has
no reason to be the exception just because it happens to be on screen.
They leave with the plaintext at lock, on both sides.
Measured on the phone, opening a cached vault: 121 ms to read the sealed file,
14 ms to display it - and then 5718 ms decrypting a snapshot it already had.
The cause is upstream. A sync pushes unconditionally at the end, and the push
is never a no-op: the container is re-encrypted with a fresh salt and IV, so an
identical vault produces completely different bytes. On Gitea that is one more
commit and a new blob sha, so every other device sees a "new" snapshot and pays
a full key derivation to arrive back where it started. It also grew the
repository on every sync that changed nothing, which is the history cost the
plan warns about.
The merged state is now compared against what was pulled, as plaintext -
the only place equality is observable - ignoring snapshot_at, which changes by
construction. Equal means the push is skipped and the sync still counts as done,
timestamp included: having nothing to write is a success, not an absence.
The comparison is deliberately strict. A false negative costs one commit; a
false positive would silence a push that carried a change and lose it for the
other devices. Four tests fix where that line sits, including a modified
updated_at (which arbitrates the merge elsewhere) and a tombstone (a deletion
that has to propagate).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Renaming the label from "Sync (WebDAV)" to "Sync (WebDAV or Gitea)" hid the
whole section from the Sync & Backup tab: applySettingsTab maps sections to
tabs BY THEIR VISIBLE LABEL TEXT, so the miss fell through to 'general'. The
section was not lost - it moved to the General tab, which is worse than
missing, because it is nowhere you would look for it.
The map is updated, and the coupling is now written down where the map lives:
these keys are the displayed strings, word for word, and rewording one means
touching both sides.
Checked every key against the labels in index.html; the only sections without
an entry are "Getting started" and "Startup & tray", which belong in General
anyway.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Same discriminant as the Android client - "/api/v1/repos/" in the URL - so one
address tells both ends which transport to speak, and nothing new appears in
Settings beyond a placeholder and a warning.
The bridge stays shared. Three things differ, and they are confined: the header
is "token <PAT>" rather than Basic (Gitea answers a silent 401 to the other),
a PUT carries a JSON envelope with the file base64-encoded inside it, and the
connection test uses GET because the contents API does not answer HEAD.
The sha does the ETag's whole job. It comes back from the pull, rides in the
PUT body instead of an If-Match, and a stale one earns a 409 - handled exactly
like WebDAV's 412, by re-running pull->merge->push so the other device's
changes are folded in rather than clobbered. On a first push `sha` is OMITTED
rather than sent empty: Gitea reads its presence as "replace that exact
version". `branch` is omitted too, so the repo's default is used instead of a
name we would be guessing.
A failing PUT now reads its response body: Gitea explains a 409 or a 422 there,
and without it the error was just a number.
The test button checks the SHAPE of what came back, not only the status. A URL
one path segment off answers 200 with valid JSON that is not a file - which
would otherwise surface much later as "wrong sync password" and send you
looking on the wrong side. Four tests cover that parsing, including the
line-wrapped base64 Gitea sends and the array an accidental directory URL
returns.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The map was built from state.entries only, so a container entry whose uuid existed just in the trash looked fresh and inserted a second row sharing that uuid - the identity the sync engine relies on. Same blind spot as the resurrection bug in applyRemoteSnapshot.
Also documents F9 and warns that emptying the trash is the wrong way to clean duplicate uuids: the tombstone would kill the live twin on the other device.
A backup that silently drops the trash is not a backup: export, empty the vault, import, and the trashed entries were gone for good with nothing said. Both containers now carry state.trashed with deleted/deleted_at, and deleted_at is preserved so a restore resumes the auto-purge clock instead of restarting it - hence its addition to the GET echo.
The sync snapshot deliberately does NOT: the trash is device-local, and re-publishing deleted rows is what produced the duplicate-uuid bug. The two identical export loops were folded into buildBackupEntry first, so the trash handling exists once.
The autofill hotkey already picked whichever field still needed filling; quick search forced the password, so a first unlock or one without a remembered username landed the cursor in the second field.
Both locked-vault paths cancelled the autofill with the default refocus, which returns the foreground to the captured target - right after Delphi had raised our window. Ctrl+Shift+Q flashed for a frame then Notepad took over; Ctrl+Shift+L moved the window twice because focusApp had to take it back.
The trash is a soft delete: no tombstone is written and the rows are absent from state.entries, so the merge read their uuids as missing locally and POSTed them back as fresh live entries. Trashing 30 entries then syncing produced 30 live duplicates alongside the trashed originals - two rows sharing one uuid.
applyRemoteSnapshot now loads the trash and skips any uuid found there. Emptying the trash is what writes the tombstone that propagates the delete.
Assigning .value fires nothing, and the code called soDirtyCheck by hand - so the strength bar, which listens to input, stayed frozen on the previous password. Dispatching the event covers every listener on the field instead of one named handler.
inert removes elements from focus and hit testing, but shortcuts are listened for on the document and never traverse an element. Ctrl+A still selected every entry under the overlay, Ctrl+K would open the palette behind it, j/k moved the cursor - and Delete ran a batch delete while a sync was writing to the same rows.
A renderer dump taken after a quick-unlock, with nothing opened, contained the auto-backup password in cleartext - inside an ExecuteJavaScript string and in the DevTools protocol buffers - and it survived the auto-lock that wipes the vault key.
New prefs/has answers set-or-not without moving the value. The settings panel uses that; only the paths that encrypt a backup, test the WebDAV connection or run a sync fetch the secrets themselves.
renderGrid replaces every card, so the scroll container jumped back to the top: ticking a card near the bottom of a long list threw the user back to the first row. Only the affected card and the selection bar are updated now.
content-header and pagination were exempt from click-away in full, so their large empty areas were dead: clicking beside the title or next to the page buttons left the editor open. Only the actual controls are exempt now.
They accepted any drag then looked for ids in text/plain: dragging text selected in a cell produced NaN, filtered out, so the target invited a drop that could do nothing. Entry drags now carry application/x-pm-entries, mirroring the folder MIME, and all four targets check it before accepting.
Checked ids belong to the view they were made in. Carried into the trash they were LIVE entries while the trash bar offers Delete forever, so a purge could destroy rows that had never been trashed. The checkbox-less views were the visible symptom of the same problem; one rule replaces the special case.
The batch bar survived into vault health, the authenticator grid and the audit log, none of which render checkboxes - it kept offering Delete, Move to folder and Add tag on entries the user could no longer see selected.
Free-floating it either covered the cards or the sidebar navigation. Inside .toast-container it shares the corner with the toasts, and column-reverse keeps it above them.
The pill was pinned to the left edge and covered the navigation, hiding Trash. It also kept counting after Breach check disabled, the workers carrying on with one network round-trip per entry.
Bottom-left, out of the way of the toasts, with a live count. Deliberately not the busy overlay: the scan is network-bound and can take a while, and there is no reason to lock the vault while it runs.
The dashboard reads its pwned column from state.hibpResults but its cache was computed before the scan finished, so enabling the breach check from the health view showed 0 pwned while the entry cards already carried the flag. Disabling had the mirror problem: results cleared, dashboard still counting them.
Each category is capped, so past the cap fixing one more password leaves the total unchanged and the work looks wasted. The breakdown lists count and points per category, marks a capped deduction with (max), and explains in a tooltip why the score is standing still.
inert removes the controls from focus but not the document scroll: Space, PageUp/Down, Home/End and the arrows scroll without any focused element, so the page still moved under the overlay.
The overlay covers the page, which stops the mouse but not Tab and Space. The inert attribute removes the subtree from focus, hit-testing and the accessibility tree in one go - applied to every body-level sibling except the overlay, and always cleared in hideBusy.
The busy guard cancelled through the path that hands the foreground back to the captured target. During an export that pushed the native Save dialog behind the other application, leaving the app disabled by a modal the user could no longer find. refocus=0 separates refusing a request from ending a user interaction.
Guarded in autofillHandleRequest, upstream of the single-match versus picker split, so it covers Ctrl+Shift+L/P/T and both outcomes: the picker would open behind the overlay, and a single-match fill would minimise our window mid-operation. Cancels the request so Delphi drops the captured target HWND.
Global hotkeys come from Win32, outside the window, so the busy overlay intercepts nothing: Ctrl+Shift+A opened the editor behind it, unreachable. Guarded in openSlideOver and openQuickSearchModal rather than at each trigger - every caller routes through them, and no save path raises the overlay.
There are two Escape handlers: a capture-phase one for the editor and the global one-surface-per-keystroke chain. Only the first was guarded, so export - which reaches the editor through the global chain - still popped the discard-confirm under the overlay.
The discard-confirm opened underneath the busy overlay, which sits at z-index 200 - blurred and half-clickable. Worse than the display: it let the user arbitrate closing the editor while a sync or a rotation was writing to the same entry.
A 404 on save meant the work had to be copied out by hand. The POST mints a fresh uuid, so this is a new entry rather than a resurrection and cannot fight the tombstone the delete left behind. Gated on status 404 AND the server message: the HTTP layer answers 404 to any request it refuses, which the status alone would misread.
Same placement as the TOTP field, which buys back a column on a dense row: label | value (eye inside) | copy | menu. Pinned top-right on a multiline field so it stays clear of the resize grip.
The inline copy button stays: copy is the frequent action and burying it costs a click every time. Unlike the old lock-versus-eye pair, two ways to copy are not confusable.
The lock button sat next to the eye and both masked the value, so they read as the same control. The eye now carries the state - cyan ring when the field is stored as secret - and flipping the flag moved into a row menu alongside Remove field, one deliberate step away from the everyday reveal. The row drops from six columns to five.
The menu reuses the card kebab classes, so closing it on an outside click is already handled.
A manual resize was lost on reopen: the row is rebuilt on every open, and remembering the height would not survive a restart either since localStorage is wiped when the ephemeral port changes the origin. Deriving it from the content removes the state entirely - the field is always the right size.
The note body of a note entry is the encrypted_password column, so logins had no way to carry notes. Rather than a dedicated encrypted column and the seven call sites that come with it, this rides on custom_fields, which rotation, duplicate, export/import and sync already carry.
Both the open-time normalisation and the save-time rebuild drop any key they do not name, so multiline is carried explicitly in each - otherwise the field silently reopens single-line.
r-zakarya