Live test checklist: open tests from v0.6.351 to v0.6.354 (Windows DC, Entra, Keycloak, Kubernetes, accessibility) #1005

Open
opened 2026-10-25 22:45:45 +00:00 by Cordy · 54 comments
Owner

One place for every live test that unit/integration tests cannot replace. Each line names what it unblocks and where the exact steps are. Tick when done and post the result on the linked issue.

Research files are in scratch/b7e/backlog/.

A. Real Windows domain controller (Manuel's AD) — unblocks #741, #991

Steps: research-985-1 §3 (about 15 min, plus 1 day if 532 needs a real expiry).

  • AD sub-codes after a correct password vs a wrong one.
    • Setup: 9 prepared accounts plus one PSO; for each account, ldapwhoami with the right and with a wrong password; read badPwdCount / lockoutTime.
    • Unblocks: the "password must be changed" message (532/773 switch, off today); 533/701 as "disabled"; whether a correct password on a disabled account touches badPwdCount.
  • Lock badge attribute.
    • Check: does AD return msDS-User-Account-Control-Computed (plus lockoutTime) in Cairn's re-check search?
    • Lock a test account; the badge must appear in Users & access › Accounts and directory-locked must be audited once.
    • If the attribute is missing, the badge never shows: say so in the docs.
  • Timing (#991). 50 sign-ins each: unknown name, wrong password, correct-but-not-allowed. Report median and spread.
  • AD lockout cache (#992). With a PSO lockout threshold of 3, a device repeating one wrong password must not lock the Windows account. The AD refusal log line must carry remote= and client=.
  • Four-field config end to end (#741 comment 2026-10-24).
    • Sign in as a member and as a non-member.
    • Remove a member: sessions end and devices pause at the next re-check.
    • Nested group; UPN login; two DCs with the first one down (facts block names the second).
  • TLS failure class over LDAPS and StartTLS with a CA that is not trusted (research-985-3 §11 T2). Certificates page and Check connection must say "unknown authority".
  • Samba AD (optional, #990). Groups A ⊃ B ⊃ user, service account denied read on B's OU. The re-check must report an outage (WARN about bug 15515), not "not a member".

B. Microsoft Entra ID test tenant — confirms #989, #1003, research-985-7/-8

Steps: research-985-8 §8.1 (steps 1-5) and research-985-7 §9.

  • App roles with Assignment required = No: remove the role, and the token carries no roles claim.
  • Assignment required = Yes: removed user gets AADSTS50105 at Entra (screenshot the page); a Global Administrator with no assignment arrives without roles.
  • groups claim with a user in zero groups, and "Groups assigned to the application" with no assigned group: claim absent, not [].
  • Overage: a user in more than 200 groups produces _claim_names; Cairn logs the overage WARN and the Users & access counter increases.
  • prompt=select_account shows the picker; sign-out without a hint returns to /login.
  • Client secret expiry wording (AADSTS7000222) for the docs (#996 context).

C. Keycloak dogfood (can be done now; needs Nikola's sign-ins and GitOps config changes)

  • Refusal page (mockup A frames 8a/8b/10b/11): set oidcAllowedGroups on a dogfood to a group nikola-test is NOT in. Check:

    • the sign-in lands on /login?refused=not-allowed;
    • "Sign out of single sign-on" goes through Keycloak's logout confirmation and back to /login;
    • the session ends and devices pause;
    • auth-signin-refused is in the audit.

    Then add the user back: the next accepted sign-in lifts the pause. Revert the config afterwards.

  • Removed at re-check: remove the test user from the allowed group while signed in. The next visit shows "Your access to Cairn was removed".

  • #987 / #988: admin group as a subgroup /org/cairn-admins written as its full path: the admin list shows its members. A bare entry that exists only as a subgroup triggers the start WARN and the facts row "subgroup only".

  • prompt=login / prompt=select_account behaviour on Keycloak 26 (research-985-7 §9 item 3).

  • Login CSRF manual check (#999): a callback URL pasted into another browser gives "login failed".

  • #1003 (after v0.6.355): research-985-8 §8.2 Keycloak steps 1-5 (empty group list = claim absent, the safety stop after 3).

D. Kubernetes

  • Vault or OpenBao Agent injector (#994). The chart's serviceAccount values plus projected token: the pod is mutated, /vault/secrets/oidc exists, and the Cairn container has no /var/run/secrets/kubernetes.io. Until then the CHANGELOG says "documented, not yet tested against a live injector".
  • Converter with a read-only root (#997).
  • TLS rotation timing with 3 replicas (research-985-3 §11 T3): the time for tls: server certificate reloaded on each pod, which sets the "instances disagree" grace.

E. Accessibility

  • VoiceOver and NVDA: the sign-in error with role="alert" is announced, including the same message twice in a row; focus stays in the form.

F. Older open live checks (Nikola)

  • #914: Finder and rclone steps for #903 and #908.
  • #915 / #972: two-account check (a grantee writes through the owner's lock; the actor is recorded in Deleted files).
One place for every live test that unit/integration tests cannot replace. Each line names what it unblocks and where the exact steps are. Tick when done and post the result on the linked issue. Research files are in `scratch/b7e/backlog/`. ## A. Real Windows domain controller (Manuel's AD) — unblocks #741, #991 Steps: research-985-1 §3 (about 15 min, plus 1 day if 532 needs a real expiry). - [ ] **AD sub-codes after a correct password vs a wrong one.** - Setup: 9 prepared accounts plus one PSO; for each account, `ldapwhoami` with the right and with a wrong password; read `badPwdCount` / `lockoutTime`. - Unblocks: the "password must be changed" message (532/773 switch, off today); 533/701 as "disabled"; whether a correct password on a disabled account touches `badPwdCount`. - [ ] **Lock badge attribute.** - Check: does AD return `msDS-User-Account-Control-Computed` (plus `lockoutTime`) in Cairn's re-check search? - Lock a test account; the badge must appear in Users & access › Accounts and `directory-locked` must be audited once. - If the attribute is missing, the badge never shows: say so in the docs. - [ ] **Timing (#991).** 50 sign-ins each: unknown name, wrong password, correct-but-not-allowed. Report median and spread. - [ ] **AD lockout cache (#992).** With a PSO lockout threshold of 3, a device repeating one wrong password must not lock the Windows account. The AD refusal log line must carry `remote=` and `client=`. - [ ] **Four-field config end to end (#741 comment 2026-10-24).** - Sign in as a member and as a non-member. - Remove a member: sessions end and devices pause at the next re-check. - Nested group; UPN login; two DCs with the first one down (facts block names the second). - [ ] **TLS failure class over LDAPS and StartTLS** with a CA that is not trusted (research-985-3 §11 T2). Certificates page and Check connection must say "unknown authority". - [ ] **Samba AD (optional, #990).** Groups A ⊃ B ⊃ user, service account denied read on B's OU. The re-check must report an outage (WARN about bug 15515), not "not a member". ## B. Microsoft Entra ID test tenant — confirms #989, #1003, research-985-7/-8 Steps: research-985-8 §8.1 (steps 1-5) and research-985-7 §9. - [ ] App roles with Assignment required = No: remove the role, and the token carries no `roles` claim. - [ ] Assignment required = Yes: removed user gets AADSTS50105 at Entra (screenshot the page); a Global Administrator with no assignment arrives without `roles`. - [ ] `groups` claim with a user in zero groups, and "Groups assigned to the application" with no assigned group: claim absent, not `[]`. - [ ] Overage: a user in more than 200 groups produces `_claim_names`; Cairn logs the overage WARN and the Users & access counter increases. - [ ] `prompt=select_account` shows the picker; sign-out without a hint returns to `/login`. - [ ] Client secret expiry wording (AADSTS7000222) for the docs (#996 context). ## C. Keycloak dogfood (can be done now; needs Nikola's sign-ins and GitOps config changes) - [ ] **Refusal page** (mockup A frames 8a/8b/10b/11): set `oidcAllowedGroups` on a dogfood to a group `nikola-test` is NOT in. Check: - the sign-in lands on `/login?refused=not-allowed`; - "Sign out of single sign-on" goes through Keycloak's logout confirmation and back to `/login`; - the session ends and devices pause; - `auth-signin-refused` is in the audit. Then add the user back: the next accepted sign-in lifts the pause. Revert the config afterwards. - [ ] **Removed at re-check:** remove the test user from the allowed group while signed in. The next visit shows "Your access to Cairn was removed". - [ ] **#987 / #988:** admin group as a subgroup `/org/cairn-admins` written as its full path: the admin list shows its members. A bare entry that exists only as a subgroup triggers the start WARN and the facts row "subgroup only". - [ ] **`prompt=login` / `prompt=select_account` behaviour on Keycloak 26** (research-985-7 §9 item 3). - [ ] **Login CSRF manual check (#999):** a callback URL pasted into another browser gives "login failed". - [ ] **#1003 (after v0.6.355):** research-985-8 §8.2 Keycloak steps 1-5 (empty group list = claim absent, the safety stop after 3). ## D. Kubernetes - [ ] **Vault or OpenBao Agent injector (#994).** The chart's `serviceAccount` values plus projected token: the pod is mutated, `/vault/secrets/oidc` exists, and the Cairn container has no `/var/run/secrets/kubernetes.io`. Until then the CHANGELOG says "documented, not yet tested against a live injector". - [ ] **Converter with a read-only root (#997).** - [ ] **TLS rotation timing with 3 replicas** (research-985-3 §11 T3): the time for `tls: server certificate reloaded` on each pod, which sets the "instances disagree" grace. ## E. Accessibility - [ ] VoiceOver and NVDA: the sign-in error with `role="alert"` is announced, including the same message twice in a row; focus stays in the form. ## F. Older open live checks (Nikola) - [ ] #914: Finder and rclone steps for #903 and #908. - [ ] #915 / #972: two-account check (a grantee writes through the owner's lock; the actor is recorded in Deleted files).
Author
Owner

G. Group identity (#1002, #919, #933, #965, #967), added 2026-10-26

Steps are in research-1002-1-cairn.md (live-test list), research-1002-2-market.md §6 and research-1002-3-standards.md §5.

  • #1002 repro on the Keycloak dogfood test realm. Create a top-level /proj directory space and a group /other/proj with a test user only in the second. Sign in: today the user sees the space; after the fix they must not.
  • Keycloak "Full group path" off: the claim carries proj for /other/proj (so Cairn cannot place it without the admin client).
  • Keycloak sibling rule: staff twice at top level gives 409. Also check whether top-level names differing only in case are allowed.
  • Keycloak group rename with Relink: a space whose group is renamed in the IdP; check Relink and the 409 case (spacerename.go:136).
  • Two replicas after a group is deleted and recreated: check which key each instance uses (keyset cache without TTL, keycloak_group.go:70).
  • AD: the same group with cn differing from sAMAccountName, with ldapGroupBaseDn off and then on; check the space membership.
  • AD: objectGUID survives rename and move; a group deleted and recreated gets a new GUID; Recycle Bin restore keeps it (standards §5).
  • OpenLDAP entryUUID / 389-ds nsUniqueId / FreeIPA ipaUniqueID: read via the service account; 389-ds entryUUID plugin with a client-supplied value (pitfall).
  • Entra: cloud-only group display names vs object ids in the groups claim.
  • Optional market checks (only if a decision depends on them): Nextcloud LDAP recreate at the same DN, Nextcloud duplicate cn labels, OpenCloud autoprovision rename, Synology rename/recreate, Google group rename, GitLab duplicate CN.
## G. Group identity (#1002, #919, #933, #965, #967), added 2026-10-26 Steps are in `research-1002-1-cairn.md` (live-test list), `research-1002-2-market.md` §6 and `research-1002-3-standards.md` §5. - [ ] **#1002 repro on the Keycloak dogfood test realm.** Create a top-level `/proj` directory space and a group `/other/proj` with a test user only in the second. Sign in: today the user sees the space; after the fix they must not. - [ ] **Keycloak "Full group path" off:** the claim carries `proj` for `/other/proj` (so Cairn cannot place it without the admin client). - [ ] **Keycloak sibling rule:** `staff` twice at top level gives 409. Also check whether top-level names differing only in case are allowed. - [ ] **Keycloak group rename with Relink:** a space whose group is renamed in the IdP; check Relink and the 409 case (spacerename.go:136). - [ ] **Two replicas after a group is deleted and recreated:** check which key each instance uses (keyset cache without TTL, keycloak_group.go:70). - [ ] **AD:** the same group with `cn` differing from `sAMAccountName`, with `ldapGroupBaseDn` off and then on; check the space membership. - [ ] **AD:** `objectGUID` survives rename and move; a group deleted and recreated gets a new GUID; Recycle Bin restore keeps it (standards §5). - [ ] **OpenLDAP `entryUUID` / 389-ds `nsUniqueId` / FreeIPA `ipaUniqueID`:** read via the service account; 389-ds entryUUID plugin with a client-supplied value (pitfall). - [ ] **Entra:** cloud-only group display names vs object ids in the groups claim. - [ ] **Optional market checks** (only if a decision depends on them): Nextcloud LDAP recreate at the same DN, Nextcloud duplicate `cn` labels, OpenCloud autoprovision rename, Synology rename/recreate, Google group rename, GitLab duplicate CN.
Author
Owner

Additions (2026-10-26)

G. Group identity, case and Unicode (research-1002-4 live tests):

  • Keycloak on Postgres (and MariaDB if available): can Finance and finance, and Zürich in NFC and NFD, exist side by side?
  • OpenLDAP: does cn matching really fold case and normalise (RFC 4518), and does a stored NFD name come back unchanged?
  • AD: after a case-only group rename, does memberOf change case, and do members lose the space?
  • Entra: are group GUIDs in the token upper- or lower-case?

B / C. Safety stop (research-985-9 live tests):

  • Entra with Assignment required off: unassigned members, guests and a Global Administrator sign in; consent behaviour; a real mapper outage.
  • Keycloak: a user with no group (no claim since KC 22), with and without the admin client; the re-check re-pauses within 5 minutes.
## Additions (2026-10-26) **G. Group identity, case and Unicode** (research-1002-4 live tests): - [ ] Keycloak on Postgres (and MariaDB if available): can `Finance` and `finance`, and `Zürich` in NFC and NFD, exist side by side? - [ ] OpenLDAP: does `cn` matching really fold case and normalise (RFC 4518), and does a stored NFD name come back unchanged? - [ ] AD: after a case-only group rename, does `memberOf` change case, and do members lose the space? - [ ] Entra: are group GUIDs in the token upper- or lower-case? **B / C. Safety stop** (research-985-9 live tests): - [ ] Entra with Assignment required off: unassigned members, guests and a Global Administrator sign in; consent behaviour; a real mapper outage. - [ ] Keycloak: a user with no group (no claim since KC 22), with and without the admin client; the re-check re-pauses within 5 minutes.
Author
Owner

F. Addition (2026-10-26):

  • #1009: after the next dogfood version bump (after v0.6.356), Administration › Certificates and Users & access › "Other instances" show no "not reporting" entry for the replaced pod, and the old pod's log contains removed this instance's status entries on shutdown.
**F. Addition (2026-10-26):** - [ ] **#1009:** after the next dogfood version bump (after v0.6.356), Administration › Certificates and Users & access › "Other instances" show no "not reporting" entry for the replaced pod, and the old pod's log contains `removed this instance's status entries on shutdown`.
Author
Owner

G. v0.6.358 space keys (#1002 stage 3), added 2026-10-26

Done on 2026-10-26:

  • Move counts: files-bao copied 5, missing 8 (all empty, see #1018); files copied 4, missing 2 (Café-922, cairn-admins). No WARN or ERROR.
  • Check keys: files-bao 5/5, files 4/4.
  • Every file in the visible spaces opens on files-bao (34 files including trash).

Open (each changes state; needs Nikola's go and his sign-in):

  • #967 step 1: on files (Keycloak), create an app-owned test space and upload a file. Expect /cairn-keys/<id>, state ok, and the file opens. Then purge the space and destroy its key (also covers the next item).
  • Purge and destroy: the key is listed under Keys of purged spaces; typed confirm; key-destroyed in the audit; the holder is gone.
  • Rotate a test space: generation 2, the older key kept, old and new files open, and a second admin tab with a stale generation gets 409.
  • Pin-mismatch drill: edit a test space's holder attribute in Keycloak. Uploads get 503 key-changed, reads still work, the red card shows, and Check again clears it after you restore the value.
  • Remove old copies: only after the items above. Both pods must report v0.6.358, Café-922 must have had no upload, and no group or space may be named cairn-keys. Expect files: 4 removed; files-bao: 5 removed. Every file still opens afterwards. Do not roll back to v0.6.357 after this without the recovery key.
  • Rollback drill (optional): on a copy of files-bao state only.
## G. v0.6.358 space keys (#1002 stage 3), added 2026-10-26 Done on 2026-10-26: - [x] Move counts: files-bao copied 5, missing 8 (all empty, see #1018); files copied 4, missing 2 (Café-922, cairn-admins). No WARN or ERROR. - [x] Check keys: files-bao 5/5, files 4/4. - [x] Every file in the visible spaces opens on files-bao (34 files including trash). Open (each changes state; needs Nikola's go and his sign-in): - [ ] **#967 step 1:** on files (Keycloak), create an app-owned test space and upload a file. Expect `/cairn-keys/<id>`, state `ok`, and the file opens. Then purge the space and destroy its key (also covers the next item). - [ ] **Purge and destroy:** the key is listed under Keys of purged spaces; typed confirm; `key-destroyed` in the audit; the holder is gone. - [ ] **Rotate a test space:** generation 2, the older key kept, old and new files open, and a second admin tab with a stale generation gets 409. - [ ] **Pin-mismatch drill:** edit a test space's holder attribute in Keycloak. Uploads get 503 `key-changed`, reads still work, the red card shows, and Check again clears it after you restore the value. - [ ] **Remove old copies:** only after the items above. Both pods must report v0.6.358, Café-922 must have had no upload, and no group or space may be named `cairn-keys`. Expect files: 4 removed; files-bao: 5 removed. Every file still opens afterwards. Do not roll back to v0.6.357 after this without the recovery key. - [ ] **Rollback drill (optional):** on a copy of files-bao state only.
Author
Owner

Section G, run on 2026-10-26:

  • #967 key path on files (Keycloak): test space zz-967-test. Its first upload created /cairn-keys/<id>, the state went from new to ok, and the file reads. #967 is closed. Files runs power mode, so app-owned spaces don't exist there.
  • Purge and destroy on files-bao (OpenBao): test space zz-destroy-test.
    • Its first upload made spaces/<id>; the state went from new to ok.
    • Deactivate, delete and purge all succeeded; the key was then listed as retired (key-retired audited).
    • Destroy with a wrong name answered 409 confirm-mismatch. With the right name it answered 200, state destroyed, key-destroyed generations=1 audited.

Side effect to clean up (Nikola): both dogfoods use the realm cordyceps. Creating zz-967-test on files (power mode) created the Keycloak groups /zz-967-test and /zz-967-test-ro. files-bao saw them and bound a directory space zz-967-test too (space-group-bind ... by=system, 23:28:38Z). Deleting the two groups in Keycloak sends the space to Needs review on both instances; unbind it there.

Still open in G: rotate, the pin-mismatch drill, Remove old copies, and the optional rollback drill.

**Section G, run on 2026-10-26:** - [x] **#967 key path on files (Keycloak):** test space `zz-967-test`. Its first upload created `/cairn-keys/<id>`, the state went from `new` to `ok`, and the file reads. #967 is closed. Files runs power mode, so app-owned spaces don't exist there. - [x] **Purge and destroy on files-bao (OpenBao):** test space `zz-destroy-test`. - Its first upload made `spaces/<id>`; the state went from `new` to `ok`. - Deactivate, delete and purge all succeeded; the key was then listed as retired (`key-retired` audited). - Destroy with a wrong name answered 409 `confirm-mismatch`. With the right name it answered 200, state `destroyed`, `key-destroyed generations=1` audited. **Side effect to clean up (Nikola):** both dogfoods use the realm `cordyceps`. Creating `zz-967-test` on files (power mode) created the Keycloak groups `/zz-967-test` and `/zz-967-test-ro`. files-bao saw them and bound a directory space `zz-967-test` too (`space-group-bind ... by=system`, 23:28:38Z). Deleting the two groups in Keycloak sends the space to Needs review on both instances; unbind it there. Still open in G: rotate, the pin-mismatch drill, Remove old copies, and the optional rollback drill.
Author
Owner

H. Live checks for v0.6.359 to v0.6.364 (added 2026-10-27)

v0.6.362, Add files only (#900):

  • Connect an rclone device as Add files only, with a new folder. Then check:
    • rclone copy --immutable works;
    • replacing or deleting an old file is refused (403 device-limited);
    • its own new file can be replaced within 60 minutes;
    • after 10 refusals the bell item appears.
  • Widen from read only to add: "Sign in again" (IdP reauth) keeps add.

v0.6.363, account policy:

  • Reset a local test account's password: the dialog says one-time, the row shows "New password pending", the person gets the bell, and the sign-in asks for a new password of at least 15 characters (frame K).
  • Rollback check: neither tag is shown before rolling back.

v0.6.364, storage restored (#904): needs Nikola, and test data on a dogfood.

  • Declare a test restore in Storage & quotas (scope all, links paused, short notice). Then check:
    • a connected device gets 403 device-paused, and its resume shows the advice;
    • a sync app using the account password is paused and resumes from Connected devices;
    • every user sees the banner, and "Got it" only hides it;
    • the checklist's Accounts row lists accounts changed after the backup;
    • "Resume public links" works;
    • Close the checklist.
  • cairnd restored --backup-time … --notice-days 14 while Cairn is stopped (Kubernetes: scale to 0, then a Job), then start Cairn.
  • On S3 (files-bao), with a copy of the bucket: roll the bucket back, and the suspicion banner appears within about a minute; "Not a restore" lifts it.
  • Incident box on a test realm: every local account must choose a new password at sign-in.
## H. Live checks for v0.6.359 to v0.6.364 (added 2026-10-27) **v0.6.362, Add files only (#900):** - [ ] Connect an rclone device as Add files only, with a new folder. Then check: - `rclone copy --immutable` works; - replacing or deleting an old file is refused (403 `device-limited`); - its own new file can be replaced within 60 minutes; - after 10 refusals the bell item appears. - [ ] Widen from read only to add: "Sign in again" (IdP reauth) keeps add. **v0.6.363, account policy:** - [ ] Reset a local test account's password: the dialog says one-time, the row shows "New password pending", the person gets the bell, and the sign-in asks for a new password of at least 15 characters (frame K). - [ ] Rollback check: neither tag is shown before rolling back. **v0.6.364, storage restored (#904):** needs Nikola, and test data on a dogfood. - [ ] Declare a test restore in Storage & quotas (scope all, links paused, short notice). Then check: - a connected device gets 403 `device-paused`, and its resume shows the advice; - a sync app using the account password is paused and resumes from Connected devices; - every user sees the banner, and "Got it" only hides it; - the checklist's Accounts row lists accounts changed after the backup; - "Resume public links" works; - Close the checklist. - [ ] `cairnd restored --backup-time … --notice-days 14` while Cairn is stopped (Kubernetes: scale to 0, then a Job), then start Cairn. - [ ] On S3 (files-bao), with a copy of the bucket: roll the bucket back, and the suspicion banner appears within about a minute; "Not a restore" lifts it. - [ ] Incident box on a test realm: every local account must choose a new password at sign-in.
Author
Owner

v0.6.365, editor page around a refused save (#913): files dogfood, Collabora 26.04.5.2.1.

  • Open an .odt in a test folder and make its .cairn-versions read-only (0500). Type, then stop typing, and wait for the autosave. Expect:
    • within about 5 s, Cairn's popup with the reason, the rescue copy and "Show in folder" (which opens a new tab);
    • no Collabora "check your permissions" popup and no "changed in storage" dialog;
    • the chip and the banner.
  • Keep typing: no second popup, and "Last attempt at" updates.
  • Back shows the close warning naming the copy; a tab reload asks first.
  • Restore the permissions: the green "Saved…" row appears, then goes.
  • A lock by another user shows "{user} has locked…" with "Download a copy".
  • Optional: a file named <b>x</b>.odt that hits the size limit shows its name literally in Collabora's own popup.
**v0.6.365, editor page around a refused save (#913):** files dogfood, Collabora 26.04.5.2.1. - [ ] Open an .odt in a test folder and make its `.cairn-versions` read-only (0500). Type, then stop typing, and wait for the autosave. Expect: - within about 5 s, Cairn's popup with the reason, the rescue copy and "Show in folder" (which opens a new tab); - no Collabora "check your permissions" popup and no "changed in storage" dialog; - the chip and the banner. - [ ] Keep typing: no second popup, and "Last attempt at" updates. - [ ] Back shows the close warning naming the copy; a tab reload asks first. - [ ] Restore the permissions: the green "Saved…" row appears, then goes. - [ ] A lock by another user shows "{user} has locked…" with "Download a copy". - [ ] Optional: a file named `<b>x</b>.odt` that hits the size limit shows its name literally in Collabora's own popup.
Author
Owner

v0.6.366 live checks (#983, #973, #978, #981)

Needs Nikola signed in on both dogfoods (pods restarted, sessions reset).

I. Version and #983

  • Admin page shows v0.6.366 on files and files-bao.
  • Audit log, filter share-grant-retired: note how many lines on each (any grants on space content made before v0.6.215). "Shared with me" shows no /spaces/... entries.

J. #973 share roots (as a grantee with write on a shared folder and a shared file)

  • Row menu on the shared folder and on the shared file: no Delete, Move, Rename.
  • Finder/WebDAV: deleting the shared folder is refused; the text in the web says "This folder belongs to {owner}…".
  • Editing the shared file's content still saves.

K. #978 locks

  • Owner opens the file menu on a file locked by a grantee: "Break lock" asks first, naming the holder.
  • Grantee gets the bell item "{owner} removed your lock on {name}."
  • Grantee takes a lock, owner revokes the share: the lock is gone at once; audit shows lock-released.

L. #981 Activity

  • Delete a local test account, create it again with the same name: it sees no Activity from before (admin still does). Revert test state afterwards.
## v0.6.366 live checks (#983, #973, #978, #981) Needs Nikola signed in on both dogfoods (pods restarted, sessions reset). **I. Version and #983** - [ ] Admin page shows v0.6.366 on files and files-bao. - [ ] Audit log, filter `share-grant-retired`: note how many lines on each (any grants on space content made before v0.6.215). "Shared with me" shows no `/spaces/...` entries. **J. #973 share roots** (as a grantee with write on a shared folder and a shared file) - [ ] Row menu on the shared folder and on the shared file: no Delete, Move, Rename. - [ ] Finder/WebDAV: deleting the shared folder is refused; the text in the web says "This folder belongs to {owner}…". - [ ] Editing the shared file's content still saves. **K. #978 locks** - [ ] Owner opens the file menu on a file locked by a grantee: "Break lock" asks first, naming the holder. - [ ] Grantee gets the bell item "{owner} removed your lock on {name}." - [ ] Grantee takes a lock, owner revokes the share: the lock is gone at once; audit shows `lock-released`. **L. #981 Activity** - [ ] Delete a local test account, create it again with the same name: it sees no Activity from before (admin still does). Revert test state afterwards.
Author
Owner

v0.6.367 live checks (#996, #1001, #1034)

  • After the upgrade, audit filter share-link-bound / share-link-unbound: note counts. Administration › Sharing › Shares: any "Space not found" rows?
  • An existing link on space content still opens. Rename that space, wait for the folder move: the link still opens (before: 404).

N. #996 Keycloak "Signed JWT" (dogfood, sign-in client only)

  1. openssl req -x509 -newkey rsa:2048 -nodes -sha256 -days 365 -subj "/CN=cairn-kc-test" -keyout tls.key -out tls.crt
  2. Keycloak › realm › clients › Cairn sign-in client › Credentials › Client Authenticator Signed JWT (RS256); Keys › Use JWKS URL off › Import key › Certificate PEM › tls.crt.
  3. Mount the directory (Secret: kubectl create secret tls cairn-oidc-client --cert=tls.crt --key=tls.key), set CAIRN_OIDC_CLIENT_CERT_DIR, remove the sign-in client secret, restart. (GitOps change; I can prepare it.)
  • Sign-in works; Users & access › Connection › Client row shows the certificate, no Entra upload sentence.
  • Negative: secret and cert dir both set → start refused with a clear message.
  • Restore the secret afterwards if the dogfood should stay on it.

O. #996 Entra live test (needs a tenant; later)

  1. Tenant: Microsoft 365 Developer Program sandbox, or a free Azure account (comes with a Microsoft Entra ID Free tenant; Free has no group assignment, assign users directly).
  2. App registration: New registration "Cairn-test", single tenant, platform Web, redirect URI https://<cairn>/auth/callback (must equal auth.oidcRedirectUrl).
  3. Claims as directory-entra-id.md: optional claim auth_time (ID token); app roles cairn.user, cairn.admin assigned to test users; Enterprise app › Properties › Assignment required = Yes. Config: oidcIssuer https://login.microsoftonline.com/<tenant-id>/v2.0, oidcClientId <application-id>, oidcUsernameClaim preferred_username, oidcGroupsClaim roles, oidcAllowedGroups ["cairn.user","cairn.admin"], adminGroups ["cairn.admin"].
  4. Certificate: openssl req -x509 -newkey rsa:2048 -nodes -sha256 -days 365 -subj "/CN=cairn-entra-test" -keyout tls.key -out tls.crt
  5. Upload tls.crt: Certificates & secrets › Certificates › Upload certificate.
  6. Cairn: mount the directory, set auth.oidcClientCertDir / CAIRN_OIDC_CLIENT_CERT_DIR, remove the client secret, restart. Expect INFO oidc: client certificate loaded from auth.oidcClientCertDir.
  • Sign in as an assigned user works.
  • Client row: "Certificate CN=cairn-entra-test, SHA-256 thumbprint …, valid until …" + the upload sentence. Cairn's thumbprint = openssl x509 -in tls.crt -outform der | openssl dgst -sha256 -binary | openssl base64 -A | tr '+/' '-_' | tr -d '='. The portal shows SHA-1 (openssl x509 -in tls.crt -noout -fingerprint -sha1); note if it shows SHA-256 instead.
  • /metrics: cairn_oidc_client_cert_expiry_days ≈ 365.
  • Failure: delete the cert in the portal, sign in → WARN oidc callback failed with invalid_client (AADSTS…); note the code.
  • Rotation: make a second cert, upload it first, then replace the files; within a minute INFO oidc: client certificate reloaded, sign-in still works; then remove the old cert in the portal.
  • Clean up the app registration afterwards.

P. #1001 (only if a dogfood runs Keycloak with oidcGroupsClaim other than groups)

  • A refused sign-in pauses that person's app passwords; they stay paused past 5 minutes; an accepted sign-in lifts them.
## v0.6.367 live checks (#996, #1001, #1034) ### M. #1034 space links (dogfood) - [ ] After the upgrade, audit filter `share-link-bound` / `share-link-unbound`: note counts. Administration › Sharing › Shares: any "Space not found" rows? - [ ] An existing link on space content still opens. Rename that space, wait for the folder move: the link still opens (before: 404). ### N. #996 Keycloak "Signed JWT" (dogfood, sign-in client only) 1. `openssl req -x509 -newkey rsa:2048 -nodes -sha256 -days 365 -subj "/CN=cairn-kc-test" -keyout tls.key -out tls.crt` 2. Keycloak › realm › clients › Cairn sign-in client › Credentials › Client Authenticator **Signed JWT** (RS256); Keys › Use JWKS URL off › Import key › Certificate PEM › `tls.crt`. 3. Mount the directory (Secret: `kubectl create secret tls cairn-oidc-client --cert=tls.crt --key=tls.key`), set `CAIRN_OIDC_CLIENT_CERT_DIR`, remove the sign-in client secret, restart. (GitOps change; I can prepare it.) - [ ] Sign-in works; Users & access › Connection › Client row shows the certificate, no Entra upload sentence. - [ ] Negative: secret and cert dir both set → start refused with a clear message. - Restore the secret afterwards if the dogfood should stay on it. ### O. #996 Entra live test (needs a tenant; later) 1. **Tenant:** Microsoft 365 Developer Program sandbox, or a free Azure account (comes with a Microsoft Entra ID Free tenant; Free has no group assignment, assign users directly). 2. **App registration:** New registration "Cairn-test", single tenant, platform **Web**, redirect URI `https://<cairn>/auth/callback` (must equal `auth.oidcRedirectUrl`). 3. **Claims** as directory-entra-id.md: optional claim `auth_time` (ID token); app roles `cairn.user`, `cairn.admin` assigned to test users; Enterprise app › Properties › Assignment required = Yes. Config: `oidcIssuer https://login.microsoftonline.com/<tenant-id>/v2.0`, `oidcClientId <application-id>`, `oidcUsernameClaim preferred_username`, `oidcGroupsClaim roles`, `oidcAllowedGroups ["cairn.user","cairn.admin"]`, `adminGroups ["cairn.admin"]`. 4. **Certificate:** `openssl req -x509 -newkey rsa:2048 -nodes -sha256 -days 365 -subj "/CN=cairn-entra-test" -keyout tls.key -out tls.crt` 5. **Upload** `tls.crt`: Certificates & secrets › Certificates › Upload certificate. 6. **Cairn:** mount the directory, set `auth.oidcClientCertDir` / `CAIRN_OIDC_CLIENT_CERT_DIR`, remove the client secret, restart. Expect INFO `oidc: client certificate loaded from auth.oidcClientCertDir`. - [ ] Sign in as an assigned user works. - [ ] Client row: "Certificate CN=cairn-entra-test, SHA-256 thumbprint …, valid until …" + the upload sentence. Cairn's thumbprint = `openssl x509 -in tls.crt -outform der | openssl dgst -sha256 -binary | openssl base64 -A | tr '+/' '-_' | tr -d '='`. The portal shows SHA-1 (`openssl x509 -in tls.crt -noout -fingerprint -sha1`); note if it shows SHA-256 instead. - [ ] `/metrics`: `cairn_oidc_client_cert_expiry_days` ≈ 365. - [ ] Failure: delete the cert in the portal, sign in → WARN `oidc callback failed` with `invalid_client (AADSTS…)`; note the code. - [ ] Rotation: make a second cert, **upload it first**, then replace the files; within a minute INFO `oidc: client certificate reloaded`, sign-in still works; then remove the old cert in the portal. - Clean up the app registration afterwards. ### P. #1001 (only if a dogfood runs Keycloak with `oidcGroupsClaim` other than `groups`) - [ ] A refused sign-in pauses that person's app passwords; they stay paused past 5 minutes; an accepted sign-in lifts them.
Author
Owner

v0.6.368 live checks (#1026, #1024)

Q. #1026 Remove partial copy (S3 dogfood)

  • If a space row shows a half-copy line: open "Remove partial copy…", check the counts and the kept list, type the folder name, run it; the row afterwards and the audit line space-half-copy-removed.
  • Otherwise skip until one appears (an S3 rename that gets interrupted produces one).

R. #1024 Empty states (any instance)

  • An empty folder, Favorites, Recents, Deleted files, an empty search with filters (Clear filters), the bell with no notifications: texts read naturally in EN and DE.
  • German admin tab reads "Spaces".
## v0.6.368 live checks (#1026, #1024) ### Q. #1026 Remove partial copy (S3 dogfood) - [ ] If a space row shows a half-copy line: open "Remove partial copy…", check the counts and the kept list, type the folder name, run it; the row afterwards and the audit line `space-half-copy-removed`. - [ ] Otherwise skip until one appears (an S3 rename that gets interrupted produces one). ### R. #1024 Empty states (any instance) - [ ] An empty folder, Favorites, Recents, Deleted files, an empty search with filters (Clear filters), the bell with no notifications: texts read naturally in EN and DE. - [ ] German admin tab reads "Spaces".
Author
Owner

v0.6.369 live checks (#654)

S. Departed users

  • Users & access › Departed users lists departed people (if any) with home, shares, spaces, seat; the old Sharing › Departed owners link lands here.
  • Open a person: panel actions match the mockup; during an IdP outage every action is disabled.
  • On a test departed account with a few files (one held, one shared by someone else): Delete unread files shows the right counts, type the username, run. Held and shared items stay; the rest is in that person's Deleted files (admin purge view filtered by owner); audit owner-delete-unread. Revert test state afterwards (restore from Deleted files or purge the test account's leftovers).
## v0.6.369 live checks (#654) ### S. Departed users - [ ] Users & access › Departed users lists departed people (if any) with home, shares, spaces, seat; the old Sharing › Departed owners link lands here. - [ ] Open a person: panel actions match the mockup; during an IdP outage every action is disabled. - [ ] On a test departed account with a few files (one held, one shared by someone else): Delete unread files shows the right counts, type the username, run. Held and shared items stay; the rest is in that person's Deleted files (admin purge view filtered by owner); audit `owner-delete-unread`. Revert test state afterwards (restore from Deleted files or purge the test account's leftovers).
Author
Owner

v0.6.370 live checks (#1035 sharing from spaces)

Needs two accounts: a space owner and a non-member (and ideally a write member).

T. Owner

  • Space header › Sharing: Settings shows "Only the owner" (default). Shares tab empty.
  • Share a file inside the space with the non-member (Can view, Until 30 days): the pane says the share belongs to the space; the row shows "Shared outside (1)".
  • Try to share the space root with a person: refused with "This is the whole space… add them as a member".
  • Copy link on the space root: whole-space confirm; the link row shows "Whole space".

U. Recipient

  • Shared with me shows the item under " · Space", shared by the owner; no /spaces path visible; can open it; cannot share it further (no link, no share).

V. Member with write

  • With "Only the owner": the pane says only the owner shares. Switch Settings to "members with Read + write": the member can now share; the owner gets the bell.

W. Lifecycle

  • Move the shared item to another folder inside the space: the recipient still has it. Delete it: gone for the recipient; restore: back.
  • Finder/Explorer: move the shared item out of the space: refused.
  • Revoke from the owner's Shares tab; CSV export downloads.
    Revert test state afterwards (revoke shares and links, reset the rule).
## v0.6.370 live checks (#1035 sharing from spaces) Needs two accounts: a space owner and a non-member (and ideally a write member). ### T. Owner - [ ] Space header › Sharing: Settings shows "Only the owner" (default). Shares tab empty. - [ ] Share a file inside the space with the non-member (Can view, Until 30 days): the pane says the share belongs to the space; the row shows "Shared outside (1)". - [ ] Try to share the space root with a person: refused with "This is the whole space… add them as a member". - [ ] Copy link on the space root: whole-space confirm; the link row shows "Whole space". ### U. Recipient - [ ] Shared with me shows the item under "<space> · Space", shared by the owner; no /spaces path visible; can open it; cannot share it further (no link, no share). ### V. Member with write - [ ] With "Only the owner": the pane says only the owner shares. Switch Settings to "members with Read + write": the member can now share; the owner gets the bell. ### W. Lifecycle - [ ] Move the shared item to another folder inside the space: the recipient still has it. Delete it: gone for the recipient; restore: back. - [ ] Finder/Explorer: move the shared item out of the space: refused. - [ ] Revoke from the owner's Shares tab; CSV export downloads. Revert test state afterwards (revoke shares and links, reset the rule).
Author
Owner

v0.6.371 live checks (#638 phase 0+1)

  • Keyboard only, in the file browser: Tab to Upload and press Enter (file picker opens); Tab to a row's ⋯, press Enter/ArrowDown, move with arrows, Escape returns focus to ⋯; same for New, sort, view, user menu.
  • Open a folder with Enter: focus stays usable and (with VoiceOver on macOS: Cmd+F5) the folder name and item count are announced.
  • The look is unchanged (toolbar, rows, menus, profile-picture buttons).
## v0.6.371 live checks (#638 phase 0+1) - [ ] Keyboard only, in the file browser: Tab to Upload and press Enter (file picker opens); Tab to a row's ⋯, press Enter/ArrowDown, move with arrows, Escape returns focus to ⋯; same for New, sort, view, user menu. - [ ] Open a folder with Enter: focus stays usable and (with VoiceOver on macOS: Cmd+F5) the folder name and item count are announced. - [ ] The look is unchanged (toolbar, rows, menus, profile-picture buttons).
Author
Owner

v0.6.372 (#1022 virus scanning): live checks for Nikola. Both dogfoods run v0.6.372 and the pods are healthy.

  1. Scanner not configured: Admin › Virus scanning reads "Not configured", with no rail dot. Uploads work as before.
  2. Optional, with a scanner (compose clamav service, or scan.clamd pointing at a clamd):
    • Check scanner shows the answer time, "EICAR detected" and the signature.
    • Upload an EICAR test file in the browser: you get the sticky banner "…was not uploaded: a virus was found", nothing is stored, the bell shows "File rejected", and the audit log has upload-rejected.
    • Do the same through WebDAV from Finder or rclone: the client gets 415, and Finder's 0-byte placeholder disappears.
    • Upload through a public drop link: you get the per-file red line, and the link's creator gets the bell.
  3. Scanner outage: stop clamd for more than 60 s, then upload. With fail closed you get 503 "not available", and the admin bell "not reachable" appears once. Start clamd again: you get "reachable again" with the count.
  4. Over the limit: set the over-limit policy to accept and upload a file larger than the limit. It is stored and listed under Unscanned files; delete and restore it, and it is still listed.
  5. Language: with the German UI, all scan texts appear in German (Sie, ß; de-CH uses ss).
**v0.6.372 (#1022 virus scanning): live checks for Nikola.** Both dogfoods run v0.6.372 and the pods are healthy. 1. **Scanner not configured:** Admin › Virus scanning reads "Not configured", with no rail dot. Uploads work as before. 2. **Optional, with a scanner** (compose clamav service, or `scan.clamd` pointing at a clamd): - Check scanner shows the answer time, "EICAR detected" and the signature. - Upload an EICAR test file in the browser: you get the sticky banner "…was not uploaded: a virus was found", nothing is stored, the bell shows "File rejected", and the audit log has `upload-rejected`. - Do the same through WebDAV from Finder or rclone: the client gets 415, and Finder's 0-byte placeholder disappears. - Upload through a public drop link: you get the per-file red line, and the link's creator gets the bell. 3. **Scanner outage:** stop clamd for more than 60 s, then upload. With fail closed you get 503 "not available", and the admin bell "not reachable" appears once. Start clamd again: you get "reachable again" with the count. 4. **Over the limit:** set the over-limit policy to accept and upload a file larger than the limit. It is stored and listed under Unscanned files; delete and restore it, and it is still listed. 5. **Language:** with the German UI, all scan texts appear in German (Sie, ß; de-CH uses ss).
Author
Owner

v0.6.373 (#638 session warning and accessibility phase 2): live checks for Nikola. Both dogfoods run v0.6.373.

  1. Warning: to see it without waiting 12 hours, leave a tab open across the end of a session. Five minutes before the end, the warning dialog appears with a countdown. Stay signed in gives one more hour, and the status line says until when. Esc turns the dialog into a banner, and clicking the backdrop does not close it.
  2. Two tabs: with the warning open in both, press Stay signed in in one. The other tab's warning closes, and the session gains one hour, not two.
  3. Signed out with the page kept: click Sign out in another tab, then delete a file in this tab. Dialog D appears, the delete waits, and after you sign in again it runs once. An upload that is running pauses and continues.
  4. Identity provider (Keycloak dogfood): Stay signed in is a single click, with no window.
  5. Look:
    • The light accent is slightly darker.
    • In a dark theme, danger buttons have dark text on a lighter red.
    • Field borders and switched-off toggles are visibly stronger.
    • The dark login page's button has dark text.
  6. Keyboard:
    • In a menu, typing a letter jumps to the matching item.
    • In the space picture crop, the arrow keys move the picture and + and - zoom.
    • Esc closes a tooltip, and the first Esc inside a dialog hides only the tooltip.
  7. Screen reader (optional): row menus are announced as "Actions for {name}". After a delete, focus lands on the next row.
**v0.6.373 (#638 session warning and accessibility phase 2): live checks for Nikola.** Both dogfoods run v0.6.373. 1. **Warning:** to see it without waiting 12 hours, leave a tab open across the end of a session. Five minutes before the end, the warning dialog appears with a countdown. Stay signed in gives one more hour, and the status line says until when. Esc turns the dialog into a banner, and clicking the backdrop does not close it. 2. **Two tabs:** with the warning open in both, press Stay signed in in one. The other tab's warning closes, and the session gains one hour, not two. 3. **Signed out with the page kept:** click Sign out in another tab, then delete a file in this tab. Dialog D appears, the delete waits, and after you sign in again it runs once. An upload that is running pauses and continues. 4. **Identity provider (Keycloak dogfood):** Stay signed in is a single click, with no window. 5. **Look:** - The light accent is slightly darker. - In a dark theme, danger buttons have dark text on a lighter red. - Field borders and switched-off toggles are visibly stronger. - The dark login page's button has dark text. 6. **Keyboard:** - In a menu, typing a letter jumps to the matching item. - In the space picture crop, the arrow keys move the picture and + and - zoom. - Esc closes a tooltip, and the first Esc inside a dialog hides only the tooltip. 7. **Screen reader (optional):** row menus are announced as "Actions for {name}". After a delete, focus lands on the next row.
Author
Owner

v0.6.374 (#896 mail server): live checks for Nikola. Both dogfoods run v0.6.374.

  1. Without mail: Basic settings › Mail server shows the "connect a mail server" steps with the buttons greyed out. Account settings › Profile shows the E-mail row with its source; on the Keycloak dogfood the address comes from the identity provider, so check that sign-in still works with the new email scope.
  2. With a mail server (any SMTP account or relay, set through the GitOps config and the secret):
    • Check connection shows the stages. Send a test e-mail to yourself and confirm it arrives in your language.
    • On a local account, use Edit e-mail address: both the old and the new address get a mail, the person gets the bell item, and the audit log has account-email-set.
    • Connect a device: the "device connected" mail arrives.
    • Point at a wrong host: after about 15 minutes the bell shows "mail server failing" and the Overview card turns amber. Fix it: "working again" appears.
  3. Microsoft 365 / Entra (when you have a tenant): follow mail.md › Microsoft 365, and the Entra page section 8 for xms_edov.
**v0.6.374 (#896 mail server): live checks for Nikola.** Both dogfoods run v0.6.374. 1. **Without mail:** Basic settings › Mail server shows the "connect a mail server" steps with the buttons greyed out. Account settings › Profile shows the E-mail row with its source; on the Keycloak dogfood the address comes from the identity provider, so check that sign-in still works with the new `email` scope. 2. **With a mail server** (any SMTP account or relay, set through the GitOps config and the secret): - Check connection shows the stages. Send a test e-mail to yourself and confirm it arrives in your language. - On a local account, use Edit e-mail address: both the old and the new address get a mail, the person gets the bell item, and the audit log has `account-email-set`. - Connect a device: the "device connected" mail arrives. - Point at a wrong host: after about 15 minutes the bell shows "mail server failing" and the Overview card turns amber. Fix it: "working again" appears. 3. **Microsoft 365 / Entra** (when you have a tenant): follow `mail.md` › Microsoft 365, and the Entra page section 8 for `xms_edov`.
Author
Owner

v0.6.375 (#1023 personal data export): live checks for Nikola. Both dogfoods run v0.6.375.

  1. Settings: Users & access › Personal data shows "README texts missing". Fill in controller, contact and purposes in EN and DE, then use Preview README.
  2. Self-service: go to Account › Your data and click "Request a copy…", then confirm with your password (or sign in again, for identity-provider accounts).
    • The progress list runs through the parts, then the copy shows as ready.
    • Download the ZIP. README.html opens in your language, manifest.json lists a hash for every file, and the CSV files open in a spreadsheet.
    • A second request the same day is refused until the 24 h are over.
  3. Admin export: in Accounts, use the row menu › "Export personal data…".
    • Choose Handover link with File contents. The code is shown once.
    • Open the link in a private window and type the code: the data ZIP and the content part download. Five wrong codes lock the link.
  4. Told: with "Tell the person" on, the person's bell shows the export and Your data lists it.
  5. Audit: the audit log under "Personal data" shows request, ready and download.
**v0.6.375 (#1023 personal data export): live checks for Nikola.** Both dogfoods run v0.6.375. 1. **Settings:** Users & access › Personal data shows "README texts missing". Fill in controller, contact and purposes in EN and DE, then use Preview README. 2. **Self-service:** go to Account › Your data and click "Request a copy…", then confirm with your password (or sign in again, for identity-provider accounts). - The progress list runs through the parts, then the copy shows as ready. - Download the ZIP. README.html opens in your language, `manifest.json` lists a hash for every file, and the CSV files open in a spreadsheet. - A second request the same day is refused until the 24 h are over. 3. **Admin export:** in Accounts, use the row menu › "Export personal data…". - Choose Handover link with File contents. The code is shown once. - Open the link in a private window and type the code: the data ZIP and the content part download. Five wrong codes lock the link. 4. **Told:** with "Tell the person" on, the person's bell shows the export and Your data lists it. 5. **Audit:** the audit log under "Personal data" shows request, ready and download.
Author
Owner

v0.6.376 live-check list (#835 links follow their file, #1045 purge verbs), both dogfoods are on v0.6.376.

  1. Personal: make a public link on a file, rename the file, open the link: it serves the file.
  2. Move the file into a subfolder of Personal: the link still serves.
  3. Move the linked file into a space: the confirm lists the link that ends; Cancel keeps it in Personal; confirming moves it and the link answers 404.
  4. Delete the linked file: the confirm says its shares and links pause; the link answers 404; Deleted files shows the "paused" chip; restore: the link serves again.
  5. Delete forever: the confirm lists what ends; afterwards the link is gone from Shares.
  6. Path reuse: link on a.pdf, rename it to b.pdf, upload a new a.pdf: the link serves b.pdf (followed), never the new a.pdf.
  7. Outside Cairn (server disk or s3), replace the linked file: Shares shows Item replaced, the owner gets a bell, the link answers 404.
  8. Shares view: state chips (Paused in Deleted files, Item missing, Item replaced); Copy link dimmed for those.
  9. Admin: transfer a departed user's files to someone: the dialog shows links paused for the new owner (or "Keep the public links working"); the new owner gets a bell and can Turn on each link.
  10. Admin Shares: chips "Paused: handed over · n" and "Item replaced · n"; export carries ?status=.
  11. Audit (#1045): with access logging off, a Delete forever still writes manual-purge.
  12. OpenBao dogfood: the start log shows the expected stamp WARN for pre-release items (#1058).
**v0.6.376 live-check list (#835 links follow their file, #1045 purge verbs)**, both dogfoods are on v0.6.376. 1. Personal: make a public link on a file, rename the file, open the link: it serves the file. 2. Move the file into a subfolder of Personal: the link still serves. 3. Move the linked file into a space: the confirm lists the link that ends; Cancel keeps it in Personal; confirming moves it and the link answers 404. 4. Delete the linked file: the confirm says its shares and links pause; the link answers 404; Deleted files shows the "paused" chip; restore: the link serves again. 5. Delete forever: the confirm lists what ends; afterwards the link is gone from Shares. 6. Path reuse: link on `a.pdf`, rename it to `b.pdf`, upload a new `a.pdf`: the link serves `b.pdf` (followed), never the new `a.pdf`. 7. Outside Cairn (server disk or s3), replace the linked file: Shares shows **Item replaced**, the owner gets a bell, the link answers 404. 8. Shares view: state chips (Paused in Deleted files, Item missing, Item replaced); Copy link dimmed for those. 9. Admin: transfer a departed user's files to someone: the dialog shows links paused for the new owner (or "Keep the public links working"); the new owner gets a bell and can Turn on each link. 10. Admin Shares: chips "Paused: handed over · n" and "Item replaced · n"; export carries `?status=`. 11. Audit (#1045): with access logging off, a Delete forever still writes `manual-purge`. 12. OpenBao dogfood: the start log shows the expected stamp WARN for pre-release items (#1058).
Author
Owner

v0.6.377 live-check list (#975 conflict prompt), both dogfoods are on v0.6.377.

  1. Move a file into a folder that already has that name: the dialog shows both cards (size, date, Newer/Older), Keep both has focus; Keep both gives "name 2.pdf".
  2. Same with Replace: the earlier file appears under Versions of the destination.
  3. Bulk move of three files, two names taken: two questions first, "apply to the others" works, one result line.
  4. Copy into the same folder: no question, "Copied as … 2".
  5. Upload two files whose names exist: the questions come before any upload starts; a large file (TUS) with Keep both lands as "… 2".
  6. Restore from Deleted files next to a taken "a 2.pdf": restores as "a 3.pdf".
  7. A file on legal hold or locked by someone else: Replace is disabled with the reason.
  8. Move a linked Personal file into a space where the name is taken: name question first, then the shares confirm.
  9. Admin › Audit: a replace shows conflict=replace replaced=1 surface=rest-rename after the path.
  10. Phone width: cards stack, buttons stack full width; dark theme readable.
**v0.6.377 live-check list (#975 conflict prompt)**, both dogfoods are on v0.6.377. 1. Move a file into a folder that already has that name: the dialog shows both cards (size, date, Newer/Older), Keep both has focus; Keep both gives "name 2.pdf". 2. Same with Replace: the earlier file appears under Versions of the destination. 3. Bulk move of three files, two names taken: two questions first, "apply to the others" works, one result line. 4. Copy into the same folder: no question, "Copied as … 2". 5. Upload two files whose names exist: the questions come before any upload starts; a large file (TUS) with Keep both lands as "… 2". 6. Restore from Deleted files next to a taken "a 2.pdf": restores as "a 3.pdf". 7. A file on legal hold or locked by someone else: Replace is disabled with the reason. 8. Move a linked Personal file into a space where the name is taken: name question first, then the shares confirm. 9. Admin › Audit: a replace shows `conflict=replace replaced=1 surface=rest-rename` after the path. 10. Phone width: cards stack, buttons stack full width; dark theme readable.
Author
Owner

v0.6.378 live-check list (#1033 Leave share), both dogfoods are on v0.6.378. Needs two accounts (owner A, recipient B).

  1. A shares a folder with B. B opens /shared, row menu › Leave share: the confirm lists B's devices bound to it and B's locks there; Leave: the row goes, the status line offers Undo.
  2. Undo: the row is back, same role.
  3. Leave again, then Shares › Shared with me › Left: the entry shows "Left on …"; Rejoin brings it back to Current.
  4. A's Shares list and the item's Shares pane show the quiet "Left on …" label for B; admin Shares has the "Left by recipient" chip and the CSV leftAt.
  5. A changes B's role while B has left: B stays out (label kept).
  6. A group share: B leaves, a second group member still sees it.
  7. Space share: B leaves an item shared from a space; the space's other members are unaffected.
  8. A device of B bound to that share shows "You left …" with Rejoin in Account › Devices.
  9. Leave from Favourites (a starred share root): the row drops; Undo brings it back.
  10. WebDAV: after leaving, /shared/<A>/<name> answers 404 for B; the share-root refusal text names Leave share.
  11. Admin › Audit: share-leave and share-rejoin with the entry's path.
**v0.6.378 live-check list (#1033 Leave share)**, both dogfoods are on v0.6.378. Needs two accounts (owner A, recipient B). 1. A shares a folder with B. B opens /shared, row menu › Leave share: the confirm lists B's devices bound to it and B's locks there; Leave: the row goes, the status line offers Undo. 2. Undo: the row is back, same role. 3. Leave again, then Shares › Shared with me › Left: the entry shows "Left on …"; Rejoin brings it back to Current. 4. A's Shares list and the item's Shares pane show the quiet "Left on …" label for B; admin Shares has the "Left by recipient" chip and the CSV `leftAt`. 5. A changes B's role while B has left: B stays out (label kept). 6. A group share: B leaves, a second group member still sees it. 7. Space share: B leaves an item shared from a space; the space's other members are unaffected. 8. A device of B bound to that share shows "You left …" with Rejoin in Account › Devices. 9. Leave from Favourites (a starred share root): the row drops; Undo brings it back. 10. WebDAV: after leaving, `/shared/<A>/<name>` answers 404 for B; the share-root refusal text names Leave share. 11. Admin › Audit: `share-leave` and `share-rejoin` with the entry's path.
Author
Owner

v0.6.379 live-check list (#1039 member end dates), both dogfoods are on v0.6.379. Use an app-owned space (enc dogfood runs power mode for IdP spaces, so try a Cairn-created space where available).

  1. Members › Until for a member: 30 days shows "Until: 30 days · "; the row chip appears in the last 7 days.
  2. Choose a date… with today: a confirm asks first; Cancel restores the select.
  3. Set an end date one or two minutes ahead (today, then wait past midnight is not needed: use the API with an until a minute ahead): within a minute after it passes the member loses access, the row moves to Recently ended with Add again, the bell says the membership ended (with shares kept and locks released counts).
  4. The member's bell shows "Your access ends in N days" during the last 7 days, and the space header shows the date.
  5. As a temporary member, share an item: the Until select starts at "End of your membership"; Never is gone; a later date is refused with the end-date text.
  6. Members › Export CSV downloads members--.csv with until and set-by.
  7. Admin › Spaces: "n with an end date · n ending within 7 days" and the panel with Members and Export CSV.
  8. Admin › Audit: space-member-until (from/to) and space-member-expired read as sentences.
  9. Restore a deleted space whose member's date passed meanwhile: that member is not restored and is listed under Recently ended (cause restore).
**v0.6.379 live-check list (#1039 member end dates)**, both dogfoods are on v0.6.379. Use an app-owned space (enc dogfood runs power mode for IdP spaces, so try a Cairn-created space where available). 1. Members › Until for a member: 30 days shows "Until: 30 days · <date>"; the row chip appears in the last 7 days. 2. Choose a date… with today: a confirm asks first; Cancel restores the select. 3. Set an end date one or two minutes ahead (today, then wait past midnight is not needed: use the API with an `until` a minute ahead): within a minute after it passes the member loses access, the row moves to Recently ended with Add again, the bell says the membership ended (with shares kept and locks released counts). 4. The member's bell shows "Your access ends in N days" during the last 7 days, and the space header shows the date. 5. As a temporary member, share an item: the Until select starts at "End of your membership"; Never is gone; a later date is refused with the end-date text. 6. Members › Export CSV downloads members-<space>-<date>.csv with until and set-by. 7. Admin › Spaces: "n with an end date · n ending within 7 days" and the panel with Members and Export CSV. 8. Admin › Audit: `space-member-until` (from/to) and `space-member-expired` read as sentences. 9. Restore a deleted space whose member's date passed meanwhile: that member is not restored and is listed under Recently ended (cause restore).
Author
Owner

v0.6.380 live-check list (#929 Find spaces), both dogfoods are on v0.6.380. Find spaces needs app-owned spaces; on a power-mode instance the policy box says it is not available (expected).

  1. Admin › Sharing › Policy: the Find spaces box is off; switch on for everyone, Save.
  2. Owner: Members › Requests tab appears; turn on "List in Find spaces"; the space shows the Listed badge in the overview.
  3. Another person: Spaces › Find spaces lists it; search by a word of the description; Ask to join with a short message: "Request for … sent", the row says pending.
  4. Owner bell: "1 request to join …" with the message; Members › Requests shows the person's display name; Approve (Read only): the person becomes a member and gets a bell.
  5. Second request from another person: Decline with a note: their bell shows the note and the ask-again date; Find spaces shows the row disabled until then.
  6. Withdraw a pending request: no confirm, "Request … withdrawn."
  7. Owner turns listing off with a request waiting: asks first; the request closes, the requester's bell says the space no longer takes requests.
  8. Admin turns the policy off with listed spaces: confirm lists counts; afterwards no Requests tab and nothing listed.
  9. Admin › Spaces: Listed filter and "n requests waiting" while on.
  10. Data export of the requester: spaces/space-requests has the request rows.
  11. A non-member opening an unlisted space id through the API gets the plain 404.
**v0.6.380 live-check list (#929 Find spaces)**, both dogfoods are on v0.6.380. Find spaces needs app-owned spaces; on a power-mode instance the policy box says it is not available (expected). 1. Admin › Sharing › Policy: the Find spaces box is off; switch on for everyone, Save. 2. Owner: Members › Requests tab appears; turn on "List in Find spaces"; the space shows the Listed badge in the overview. 3. Another person: Spaces › Find spaces lists it; search by a word of the description; Ask to join with a short message: "Request for … sent", the row says pending. 4. Owner bell: "1 request to join …" with the message; Members › Requests shows the person's display name; Approve (Read only): the person becomes a member and gets a bell. 5. Second request from another person: Decline with a note: their bell shows the note and the ask-again date; Find spaces shows the row disabled until then. 6. Withdraw a pending request: no confirm, "Request … withdrawn." 7. Owner turns listing off with a request waiting: asks first; the request closes, the requester's bell says the space no longer takes requests. 8. Admin turns the policy off with listed spaces: confirm lists counts; afterwards no Requests tab and nothing listed. 9. Admin › Spaces: Listed filter and "n requests waiting" while on. 10. Data export of the requester: `spaces/space-requests` has the request rows. 11. A non-member opening an unlisted space id through the API gets the plain 404.
Author
Owner

v0.6.381 and v0.6.382 live-check list (#1058 custody stamp, #1059 feed vs item guard), both dogfoods are on v0.6.382.

Already checked: the OpenBao dogfood's start log now reads shares: stamped item ids on existing shares and links records=5 stamped=4 missing=1 deferred=0. The per-start stamp WARN from v0.6.376 is gone.

  1. OpenBao dogfood: as the owner of a link on a file from before v0.6.376, replace that file outside Cairn (in the bucket). The link should now answer 404, and Shares should show Item replaced. Before v0.6.381 it served by path.
  2. Keycloak-custody dogfood: restart the pod twice. The stamp line appears once and is not repeated.
  3. #1059: grant a file to B, then replace it outside Cairn. B's sync client (or GET /api/v1/changes as B) no longer lists changes at that /shared path. If B has a second grant with the same name that still serves, it is listed and opens.
  4. Shared with me: a replaced grant does not show; a same-named sibling grant that serves does show.
**v0.6.381 and v0.6.382 live-check list (#1058 custody stamp, #1059 feed vs item guard)**, both dogfoods are on v0.6.382. Already checked: the OpenBao dogfood's start log now reads `shares: stamped item ids on existing shares and links records=5 stamped=4 missing=1 deferred=0`. The per-start stamp WARN from v0.6.376 is gone. 1. OpenBao dogfood: as the owner of a link on a file from before v0.6.376, replace that file outside Cairn (in the bucket). The link should now answer 404, and Shares should show **Item replaced**. Before v0.6.381 it served by path. 2. Keycloak-custody dogfood: restart the pod twice. The stamp line appears once and is not repeated. 3. #1059: grant a file to B, then replace it outside Cairn. B's sync client (or `GET /api/v1/changes` as B) no longer lists changes at that `/shared` path. If B has a second grant with the same name that still serves, it is listed and opens. 4. Shared with me: a replaced grant does not show; a same-named sibling grant that serves does show.
Author
Owner

v0.6.383 live-check list (#1063 change feed on grant and revoke), both dogfoods are on v0.6.383. Needs a sync client or GET /api/v1/changes as person B.

  1. A shares a folder with B: B's next poll has a write entry for /shared/<A>/<name> with why: "access".
  2. A revokes it: B's next poll has a delete entry with why: "access".
  3. A renames the shared folder: B's next poll has a move with from.
  4. Two shares of the same name from A: renaming one renumbers the other ("(2)"), and the feed reports it as a move.
  5. Group share: add B to the group in the IdP. The next poll has the root.
  6. B leaves a share (#1033): the feed reports a delete, and Rejoin reports a write.
  7. An unchanged poll answers "changes":[].
  8. Restart the pod: old cursors still work, and the first poll after the restart adopts the view without false entries.
**v0.6.383 live-check list (#1063 change feed on grant and revoke)**, both dogfoods are on v0.6.383. Needs a sync client or `GET /api/v1/changes` as person B. 1. A shares a folder with B: B's next poll has a `write` entry for `/shared/<A>/<name>` with `why: "access"`. 2. A revokes it: B's next poll has a `delete` entry with `why: "access"`. 3. A renames the shared folder: B's next poll has a `move` with `from`. 4. Two shares of the same name from A: renaming one renumbers the other ("(2)"), and the feed reports it as a move. 5. Group share: add B to the group in the IdP. The next poll has the root. 6. B leaves a share (#1033): the feed reports a `delete`, and Rejoin reports a `write`. 7. An unchanged poll answers `"changes":[]`. 8. Restart the pod: old cursors still work, and the first poll after the restart adopts the view without false entries.
Author
Owner

v0.6.384 live-check list (#1055 tags, favourites and recents follow their item), both dogfoods are on v0.6.384.

  1. Two accounts, each with /home/Offer.pdf. A tags theirs: B sees none of A's tags (row, details pane, tag: search, the suggestions).
  2. Tag and star a file, then rename it in the web app, rename it again over WebDAV (Finder), and move it into a subfolder. Tags and star follow each time. A new file under the old name has neither.
  3. Open a tagged .docx in the editor and save (Office temp files): the tags stay on the document.
  4. Copy a tagged file: the copy has the tags. Overwrite a tagged file with another tagged file: the tags are joined, the destination's first.
  5. Delete a tagged, starred file: it leaves Favorites. Restore: back, also under a numbered name if the name is taken.
  6. Share a folder with B; B tags a file inside. A sees B's tag (same item). B's own tags on B's files stay private.
  7. A viewer (Can view) cannot change tags ("read-only access"). An Add files only device cannot tag existing files.
  8. Rename a space: tags and stars inside it follow.
  9. Admin transfer of a departed account's files: tags and other people's stars follow.
**v0.6.384 live-check list (#1055 tags, favourites and recents follow their item)**, both dogfoods are on v0.6.384. 1. Two accounts, each with `/home/Offer.pdf`. A tags theirs: B sees none of A's tags (row, details pane, `tag:` search, the suggestions). 2. Tag and star a file, then rename it in the web app, rename it again over WebDAV (Finder), and move it into a subfolder. Tags and star follow each time. A new file under the old name has neither. 3. Open a tagged .docx in the editor and save (Office temp files): the tags stay on the document. 4. Copy a tagged file: the copy has the tags. Overwrite a tagged file with another tagged file: the tags are joined, the destination's first. 5. Delete a tagged, starred file: it leaves Favorites. Restore: back, also under a numbered name if the name is taken. 6. Share a folder with B; B tags a file inside. A sees B's tag (same item). B's own tags on B's files stay private. 7. A viewer (Can view) cannot change tags ("read-only access"). An Add files only device cannot tag existing files. 8. Rename a space: tags and stars inside it follow. 9. Admin transfer of a departed account's files: tags and other people's stars follow.
Author
Owner

v0.6.385 live-check list (#1057 federated shares out follow their item), both dogfoods are on v0.6.385. You need the lab Nextcloud as the other server (allow-list it under Federated).

  1. Share a file from Personal with an address on the lab Nextcloud. Accept it there; the Federated › You shared row shows it as accepted.
  2. Rename the file in Cairn, then move it into a subfolder. The Nextcloud side keeps reading it (see #1070 if it cannot open it at all).
  3. Upload a new file under the old name. The Nextcloud side never sees it.
  4. Delete the file. The confirm says the other server is not told and the share shows as unavailable there. Restore it: the share serves again.
  5. Delete for good. The status line says the other server is told. On Nextcloud the share disappears.
  6. Move a shared file into a space. The confirm lists the federated share. Confirm, and the share ends with "Ended: moved out of your Personal folder".
  7. Switch OCM off and end a share. The row shows "not told", and Nextcloud gets nothing.
  8. Admin: transfer a departed account that has a federated share. The dialog shows the federated row. The share is paused for the new owner, who can Turn on.
  9. Replace a shared file outside Cairn. The Shares view shows Item replaced, the owner gets the bell, and the other server gets 404.
**v0.6.385 live-check list (#1057 federated shares out follow their item)**, both dogfoods are on v0.6.385. You need the lab Nextcloud as the other server (allow-list it under Federated). 1. Share a file from Personal with an address on the lab Nextcloud. Accept it there; the Federated › You shared row shows it as accepted. 2. Rename the file in Cairn, then move it into a subfolder. The Nextcloud side keeps reading it (see #1070 if it cannot open it at all). 3. Upload a new file under the old name. The Nextcloud side never sees it. 4. Delete the file. The confirm says the other server is not told and the share shows as unavailable there. Restore it: the share serves again. 5. Delete for good. The status line says the other server is told. On Nextcloud the share disappears. 6. Move a shared file into a space. The confirm lists the federated share. Confirm, and the share ends with "Ended: moved out of your Personal folder". 7. Switch OCM off and end a share. The row shows "not told", and Nextcloud gets nothing. 8. Admin: transfer a departed account that has a federated share. The dialog shows the federated row. The share is paused for the new owner, who can Turn on. 9. Replace a shared file outside Cairn. The Shares view shows Item replaced, the owner gets the bell, and the other server gets 404.
Author
Owner

v0.6.386 live-check list (#1056 editor sessions follow the document, folder locks), both dogfoods are on v0.6.386 with Collabora.

  1. Open a .docx in the editor and rename it in another tab. The editor keeps saving, and its title shows the new name within 30 s. Then rename its parent folder: it still saves.
  2. Two people edit one document, one from Personal and one through Shares. Both save. Administration › Locks shows one session lock.
  3. Revoke the second person's share while they edit. Their page shows the end banner ("moved to a place you cannot open"), and Download a copy works.
  4. Delete the open document. The end banner says deleted, and Open folder goes to the parent.
  5. Replace the open document outside Cairn. The editor ends as replaced.
  6. While the lock holder is still editing, take away their access. The other editor keeps saving, and Locks shows them as holder.
  7. Collabora checks: its own name field after a rename, whether it passes the lock on when the first opener leaves, and how it shows a 404 on a callback.
  8. Locks: lock a file as B in A's shared folder. A cannot delete or move the folder (the message names B). A can move a folder above a document someone has open in the editor; the editor follows.
  9. Admin Delete unread files or a space delete with a lock inside: the lock is broken, and the holder gets the bell.
  10. On s3 (files-bao), rename a large folder while saving a file inside it. The save is not lost.
  11. A sync client sees id on write and move entries; WebDAV PROPFIND for cairn:item-id answers it.
**v0.6.386 live-check list (#1056 editor sessions follow the document, folder locks)**, both dogfoods are on v0.6.386 with Collabora. 1. Open a .docx in the editor and rename it in another tab. The editor keeps saving, and its title shows the new name within 30 s. Then rename its parent folder: it still saves. 2. Two people edit one document, one from Personal and one through Shares. Both save. Administration › Locks shows one session lock. 3. Revoke the second person's share while they edit. Their page shows the end banner ("moved to a place you cannot open"), and Download a copy works. 4. Delete the open document. The end banner says deleted, and Open folder goes to the parent. 5. Replace the open document outside Cairn. The editor ends as replaced. 6. While the lock holder is still editing, take away their access. The other editor keeps saving, and Locks shows them as holder. 7. Collabora checks: its own name field after a rename, whether it passes the lock on when the first opener leaves, and how it shows a 404 on a callback. 8. Locks: lock a file as B in A's shared folder. A cannot delete or move the folder (the message names B). A can move a folder above a document someone has open in the editor; the editor follows. 9. Admin Delete unread files or a space delete with a lock inside: the lock is broken, and the holder gets the bell. 10. On s3 (files-bao), rename a large folder while saving a file inside it. The save is not lost. 11. A sync client sees `id` on write and move entries; WebDAV PROPFIND for `cairn:item-id` answers it.
Author
Owner

v0.6.387 (#1054, a name used again keeps the earlier account apart)

On cairn-enc (Keycloak) unless noted. Use test accounts only.

  1. Re-created Keycloak account: as nikola-test, make a public link and a share to a second test user. In Keycloak delete nikola-test and create it again with the same username. Without signing in: the link answers 404, the recipient no longer sees the share. Administration › Users & access › Departed users shows Earlier account of nikola-test.
  2. Personal folder kept apart: sign in as the new nikola-test. Personal shows "Your Personal folder is not ready yet"; Spaces and Shared with me work.
  3. Give: as admin, open the earlier row › Give to the current nikola-test… (leave Keep the public links working unticked). The new account gets Your Personal folder is ready; the share reaches the recipient again; the link is paused until turned on.
  4. Treat as a different person: Accounts › a local test account › Treat as a different person…; its earlier shares move to Departed users; the account keeps signing in.
  5. Credentials: an app password created before the re-creation in step 1 is refused (WebDAV 401).
  6. Setting: Users & access › Access › When a name is used again shows the two options; switching writes a settings-change audit line with key=nameReuse.
  7. Optional (cairn-openbao): repeat 1 to 3 to see the same on OpenBao custody.

Expected logs: INFO account start moved with via=directory or via=signin; no ERROR lines.

### v0.6.387 (#1054, a name used again keeps the earlier account apart) On cairn-enc (Keycloak) unless noted. Use test accounts only. 1. **Re-created Keycloak account:** as `nikola-test`, make a public link and a share to a second test user. In Keycloak delete `nikola-test` and create it again with the same username. Without signing in: the link answers 404, the recipient no longer sees the share. Administration › Users & access › Departed users shows **Earlier account of nikola-test**. 2. **Personal folder kept apart:** sign in as the new `nikola-test`. Personal shows "Your Personal folder is not ready yet"; Spaces and Shared with me work. 3. **Give:** as admin, open the earlier row › **Give to the current nikola-test…** (leave Keep the public links working unticked). The new account gets **Your Personal folder is ready**; the share reaches the recipient again; the link is paused until turned on. 4. **Treat as a different person:** Accounts › a local test account › Treat as a different person…; its earlier shares move to Departed users; the account keeps signing in. 5. **Credentials:** an app password created before the re-creation in step 1 is refused (WebDAV 401). 6. **Setting:** Users & access › Access › When a name is used again shows the two options; switching writes a `settings-change` audit line with `key=nameReuse`. 7. Optional (cairn-openbao): repeat 1 to 3 to see the same on OpenBao custody. Expected logs: INFO `account start moved` with `via=directory` or `via=signin`; no ERROR lines.
Author
Owner

v0.6.388 (#1070 quick fix: Nextcloud keeps paused federated shares)

With the lab Nextcloud as the recipient server:

  1. From cairn-enc, send a federated share to a Nextcloud test user and accept it there.
  2. Live: on the Nextcloud side the share opens. Its periodic check (/index.php/apps/files_sharing/shareinfo) appears in Cairn's logs at DEBUG only.
  3. Paused: move the shared file to Deleted files in Cairn (the share pauses). Wait for Nextcloud's next check, or run occ sharing:cleanup-remote-storages on the Nextcloud host. The share must stay in the recipient's list, showing as unavailable, not removed.
  4. Turn on: restore the file. The share opens again on Nextcloud without being sent again.
  5. Ended: end the share in Cairn. Nextcloud removes it at its next check.

Expected: no 401 or 403 answers to Nextcloud in Cairn's logs; no WARN/ERROR.

### v0.6.388 (#1070 quick fix: Nextcloud keeps paused federated shares) With the lab Nextcloud as the recipient server: 1. From cairn-enc, send a federated share to a Nextcloud test user and accept it there. 2. **Live:** on the Nextcloud side the share opens. Its periodic check (`/index.php/apps/files_sharing/shareinfo`) appears in Cairn's logs at DEBUG only. 3. **Paused:** move the shared file to Deleted files in Cairn (the share pauses). Wait for Nextcloud's next check, or run `occ sharing:cleanup-remote-storages` on the Nextcloud host. The share must **stay** in the recipient's list, showing as unavailable, not removed. 4. **Turn on:** restore the file. The share opens again on Nextcloud without being sent again. 5. **Ended:** end the share in Cairn. Nextcloud removes it at its next check. Expected: no 401 or 403 answers to Nextcloud in Cairn's logs; no WARN/ERROR.
Author
Owner

v0.6.389 (#1074, item id on the change feed's /shared roots)

On cairn-enc, with two test accounts A and B.

  1. As B, take a change-feed cursor: GET /api/v1/changes?cursor=now (Settings › Devices token or the browser session).
  2. As A, share a folder Probe with B. B polls with the cursor: the /shared/<A>/Probe entry has why: "access" and an itemId.
  3. A unshares it, then shares the same folder again. B's next polls show the vanish and the appear with the same itemId.
  4. A deletes Probe, creates a new folder Probe and shares it. The new appear has a different itemId.

Expected: no WARN/ERROR in the logs.

### v0.6.389 (#1074, item id on the change feed's /shared roots) On cairn-enc, with two test accounts A and B. 1. As B, take a change-feed cursor: `GET /api/v1/changes?cursor=now` (Settings › Devices token or the browser session). 2. As A, share a folder `Probe` with B. B polls with the cursor: the `/shared/<A>/Probe` entry has `why: "access"` and an `itemId`. 3. A unshares it, then shares the **same** folder again. B's next polls show the vanish and the appear with the **same** `itemId`. 4. A deletes `Probe`, creates a new folder `Probe` and shares it. The new appear has a **different** `itemId`. Expected: no WARN/ERROR in the logs.
Author
Owner

v0.6.390 (#1071, federated shares in the admin Shares list, administrator stop, server removal)

With the lab Nextcloud as the other server, on one dogfood. Use test accounts and test shares only.

  1. The list: as a test user, send two federated shares to a Nextcloud test user; accept one there. Administration › Sharing › Shares shows both after the links and grants:
    • Type Federated, with the address and "{server} · Nextcloud" under it.
    • The unaccepted one shows Not accepted yet (the tooltip also appears on keyboard focus).
    • The Federated type button filters to them, and searching the server's name finds them.
  2. Exports: export CSV and PDF. The CSV ends with server,accepted (yes/no); the PDF summary counts "2 federated".
  3. Mark as reviewed (reviews on): the confirm adds "2 of them are shared with other servers through Open Cloud Mesh."; the audit shows share-review count=<n> federated=2.
  4. Revoke the accepted federated row. The confirm names the server and focus starts on Cancel. Afterwards:
    • The owner's Federated › You shared shows Ended by an administrator.
    • The owner's bell item opens Federated at that server.
    • Nextcloud drops the share at its next check.
    • The audit shows share-admin-revoke … ocm=… server=… told=pending.
  5. Revoke twice: revoke the same row again from a second tab: "it had already ended".
  6. Open Cloud Mesh off: the row stays, with the "switched off" note. Stopping it says the server cannot be told. Switch OCM back on afterwards.
  7. Server removal (lab Nextcloud test entry only; re-add it afterwards):
    • Servers › Remove shows "1 share to this server ends, made by {owner}." The count is read when the dialog opens.
    • Confirm: the answer does not wait for Nextcloud.
    • The owner sees Ended: the server was removed and the bell "An administrator removed {server}".
    • The audit shows share-ended reason=server-removed per share and ocm-server-remove out=1 ….
  8. Audit tiers: with access logging off, send and end a federated share. ocm-share-send and ocm-share-end are recorded with ocm=<id> recipient=<address>. After a pod restart, audit-tiers admin=ocm-share-send,ocm-share-end is in the log.
  9. Ending an ended share: end a share that already ended from Federated (second tab): the page reloads, nothing new in the audit.

Expected: no WARN admin shares: a federated share is still live after it was ended, no ledger WARN, no ERROR.

### v0.6.390 (#1071, federated shares in the admin Shares list, administrator stop, server removal) With the lab Nextcloud as the other server, on one dogfood. Use test accounts and test shares only. 1. **The list:** as a test user, send two federated shares to a Nextcloud test user; accept one there. Administration › Sharing › Shares shows both after the links and grants: - Type **Federated**, with the address and "{server} · Nextcloud" under it. - The unaccepted one shows **Not accepted yet** (the tooltip also appears on keyboard focus). - The **Federated** type button filters to them, and searching the server's name finds them. 2. **Exports:** export CSV and PDF. The CSV ends with `server,accepted` (`yes`/`no`); the PDF summary counts "2 federated". 3. **Mark as reviewed** (reviews on): the confirm adds "2 of them are shared with other servers through Open Cloud Mesh."; the audit shows `share-review count=<n> federated=2`. 4. **Revoke** the accepted federated row. The confirm names the server and focus starts on Cancel. Afterwards: - The owner's Federated › You shared shows **Ended by an administrator**. - The owner's bell item opens Federated at that server. - Nextcloud drops the share at its next check. - The audit shows `share-admin-revoke … ocm=… server=… told=pending`. 5. **Revoke twice:** revoke the same row again from a second tab: "it had already ended". 6. **Open Cloud Mesh off:** the row stays, with the "switched off" note. Stopping it says the server cannot be told. Switch OCM back on afterwards. 7. **Server removal** (lab Nextcloud test entry only; re-add it afterwards): - Servers › Remove shows "1 share to this server ends, made by {owner}." The count is read when the dialog opens. - Confirm: the answer does not wait for Nextcloud. - The owner sees **Ended: the server was removed** and the bell "An administrator removed {server}". - The audit shows `share-ended reason=server-removed` per share and `ocm-server-remove out=1 …`. 8. **Audit tiers:** with access logging off, send and end a federated share. `ocm-share-send` and `ocm-share-end` are recorded with `ocm=<id> recipient=<address>`. After a pod restart, `audit-tiers admin=ocm-share-send,ocm-share-end` is in the log. 9. **Ending an ended share:** end a share that already ended from Federated (second tab): the page reloads, nothing new in the audit. Expected: no WARN `admin shares: a federated share is still live after it was ended`, no ledger WARN, no ERROR.
Author
Owner

v0.6.391 (#1073, lock rows in administrator dialogs)

  1. Transfer files (Users & access › Departed users):
    • A row like "N files are locked for editing by …: the locks move with them and their holders keep them." appears when files are locked. Switching to Only the shared items updates it without a new request.
    • After the transfer, the status line ends "N locks moved with the files."
  2. Delete unread files:
    • The lock row reads "… the locks end with the files and each holder is told." It is amber when an editor is open.
    • Afterwards: "N locks ended; their holders were told."
    • Each holder gets exactly one bell item with a padlock and no link.
  3. Delete space, on a deactivated test space:
    • Opening Delete from the space menu and from Administration › Spaces both show the Delete space dialog with the lock row.
    • The audit space-delete line has locksBroken=<n> locksReleased=<m> (0 included).
  4. Delete group with a space: the lock row shows only while "Delete the space" is chosen. The note says the space's files go to Deleted files.
  5. Administration › Locks:
    • Labels read WebDAV and Office.
    • Break on a lock in another person's Personal folder succeeds; the audit shows break-lock holder=<name> and the holder gets a bell.
    • GET /api/v1/admin/locks has no token or session.
  6. Audit notes: owner-transfer … locks=<n> locksReleased=<m> and owner-delete-unread … locksBroken=<n> locksReleased=<m>.
  7. Wording: German shows "rechtliche Sperre" and "Space"; Italian shows "blocco legale".

Expected: no WARN … the locks inside could not be counted; no ERROR.

### v0.6.391 (#1073, lock rows in administrator dialogs) 1. **Transfer files** (Users & access › Departed users): - A row like "N files are locked for editing by …: the locks move with them and their holders keep them." appears when files are locked. Switching to Only the shared items updates it without a new request. - After the transfer, the status line ends "N locks moved with the files." 2. **Delete unread files:** - The lock row reads "… the locks end with the files and each holder is told." It is amber when an editor is open. - Afterwards: "N locks ended; their holders were told." - Each holder gets exactly one bell item with a padlock and no link. 3. **Delete space**, on a deactivated test space: - Opening Delete from the space menu and from Administration › Spaces both show the Delete space dialog with the lock row. - The audit `space-delete` line has `locksBroken=<n> locksReleased=<m>` (0 included). 4. **Delete group with a space:** the lock row shows only while "Delete the space" is chosen. The note says the space's files go to Deleted files. 5. **Administration › Locks:** - Labels read WebDAV and Office. - Break on a lock in another person's Personal folder succeeds; the audit shows `break-lock holder=<name>` and the holder gets a bell. - `GET /api/v1/admin/locks` has no `token` or `session`. 6. **Audit notes:** `owner-transfer … locks=<n> locksReleased=<m>` and `owner-delete-unread … locksBroken=<n> locksReleased=<m>`. 7. **Wording:** German shows "rechtliche Sperre" and "Space"; Italian shows "blocco legale". Expected: no WARN `… the locks inside could not be counted`; no ERROR.
Author
Owner

v0.6.392 (#1072, a storage restore can pause federated shares)

On cairn-openbao (s3) with the lab Nextcloud as the other server. Use a test share and a test declaration.

  1. Send a federated share to a Nextcloud test user; accept it there; it opens.
  2. Storage & quotas › Usage › Storage was restored…: the box Pause federated shares until their owners turn them on shows "(1 federated share to 1 server; …)", unticked. Tick This restore follows a security incident only on a test realm, otherwise tick the box by hand. Declare (short notice).
  3. Nextcloud keeps the share: its next request and its share check get 503 with Retry-After: 3600 and X-Cairn-Code: share-paused-restore. Run occ sharing:cleanup-remote-storages on the Nextcloud host: the share stays in the recipient's list as unavailable, and Cairn gets no SHARE_DECLINED. Cairn's log has one INFO ocm: share paused after a restore per minute; no 401, 403 or 404 answers for the share.
  4. Owner: Federated › You shared shows Paused: storage restored, the note, and the bell item "1 of your federated shares is paused after a restore". Turn on asks first; confirm: the share opens on Nextcloud again without being sent again, the bell item goes, and the audit has share-restore-resumed.
  5. Admin row: the checklist shows Federated shares (8 items). Declare again with a second test share, then Resume all…: the confirm lists the server; afterwards "Federated shares were resumed on … by …" and restore-federated-resumed.
  6. Optional, End again: end a share, then declare with a backup time before the end. The row lists it under Ended after the backup and back again, but only on s3, and only if the share is live again in the ledger, which needs a real bucket rollback. End again ends it, Nextcloud is told, and the owner sees Ended by an administrator.
  7. Optional: cairnd restored --pause-federated --yes while Cairn is stopped (Kubernetes: scale to 0, then a Job). The summary ends "…, and pauses 1 federated share to 1 server until their owners turn them on."; after the start the share answers 503.

Expected: no WARN/ERROR apart from the documented ones. If step 3 shows a decline, the box must stay unticked by default (OP1, see #1087).

### v0.6.392 (#1072, a storage restore can pause federated shares) On cairn-openbao (s3) with the lab Nextcloud as the other server. Use a test share and a test declaration. 1. Send a federated share to a Nextcloud test user; accept it there; it opens. 2. Storage & quotas › Usage › Storage was restored…: the box **Pause federated shares until their owners turn them on** shows "(1 federated share to 1 server; …)", unticked. Tick **This restore follows a security incident** only on a test realm, otherwise tick the box by hand. Declare (short notice). 3. **Nextcloud keeps the share:** its next request and its share check get 503 with `Retry-After: 3600` and `X-Cairn-Code: share-paused-restore`. Run `occ sharing:cleanup-remote-storages` on the Nextcloud host: the share stays in the recipient's list as unavailable, and Cairn gets **no SHARE_DECLINED**. Cairn's log has one INFO `ocm: share paused after a restore` per minute; no 401, 403 or 404 answers for the share. 4. **Owner:** Federated › You shared shows **Paused: storage restored**, the note, and the bell item "1 of your federated shares is paused after a restore". **Turn on** asks first; confirm: the share opens on Nextcloud again without being sent again, the bell item goes, and the audit has `share-restore-resumed`. 5. **Admin row:** the checklist shows **Federated shares** (8 items). Declare again with a second test share, then **Resume all…**: the confirm lists the server; afterwards "Federated shares were resumed on … by …" and `restore-federated-resumed`. 6. Optional, End again: end a share, then declare with a backup time before the end. The row lists it under **Ended after the backup and back again**, but only on s3, and only if the share is live again in the ledger, which needs a real bucket rollback. **End again** ends it, Nextcloud is told, and the owner sees **Ended by an administrator**. 7. Optional: `cairnd restored --pause-federated --yes` while Cairn is stopped (Kubernetes: scale to 0, then a Job). The summary ends "…, and pauses 1 federated share to 1 server until their owners turn them on."; after the start the share answers 503. Expected: no WARN/ERROR apart from the documented ones. If step 3 shows a decline, the box must stay unticked by default (OP1, see #1087).
Author
Owner

G. Federated shares read by Nextcloud (v0.6.393, #1070), for Nikola

Steps: nc1070 spec §14, on cairn-openbao (the OCM instance). Claude never handles a secret: record whether one appears, never its value.

Already checked on the live v0.6.393 (no secret involved): PROPFIND /dav/ocm/ with no secret answers 401 with both WWW-Authenticate: Basic realm="ocm", charset="UTF-8" and Bearer realm="ocm"; an unknown secret answers 401 with Basic and with Bearer; /status.php 200; discovery lists file with /dav/ocm/ only.

  • Before (14.1): the dogfood already runs v0.6.393, so the "before" half needs a rollback to v0.6.392 (nothing to prepare, see the CHANGELOG's rollback note) or is skipped. Either way record: NC version, the allow-list entry and dialect, ocm_discovery_enabled / enforce_signed_ocm_request, and the access-token-null check on oc_share_external. If NC asks /dav/ocm/<providerId>/ instead of /dav/ocm/, stop and report on #1070.
  • Share: a folder (write) and a file (read) to the NC user; accept both.
  • Open: PROPFIND /dav/ocm/ 207 (Depth 0, then 1); the folder shows its file with the right size and date; the file share opens.
  • Download: GET /dav/ocm/<name> 200; a large file streams.
  • Upload (write share): PUT 201; the file appears in Cairn with the upload recorded for the OCM source. On the read share NC offers no upload.
  • New folder (write share): MKCOL 201. Rename and delete: record NC's message (expected 405, follow-up #1093).
  • Change detection: add a file in Cairn inside a subfolder of the shared folder; record how long until NC shows it (expected: next NC request after at most 5 s).
  • Pause: move the shared folder to Deleted files; NC shows it unavailable and still lists it after a reload and after occ files:scan --all; restore it, NC shows it again.
  • End: End the share in Cairn; Cairn shows told yes; NC removes it.
  • curl from a shell: ended secret 401; -u '<secret>:::' without Depth 207; a path outside the share 404; file share /dav/ocm/<name> the file, /dav/ocm/<name>/x 404; Depth: infinity 403.
  • shareinfo: live 200; ended 404.
  • Logs: INFO ocm: share read with the deprecated Basic form once per share; no WARN/ERROR beyond expected refusals; the secret is in neither Cairn's nor the proxy's log (Nikola greps).
  • Version matrix (if possible): open and download from a second NC major (34 or 33); record whether NC 35 tried a token exchange (it should not; see #1089).
  • CI: run ocm-live-nc job cairn-to-nc once with you watching (secrets CAIRN_URL, CAIRN_USER, CAIRN_APP_PW; input share_pid if the API refuses the app password). If the job log API answers 404, read the runner log.

Until these pass, the CHANGELOG and handbook word the Nextcloud/ownCloud behaviour as expected. Post results to #1070 and here.

## G. Federated shares read by Nextcloud (v0.6.393, #1070), for Nikola Steps: nc1070 spec §14, on cairn-openbao (the OCM instance). Claude never handles a secret: record whether one appears, never its value. Already checked on the live v0.6.393 (no secret involved): `PROPFIND /dav/ocm/` with no secret answers 401 with both `WWW-Authenticate: Basic realm="ocm", charset="UTF-8"` and `Bearer realm="ocm"`; an unknown secret answers 401 with Basic and with Bearer; `/status.php` 200; discovery lists `file` with `/dav/ocm/` only. - [ ] **Before (14.1):** the dogfood already runs v0.6.393, so the "before" half needs a rollback to v0.6.392 (nothing to prepare, see the CHANGELOG's rollback note) or is skipped. Either way record: NC version, the allow-list entry and dialect, `ocm_discovery_enabled` / `enforce_signed_ocm_request`, and the access-token-null check on `oc_share_external`. If NC asks `/dav/ocm/<providerId>/` instead of `/dav/ocm/`, stop and report on #1070. - [ ] **Share:** a folder (write) and a file (read) to the NC user; accept both. - [ ] **Open:** `PROPFIND /dav/ocm/` 207 (Depth 0, then 1); the folder shows its file with the right size and date; the file share opens. - [ ] **Download:** `GET /dav/ocm/<name>` 200; a large file streams. - [ ] **Upload (write share):** `PUT` 201; the file appears in Cairn with the upload recorded for the OCM source. On the read share NC offers no upload. - [ ] **New folder (write share):** `MKCOL` 201. **Rename and delete:** record NC's message (expected 405, follow-up #1093). - [ ] **Change detection:** add a file in Cairn inside a subfolder of the shared folder; record how long until NC shows it (expected: next NC request after at most 5 s). - [ ] **Pause:** move the shared folder to Deleted files; NC shows it unavailable and still lists it after a reload and after `occ files:scan --all`; restore it, NC shows it again. - [ ] **End:** End the share in Cairn; Cairn shows told yes; NC removes it. - [ ] **curl from a shell:** ended secret 401; `-u '<secret>:::'` without Depth 207; a path outside the share 404; file share `/dav/ocm/<name>` the file, `/dav/ocm/<name>/x` 404; `Depth: infinity` 403. - [ ] **shareinfo:** live 200; ended 404. - [ ] **Logs:** INFO `ocm: share read with the deprecated Basic form` once per share; no WARN/ERROR beyond expected refusals; the secret is in neither Cairn's nor the proxy's log (Nikola greps). - [ ] **Version matrix (if possible):** open and download from a second NC major (34 or 33); record whether NC 35 tried a token exchange (it should not; see #1089). - [ ] **CI:** run `ocm-live-nc` job `cairn-to-nc` once with you watching (secrets `CAIRN_URL`, `CAIRN_USER`, `CAIRN_APP_PW`; input `share_pid` if the API refuses the app password). If the job log API answers 404, read the runner log. Until these pass, the CHANGELOG and handbook word the Nextcloud/ownCloud behaviour as expected. Post results to #1070 and here.
Author
Owner

v0.6.394 (#1077, an admin revoke of an earlier account's share stays quiet)

Needs a name used again (#1054) on a test realm.

  • With a name used again (folder kept apart), an admin Revoke of an earlier-account link in Sharing › Shares: the current account's bell shows nothing, and the audit has share-admin-revoke with owner=<name>.
  • The same with Stop sharing selected (a link and a grant) and with a group-review Remove: no bell item for the current account.
  • With nameReuse inherit, the earlier record revoked: the current account's Activity of that item shows no share-removed.
  • After Give, an admin revoke of a handed-over link rings the current account's bell (share-admin-revoked).
  • A share the current account made itself, revoked by an admin, still rings its bell.
### v0.6.394 (#1077, an admin revoke of an earlier account's share stays quiet) Needs a name used again (#1054) on a test realm. - [ ] With a name used again (folder kept apart), an admin Revoke of an earlier-account link in Sharing › Shares: the current account's bell shows nothing, and the audit has `share-admin-revoke` with `owner=<name>`. - [ ] The same with Stop sharing selected (a link and a grant) and with a group-review Remove: no bell item for the current account. - [ ] With `nameReuse` inherit, the earlier record revoked: the current account's Activity of that item shows no `share-removed`. - [ ] After Give, an admin revoke of a handed-over link rings the current account's bell (`share-admin-revoked`). - [ ] A share the current account made itself, revoked by an admin, still rings its bell.
Author
Owner

How to run the ocm-live-nc job (one-time setup, then reusable)

This is the automated half of section G (#1070). Normal CI never runs it; it runs only when started by hand. The secrets stay in Forgejo, so later runs only need step 6.

Before the first run (Claude): the NC→Cairn job (live) still has nikola@files.c0rdyceps.ch hard-coded. That is cairn-enc, where Open Cloud Mesh is off. It must take the recipient from CAIRN_USER and CAIRN_URL (cairn-openbao) instead. Fix this before step 6.

[Browser: Nextcloud, https://nextcloud.c0rdyceps.ch]

  1. Pick or create a test user, e.g. ocm-live. Not your own account: the job writes files.
  2. As that user: Settings › Security › Devices & sessions. Enter the app name ocm-live-ci and click Create new app password. Keep it for step 5.

[Browser: Cairn, https://files-bao.c0rdyceps.ch, as admin]
3. Federation settings: Open Cloud Mesh is on, and nextcloud.c0rdyceps.ch is on the allow-list with the policy "any user" (probably still set from #1057).

[Browser: Cairn, as the test user that will share]
4. Create an app password named ocm-live-ci with full access (not folder-limited). The job shares the folder ocm-live from this user's Personal folder and ends the share at the end.

[Browser: Forgejo, http://192.168.10.245/Cordy/Cairn/settings/actions/secrets]
5. Click Add secret six times:

Name Value
NC_URL https://nextcloud.c0rdyceps.ch
NC_USER the Nextcloud test user
NC_APP_PW the Nextcloud app password
CAIRN_URL https://files-bao.c0rdyceps.ch
CAIRN_USER the Cairn test user
CAIRN_APP_PW the Cairn app password

[Browser: Forgejo › Cairn › Actions › ocm-live-nc]
6. Run workflow on main, share_pid empty. Give Claude the run number; a green cairn-to-nc ends with cairn-to-nc: ALL GREEN. The live job creates a share from Nextcloud: accept it in Cairn under Federated › Requests.
7. If it stops with "the share API refused the app password": share ocm-live with the Nextcloud user by hand in the Cairn UI. Then run it again and enter that share's provider id as share_pid.
8. Afterwards: keep both app passwords for later runs, or revoke them in both apps.

## How to run the `ocm-live-nc` job (one-time setup, then reusable) This is the automated half of section G (#1070). Normal CI never runs it; it runs only when started by hand. The secrets stay in Forgejo, so later runs only need step 6. **Before the first run (Claude):** the NC→Cairn job (`live`) still has `nikola@files.c0rdyceps.ch` hard-coded. That is cairn-enc, where Open Cloud Mesh is off. It must take the recipient from `CAIRN_USER` and `CAIRN_URL` (cairn-openbao) instead. Fix this before step 6. **[Browser: Nextcloud, https://nextcloud.c0rdyceps.ch]** 1. Pick or create a test user, e.g. `ocm-live`. Not your own account: the job writes files. 2. As that user: Settings › Security › Devices & sessions. Enter the app name `ocm-live-ci` and click **Create new app password**. Keep it for step 5. **[Browser: Cairn, https://files-bao.c0rdyceps.ch, as admin]** 3. Federation settings: Open Cloud Mesh is on, and `nextcloud.c0rdyceps.ch` is on the allow-list with the policy "any user" (probably still set from #1057). **[Browser: Cairn, as the test user that will share]** 4. Create an app password named `ocm-live-ci` with **full access** (not folder-limited). The job shares the folder `ocm-live` from this user's Personal folder and ends the share at the end. **[Browser: Forgejo, http://192.168.10.245/Cordy/Cairn/settings/actions/secrets]** 5. Click **Add secret** six times: | Name | Value | |---|---| | `NC_URL` | `https://nextcloud.c0rdyceps.ch` | | `NC_USER` | the Nextcloud test user | | `NC_APP_PW` | the Nextcloud app password | | `CAIRN_URL` | `https://files-bao.c0rdyceps.ch` | | `CAIRN_USER` | the Cairn test user | | `CAIRN_APP_PW` | the Cairn app password | **[Browser: Forgejo › Cairn › Actions › ocm-live-nc]** 6. **Run workflow** on `main`, `share_pid` empty. Give Claude the run number; a green `cairn-to-nc` ends with `cairn-to-nc: ALL GREEN`. The `live` job creates a share from Nextcloud: accept it in Cairn under Federated › Requests. 7. If it stops with "the share API refused the app password": share `ocm-live` with the Nextcloud user by hand in the Cairn UI. Then run it again and enter that share's provider id as `share_pid`. 8. Afterwards: keep both app passwords for later runs, or revoke them in both apps.
Author
Owner

v0.6.395 (#1100, Deleted files on by default)

Both dogfoods set storage.trash explicitly (they log trash enabled source=config), so the upgrade path needs a test instance or a temporary config change.

  1. Before upgrading (a test instance): remove storage.trash from the config. Run find <storage root>/spaces -mindepth 2 -maxdepth 2 -name .trash. As a test, make a folder <storage root>/spaces/<test space>/.trash holding two files, then upgrade.
  2. Start log: INFO trash enabled source=default, INFO trash: Deleted files is on by default; set storage.trash.enabled to false to switch it off, and one WARN trash: a folder named .trash at the top of a space is not a Deleted files tree; its files are now hidden with folder=/spaces/<test space>/.trash items=2.
  3. Bell (administrator): after about 2 seconds, "Deleted files is now on" (download icon) and "A folder named “.trash” in is hidden" with How to fix ↗.
  4. How to fix ↗ opens the handbook section; clicking "Deleted files is now on" opens Administration › Deleted files › Retention with the stored settings loaded. Reload: still loaded.
  5. Retention tab: switch Delete automatically off: the amber plate (light theme) reads the corrected quota sentence and is announced once. With no stored period, switching on preselects 30 days. Discard puts the row and plate back.
  6. Restart: the folder WARN repeats, no new bell item; trash-default.json holds default-on and the folder path.
  7. Fix it: rename the stray folder on the storage; its files appear in the space and the next start logs no WARN.
  8. Deletes: delete a test file in a personal folder and in a space: both in Deleted files and restorable.
  9. Settings safety: with the browser offline after page load, open a settings tab, change a field, Save: refused with "Load failed", nothing saved.
  10. Optional, new install: no key, no settings: INFO settings: new instance; deleted files are purged automatically after 720h, Retention shows 30 days, no bell.

Expected: no WARN/ERROR apart from the documented ones; nothing on the storage moves.

### v0.6.395 (#1100, Deleted files on by default) Both dogfoods set `storage.trash` explicitly (they log `trash enabled source=config`), so the upgrade path needs a test instance or a temporary config change. 1. **Before upgrading (a test instance):** remove `storage.trash` from the config. Run `find <storage root>/spaces -mindepth 2 -maxdepth 2 -name .trash`. As a test, make a folder `<storage root>/spaces/<test space>/.trash` holding two files, then upgrade. 2. **Start log:** INFO `trash enabled source=default`, INFO `trash: Deleted files is on by default; set storage.trash.enabled to false to switch it off`, and one WARN `trash: a folder named .trash at the top of a space is not a Deleted files tree; its files are now hidden` with `folder=/spaces/<test space>/.trash items=2`. 3. **Bell (administrator):** after about 2 seconds, "Deleted files is now on" (download icon) and "A folder named “.trash” in <space> is hidden" with **How to fix ↗**. 4. **How to fix ↗** opens the handbook section; clicking "Deleted files is now on" opens Administration › Deleted files › Retention with the stored settings loaded. Reload: still loaded. 5. **Retention tab:** switch Delete automatically off: the amber plate (light theme) reads the corrected quota sentence and is announced once. With no stored period, switching on preselects 30 days. Discard puts the row and plate back. 6. **Restart:** the folder WARN repeats, no new bell item; `trash-default.json` holds `default-on` and the folder path. 7. **Fix it:** rename the stray folder on the storage; its files appear in the space and the next start logs no WARN. 8. **Deletes:** delete a test file in a personal folder and in a space: both in Deleted files and restorable. 9. **Settings safety:** with the browser offline after page load, open a settings tab, change a field, Save: refused with "Load failed", nothing saved. 10. Optional, new install: no key, no settings: INFO `settings: new instance; deleted files are purged automatically after 720h`, Retention shows 30 days, no bell. Expected: no WARN/ERROR apart from the documented ones; nothing on the storage moves.
Author
Owner

v0.6.396 (#1099, WebDAV for sync clients)

Release gate already passed: interop-rclone run #3980 on the branch.

  1. rclone, new remote from Connect a device › Ready (vendor=owncloud): copy three files dated 2019, rclone lsl shows 2019; a same-size edit then rclone bisync reaches Cairn; rclone touch -t 2018-01-02T03:04:05 shows in lsl and the audit has modified-time-set (mtime=… from=…); rclone moveto keeps the date; rclone check is clean.
  2. Existing vendor=other remote: usual bisync, rclone config update cairn vendor owncloud, --resync --resync-mode path1 --dry-run, then without --dry-run; a same-size edit is then seen.
  3. Add-only device: an existing vendor=other remote keeps working with rclone copy --immutable (no "immutable file modified"); a new add-only device on vendor=owncloud keeps dates and a second run reports nothing.
  4. Windows File Explorer: copy a folder with files onto the mapped drive: no error dialog, files keep their dates, the folder shows the server time; a single file keeps its date; the same into an Add files only folder works.
  5. Finder: copy files and a folder into the mount: no error, files open; note what Modified shows (upload time expected).
  6. Cyberduck with "preserve modification date": date kept.
  7. cadaver: propset of a custom property gets a 403 propstat inside a 207.
  8. Web app: an old file uploaded (small and resumable) shows the device date; Copy to… keeps the original's date; a drop-link upload shows the upload time; a file dated 1970-01-01 shows the upload time and logs INFO client file date ignored.
  9. Caching: downloads and the preview carry Cache-Control: private, no-cache; a PDF preview has an ETag and answers 304 on reload; after replacing a file with older-dated content the preview shows the new content.
  10. Change feed: a PROPPATCH date change is a write entry with an id; no Activity entry.
  11. Federation regression (folder tags moved to internal/foldertag): the lab Nextcloud still lists the Cairn share and a deep change moves the root ETag.

Expected: no WARN/ERROR apart from the documented ones.

### v0.6.396 (#1099, WebDAV for sync clients) Release gate already passed: `interop-rclone` run #3980 on the branch. 1. **rclone, new remote from Connect a device › Ready** (`vendor=owncloud`): copy three files dated 2019, `rclone lsl` shows 2019; a same-size edit then `rclone bisync` reaches Cairn; `rclone touch -t 2018-01-02T03:04:05` shows in `lsl` and the audit has `modified-time-set` (`mtime=… from=…`); `rclone moveto` keeps the date; `rclone check` is clean. 2. **Existing `vendor=other` remote:** usual bisync, `rclone config update cairn vendor owncloud`, `--resync --resync-mode path1 --dry-run`, then without `--dry-run`; a same-size edit is then seen. 3. **Add-only device:** an existing `vendor=other` remote keeps working with `rclone copy --immutable` (no "immutable file modified"); a new add-only device on `vendor=owncloud` keeps dates and a second run reports nothing. 4. **Windows File Explorer:** copy a folder with files onto the mapped drive: no error dialog, files keep their dates, the folder shows the server time; a single file keeps its date; the same into an Add files only folder works. 5. **Finder:** copy files and a folder into the mount: no error, files open; note what Modified shows (upload time expected). 6. **Cyberduck** with "preserve modification date": date kept. 7. **cadaver:** `propset` of a custom property gets a 403 propstat inside a 207. 8. **Web app:** an old file uploaded (small and resumable) shows the device date; Copy to… keeps the original's date; a drop-link upload shows the upload time; a file dated 1970-01-01 shows the upload time and logs INFO `client file date ignored`. 9. **Caching:** downloads and the preview carry `Cache-Control: private, no-cache`; a PDF preview has an `ETag` and answers 304 on reload; after replacing a file with older-dated content the preview shows the new content. 10. **Change feed:** a PROPPATCH date change is a `write` entry with an `id`; no Activity entry. 11. **Federation regression** (folder tags moved to `internal/foldertag`): the lab Nextcloud still lists the Cairn share and a deep change moves the root ETag. Expected: no WARN/ERROR apart from the documented ones.
Author
Owner

Dogfood checks for v0.6.397 (#1097) and v0.6.398 (#1098), both live on files.c0rdyceps.ch (posix) and files-bao.c0rdyceps.ch (s3).

v0.6.397, ZIP downloads:

  • The ZIP64 gate on #1097 (the release check for #1097; it stays open until then).
  • Select two files and a folder, Download: one ZIP, folders kept, names with umlauts correct.
  • A folder over 1 GB asks first ("n files in m folders, x GB in total"); Cancel sends nothing.
  • Search, Favourites and Recents: Download of a selection takes the selected items (the old wrong-list bug).
  • A public folder link: Download all as ZIP; a password link asks the password first; a strong (encrypted) link decrypts in the page.
  • Admin › ZIP downloads box: change a limit, save, reload; Audit shows download-zip.

v0.6.398, folder upload:

  • Toolbar arrow › Upload folder: pick a folder with subfolders; the plan question appears only over 1,000 files or 10 GB; everything arrives with its folders.
  • Drag a folder from Finder onto the list; while it says "Reading…", open another folder: the upload still goes where you dropped it.
  • Upload the same folder again: Add to it with "Skip files that are already there…" ticked sends only what's new; Keep both makes "Name 2".
  • Press Stop during a big upload: stopped files show in Not uploaded, Try again sends them.
  • Drop a single empty (0-byte) text file: it uploads (was skipped before).
  • An upload link: Choose a folder and a folder drop; uploading a name that exists keeps both and never replaces; an empty folder pick says "The folder you chose has no files to upload."
**Dogfood checks for v0.6.397 (#1097) and v0.6.398 (#1098)**, both live on files.c0rdyceps.ch (posix) and files-bao.c0rdyceps.ch (s3). v0.6.397, ZIP downloads: - [ ] The ZIP64 gate on #1097 (the release check for #1097; it stays open until then). - [ ] Select two files and a folder, **Download**: one ZIP, folders kept, names with umlauts correct. - [ ] A folder over 1 GB asks first ("n files in m folders, x GB in total"); Cancel sends nothing. - [ ] Search, Favourites and Recents: **Download** of a selection takes the selected items (the old wrong-list bug). - [ ] A public folder link: **Download all as ZIP**; a password link asks the password first; a strong (encrypted) link decrypts in the page. - [ ] Admin › ZIP downloads box: change a limit, save, reload; Audit shows `download-zip`. v0.6.398, folder upload: - [ ] Toolbar arrow › **Upload folder**: pick a folder with subfolders; the plan question appears only over 1,000 files or 10 GB; everything arrives with its folders. - [ ] Drag a folder from Finder onto the list; while it says "Reading…", open another folder: the upload still goes where you dropped it. - [ ] Upload the same folder again: **Add to it** with "Skip files that are already there…" ticked sends only what's new; **Keep both** makes "Name 2". - [ ] Press **Stop** during a big upload: stopped files show in **Not uploaded**, **Try again** sends them. - [ ] Drop a single empty (0-byte) text file: it uploads (was skipped before). - [ ] An upload link: **Choose a folder** and a folder drop; uploading a name that exists keeps both and never replaces; an empty folder pick says "The folder you chose has no files to upload."
Author
Owner

Dogfood checks for v0.6.399 (#1102 name search), live on both instances. Sign in again first (the redeploy ended all sessions).

  • Admin › Storage & quotas › Usage › Name search: Ready, "Files and folders: n" roughly matches your data (posix logged only 1 entry over 9 roots at start, see #1102).
  • Search a name that lives deep in a subfolder: found, with a count ("n results").
  • Search muller finds Müller…; a name with an accent typed without it is found.
  • Filters: Type = PDF and Last modified = Last year change the results without a reload; Clear filters brings them back.
  • A common word (e.g. scan or img): "More than 1,000 matches…" notice; with a filter set, the line "Only the best 1,000 name matches were checked against the filters…".
  • Load more adds rows and keeps focus; Escape cancels a slow search.
  • Prepare again on the card asks first, then shows Being prepared and returns to Ready; Audit shows search-index-rebuild.
  • Overview shows the Name search row.
**Dogfood checks for v0.6.399 (#1102 name search)**, live on both instances. Sign in again first (the redeploy ended all sessions). - [ ] Admin › Storage & quotas › Usage › **Name search**: Ready, "Files and folders: n" roughly matches your data (posix logged only 1 entry over 9 roots at start, see #1102). - [ ] Search a name that lives deep in a subfolder: found, with a count ("n results"). - [ ] Search `muller` finds `Müller…`; a name with an accent typed without it is found. - [ ] Filters: Type = PDF and Last modified = Last year change the results without a reload; Clear filters brings them back. - [ ] A common word (e.g. `scan` or `img`): "More than 1,000 matches…" notice; with a filter set, the line "Only the best 1,000 name matches were checked against the filters…". - [ ] Load more adds rows and keeps focus; Escape cancels a slow search. - [ ] Prepare again on the card asks first, then shows Being prepared and returns to Ready; Audit shows `search-index-rebuild`. - [ ] Overview shows the Name search row.
Author
Owner

v0.6.400 live checks (both dogfoods on v0.6.400, ready). Needs your sign-in:

  1. Search, Favourites, Recents: select 3 rows, Delete. The confirm lists them with their folders; after Delete you stay in the view; the status line names what failed, then what worked. Same for Move to and Copy to (picker starts at the folder holding the selection).
  2. Select a space plus a file in search, Delete: "Left out: 1. Spaces and items shared with you..." shows in the confirm.
  3. Favourites view: the bar reads "Remove from favourites"; removed rows leave in place.
  4. Name search for a space you are a member of, by its name: it is found. Rename it, search the new name: found.
  5. tag:<tag> with Type and Last modified pills, and with This folder.
  6. On the S3 dogfood (files-enc): search a folder name with Last modified "7 days": the folder is listed.
  7. On the S3 dogfood: download a folder as ZIP, unpack with Archive Utility: the top folder is dated (not 30 Nov 1979).
  8. Dark theme: selected rows in search stay readable.
**v0.6.400 live checks** (both dogfoods on v0.6.400, ready). Needs your sign-in: 1. Search, Favourites, Recents: select 3 rows, Delete. The confirm lists them with their folders; after Delete you stay in the view; the status line names what failed, then what worked. Same for Move to and Copy to (picker starts at the folder holding the selection). 2. Select a space plus a file in search, Delete: "Left out: 1. Spaces and items shared with you..." shows in the confirm. 3. Favourites view: the bar reads "Remove from favourites"; removed rows leave in place. 4. Name search for a space you are a member of, by its name: it is found. Rename it, search the new name: found. 5. `tag:<tag>` with Type and Last modified pills, and with This folder. 6. On the S3 dogfood (files-enc): search a folder name with Last modified "7 days": the folder is listed. 7. On the S3 dogfood: download a folder as ZIP, unpack with Archive Utility: the top folder is dated (not 30 Nov 1979). 8. Dark theme: selected rows in search stay readable.
Author
Owner

v0.6.401 live checks (both dogfoods on v0.6.401, ready). Needs your sign-in:

  1. Both dogfoods: sign in through the identity provider (the sign-in state is now a sealed cookie); open a document in the editor, restart nothing, keep editing for 2 minutes.
  2. Administration: End all editor sessions. The open editor ends; opening the document again works.
  3. Two admin tabs: change a setting in tab 1, then save a different setting in tab 2: tab 2 is told the settings changed (409) and nothing is lost.
  4. On the S3 dogfood (files-enc): copy a folder of about 100 files that have older versions; note how long it takes, then open Versions on one copied file: its history is there (#950, cost follow-up #1135).
  5. A local test account with an app password: delete the account; the app password no longer signs in (WebDAV).
**v0.6.401 live checks** (both dogfoods on v0.6.401, ready). Needs your sign-in: 1. Both dogfoods: sign in through the identity provider (the sign-in state is now a sealed cookie); open a document in the editor, restart nothing, keep editing for 2 minutes. 2. Administration: **End all editor sessions**. The open editor ends; opening the document again works. 3. Two admin tabs: change a setting in tab 1, then save a different setting in tab 2: tab 2 is told the settings changed (409) and nothing is lost. 4. On the S3 dogfood (files-enc): copy a folder of about 100 files that have older versions; note how long it takes, then open Versions on one copied file: its history is there (#950, cost follow-up #1135). 5. A local test account with an app password: delete the account; the app password no longer signs in (WebDAV).
Author
Owner

v0.6.402 live checks (#1139), in Safari and Firefox (Chromium done), on a dogfood while signed in:

  1. Download a file, a folder as ZIP and an older version: each saves normally.
  2. Open (new tab) a .png, an .mp4 and an .mp3: the image shows, video and audio play.
  3. Open a .pdf: it shows in the browser's viewer.
  4. Upload an .html and an .svg, then Open each: both download, neither renders.
  5. In Finder or another WebDAV client: copying an .html file out still works.
**v0.6.402 live checks (#1139)**, in Safari and Firefox (Chromium done), on a dogfood while signed in: 1. Download a file, a folder as ZIP and an older version: each saves normally. 2. Open (new tab) a .png, an .mp4 and an .mp3: the image shows, video and audio play. 3. Open a .pdf: it shows in the browser's viewer. 4. Upload an .html and an .svg, then Open each: both download, neither renders. 5. In Finder or another WebDAV client: copying an .html file out still works.
Author
Owner

v0.6.403 live checks (#1105, #1140, #1106, #1107), for Nikola on the dogfoods (sign-in needed):

Links (#1105, #1140)

  • Turn on View-only links (Administration › Sharing). Read the editor's /hosting/discovery: does the dogfood Collabora publish proof keys (modulus/exponent)? If not, set office.serverAddresses and check the address-list fallback.
  • Make a view-only link to a .docx and a .pdf; open it in a private window: read only, watermark visible, visitor named "Visitor" in the page's language (lang), keyboard focus can leave the editor frame.
  • Revoke the link while the visitor page is open: the next editor call fails.
  • A browser replaying the visitor token straight to WOPI gets 500 with X-Cairn-Code: wopi-proof-required (proof-key mode).
  • Download limit 2 across two browser sessions: the third download is refused (410), the bell arrives.
  • Changing a link password signs out a second browser.
  • Copy link again in Safari and Firefox (http dogfood: the selected-field fallback).
  • Edit a link (label, note, role, limit) from the share pane and from the Shares view.

Personal folder sizes (#1106)

  • Give one account an own size below its usage: uploads refused, nothing deleted, red banner in Personal.
  • A Keycloak group size: applies after the member's next sign-in; Accounts shows "Size from the group …".

Space managers (#1107)

  • Make a member a manager, then hand the space over; check bells, Members dialog, lock menu, dark theme selected rows.
  • A manager cannot delete the space; Stop managing works.
**v0.6.403 live checks** (#1105, #1140, #1106, #1107), for Nikola on the dogfoods (sign-in needed): Links (#1105, #1140) - [ ] Turn on View-only links (Administration › Sharing). Read the editor's `/hosting/discovery`: does the dogfood Collabora publish proof keys (`modulus`/`exponent`)? If not, set `office.serverAddresses` and check the address-list fallback. - [ ] Make a view-only link to a .docx and a .pdf; open it in a private window: read only, watermark visible, visitor named "Visitor" in the page's language (`lang`), keyboard focus can leave the editor frame. - [ ] Revoke the link while the visitor page is open: the next editor call fails. - [ ] A browser replaying the visitor token straight to WOPI gets 500 with `X-Cairn-Code: wopi-proof-required` (proof-key mode). - [ ] Download limit 2 across two browser sessions: the third download is refused (410), the bell arrives. - [ ] Changing a link password signs out a second browser. - [ ] Copy link again in Safari and Firefox (http dogfood: the selected-field fallback). - [ ] Edit a link (label, note, role, limit) from the share pane and from the Shares view. Personal folder sizes (#1106) - [ ] Give one account an own size below its usage: uploads refused, nothing deleted, red banner in Personal. - [ ] A Keycloak group size: applies after the member's next sign-in; Accounts shows "Size from the group …". Space managers (#1107) - [ ] Make a member a manager, then hand the space over; check bells, Members dialog, lock menu, dark theme selected rows. - [ ] A manager cannot delete the space; Stop managing works.
Author
Owner

v0.6.404 live-check list (both dogfoods; Nikola signs in, Claude never enters passwords). Automatic part done: both pods on v0.6.404, 512Mi, 0 restarts, healthz 200, caches sealed, v0.6.403 plaintext caches removed, only the pre-existing WARN lines.

Branding (#1109)

  • Upload PNG logo, SVG logo, dark logo, icon; a GIF, an SVG with <script> and an old Illustrator SVG with DOCTYPE are refused (audit branding-image-refused, hint shown); Save writes branding-change.
  • Sign-in page: logo, title, icon, accent, footer Imprint / Privacy / Runs on Cairn; credit off removes it (and titles show the host when no name is set).
  • Header in a light and a dark theme; Imprint and Privacy in the user menu open a new tab; Appearance "Instance default" marker.
  • Unknown, expired and revoked link tokens give the same branded 404.
  • A security notice shows -- once and the Imprint/Privacy lines, no image, no credit.

Sharing e-mails (#1103)

  • Share to a person and to a small group: mail arrives, From "… via …", Reply-To, inline logo, footer links.
  • Gmail's and Outlook's Unsubscribe post one-click (relay DKIM must sign both List headers); Stop page GET changes nothing, Stop / Turn on again / Manage opens ?row=.
  • Send a link to an outside address with and without a password (no password in the mail; key link carries #key); Preview sends nothing.
  • Upload burst: one mail after 10 quiet minutes. Link ending in 3 days 1 hour: bell item (also with mail off) and mail.
  • Daily summary at about 07:00-07:10 server time.
  • Administration › Sharing › Policy: Send sharing e-mails off stops new optional mail; the Notifications page shows the note.
  • Outbox shows a security mail overtaking queued sharing mails.

Thumbnails (#1101)

  • Encrypted dogfood: PDF, pptx with picture, SVG, HEIC (Safari) tiles show; selecting a PDF leaves Recents unchanged.
  • Log: preview: pdf renderer compiled engine=compiler cache=on on the first PDF; a second pod start shows the warm compile.
  • No OOM on the first PDF at 512Mi; Previews tab shows sealed, "Built in", no memory warning.
  • An arm64 Pi pod renders a PDF (both dogfoods run on pi2).
**v0.6.404 live-check list** (both dogfoods; Nikola signs in, Claude never enters passwords). Automatic part done: both pods on v0.6.404, 512Mi, 0 restarts, healthz 200, caches sealed, v0.6.403 plaintext caches removed, only the pre-existing WARN lines. Branding (#1109) - [ ] Upload PNG logo, SVG logo, dark logo, icon; a GIF, an SVG with `<script>` and an old Illustrator SVG with DOCTYPE are refused (audit `branding-image-refused`, hint shown); Save writes `branding-change`. - [ ] Sign-in page: logo, title, icon, accent, footer Imprint / Privacy / Runs on Cairn; credit off removes it (and titles show the host when no name is set). - [ ] Header in a light and a dark theme; Imprint and Privacy in the user menu open a new tab; Appearance "Instance default" marker. - [ ] Unknown, expired and revoked link tokens give the same branded 404. - [ ] A security notice shows `-- ` once and the Imprint/Privacy lines, no image, no credit. Sharing e-mails (#1103) - [ ] Share to a person and to a small group: mail arrives, From "… via …", Reply-To, inline logo, footer links. - [ ] Gmail's and Outlook's Unsubscribe post one-click (relay DKIM must sign both List headers); Stop page GET changes nothing, Stop / Turn on again / Manage opens `?row=`. - [ ] Send a link to an outside address with and without a password (no password in the mail; key link carries `#key`); Preview sends nothing. - [ ] Upload burst: one mail after 10 quiet minutes. Link ending in 3 days 1 hour: bell item (also with mail off) and mail. - [ ] Daily summary at about 07:00-07:10 server time. - [ ] Administration › Sharing › Policy: Send sharing e-mails off stops new optional mail; the Notifications page shows the note. - [ ] Outbox shows a security mail overtaking queued sharing mails. Thumbnails (#1101) - [ ] Encrypted dogfood: PDF, pptx with picture, SVG, HEIC (Safari) tiles show; selecting a PDF leaves Recents unchanged. - [ ] Log: `preview: pdf renderer compiled engine=compiler cache=on` on the first PDF; a second pod start shows the warm compile. - [ ] No OOM on the first PDF at 512Mi; Previews tab shows sealed, "Built in", no memory warning. - [ ] An arm64 Pi pod renders a PDF (both dogfoods run on pi2).
Author
Owner

v0.6.405 live-check list (both dogfoods; Nikola signs in). Automatic part done: both pods on v0.6.405, ready, 0 restarts, healthz 200, only pre-existing WARN lines.

File versions (#1110, #924)

  • The one-time 7-day bell appears once for administrators; a restart does not repeat it.
  • #admin/retention/versions shows 7 / 10 / 90 / 1,000 / 20 / 10 / on; lowering a limit asks first and posts the bell.
  • Keep with a name, rename, Don't keep and Undo; the Kept badge; audit notes without the name.
  • Delete a version as a space manager; restore it back into the history; delete the file and restore the version entry as a copy; purge it.
  • A version inside 7 days shows "Possible from {date}"; a Read + write member can keep but not delete.
  • A legal hold shows "Held since … by …"; only administrators see the note.
  • The first night at 02:30: versions-sweep-run audit, Last run in the tab.

Restore to an earlier time (#1104) and G4

  • curl -u with an app password on DELETE /api/v1/trash/{id} gets 403 needs-session, a purge-refused audit line, a red bell item and one security e-mail; restoring with the device still works.
  • Restore a test folder: chart, preview sections, Untick all deleted by, Keep these, CSV, Check again, run with device pause, Stop, Continue, Undo, Restore these too.
  • A member of the space opens the result page from the bell; the 14-day banner; Deleted files group with Bring all back and Delete all forever (browser only).
  • Administration › Earlier-time restores: switches and the run list.
  • A run survives a pod restart (rewind resumed on the next start).
**v0.6.405 live-check list** (both dogfoods; Nikola signs in). Automatic part done: both pods on v0.6.405, ready, 0 restarts, healthz 200, only pre-existing WARN lines. File versions (#1110, #924) - [ ] The one-time 7-day bell appears once for administrators; a restart does not repeat it. - [ ] `#admin/retention/versions` shows 7 / 10 / 90 / 1,000 / 20 / 10 / on; lowering a limit asks first and posts the bell. - [ ] Keep with a name, rename, Don't keep and Undo; the Kept badge; audit notes without the name. - [ ] Delete a version as a space manager; restore it back into the history; delete the file and restore the version entry as a copy; purge it. - [ ] A version inside 7 days shows "Possible from {date}"; a Read + write member can keep but not delete. - [ ] A legal hold shows "Held since … by …"; only administrators see the note. - [ ] The first night at 02:30: `versions-sweep-run` audit, Last run in the tab. Restore to an earlier time (#1104) and G4 - [ ] `curl -u` with an app password on `DELETE /api/v1/trash/{id}` gets 403 `needs-session`, a `purge-refused` audit line, a red bell item and one security e-mail; restoring with the device still works. - [ ] Restore a test folder: chart, preview sections, Untick all deleted by, Keep these, CSV, Check again, run with device pause, Stop, Continue, Undo, Restore these too. - [ ] A member of the space opens the result page from the bell; the 14-day banner; Deleted files group with Bring all back and Delete all forever (browser only). - [ ] Administration › Earlier-time restores: switches and the run list. - [ ] A run survives a pod restart (`rewind resumed` on the next start).
Author
Owner

v0.6.406 live checks (both dogfoods, running since 2026-11-11 15:34 UTC). These need a person, a browser and a device.

#1046 Two-step sign-in

  • Scan the authenticator setup QR with a phone, in the light and dark themes. The code must stay black on white.
  • Hardware security key (USB or NFC): add it, sign in, Confirm it's you, remove it.
  • Synced passkey (iCloud Keychain or Google Password Manager): add it and sign in. Then turn on "Only allow keys that cannot be copied" and check that it is refused.
  • Password-manager passkey (1Password or Bitwarden): it must pass the extension-output check.
  • Safari:
    • the security key button on the sign-in page;
    • Add security key on Account › Security;
    • "Sign in again" after a session expires: the window opens on the click.
  • Require two-step sign-in for Administrators, then check the reminder e-mail and the Remind them button.
  • With a factor set up, a WebDAV client using the account password gets refused and a bell item appears. An app password still works.

#1036 Guests and OCM from Spaces

  • Turn guests on, then invite a test address.
  • As the guest: open the invitation, the code mail and the sign-in. Check the "No e-mail? Check the spam folder" line, the list, upload, the editor (Can edit) and sign-out.
  • Share from a Space with a Nextcloud, OpenCloud or oCIS server. Check how the owner and sender display names render, and that a dated share gets accepted (expiration is sent in Unix seconds).

#1108 API keys

  • Turn keys on, create a key and call /api/v1/admin/info with curl.
  • The same key from a browser tab is refused.
  • Rotate the key, then revoke it.

#1156

  • Spot check: /s/g:<some grant id> answers 404.
**v0.6.406 live checks** (both dogfoods, running since 2026-11-11 15:34 UTC). These need a person, a browser and a device. **#1046 Two-step sign-in** - [ ] Scan the authenticator setup QR with a phone, in the light and dark themes. The code must stay black on white. - [ ] Hardware security key (USB or NFC): add it, sign in, Confirm it's you, remove it. - [ ] Synced passkey (iCloud Keychain or Google Password Manager): add it and sign in. Then turn on "Only allow keys that cannot be copied" and check that it is refused. - [ ] Password-manager passkey (1Password or Bitwarden): it must pass the extension-output check. - [ ] Safari: - the security key button on the sign-in page; - Add security key on Account › Security; - "Sign in again" after a session expires: the window opens on the click. - [ ] Require two-step sign-in for Administrators, then check the reminder e-mail and the Remind them button. - [ ] With a factor set up, a WebDAV client using the account password gets refused and a bell item appears. An app password still works. **#1036 Guests and OCM from Spaces** - [ ] Turn guests on, then invite a test address. - [ ] As the guest: open the invitation, the code mail and the sign-in. Check the "No e-mail? Check the spam folder" line, the list, upload, the editor (Can edit) and sign-out. - [ ] Share from a Space with a Nextcloud, OpenCloud or oCIS server. Check how the owner and sender display names render, and that a dated share gets accepted (`expiration` is sent in Unix seconds). **#1108 API keys** - [ ] Turn keys on, create a key and call `/api/v1/admin/info` with curl. - [ ] The same key from a browser tab is refused. - [ ] Rotate the key, then revoke it. **#1156** - [ ] Spot check: `/s/g:<some grant id>` answers 404.
Author
Owner

v0.6.407 live checks (Nikola signs in; Claude cannot):

  • #1152: Administration › Deleted files › Retention: the new "Keep items deleted forever for administrators" row (dogfoods upgraded, so Off until chosen; the offer bell appears). Choose 14 days, Delete forever a test file as a person, find it under Deleted forever, Restore it, then Erase now another with a reason. Turn on "Erase now needs a second administrator" (needs two admins) and check the request card, approval with second step, and that the security mail subject names no person.
  • #1044: Administration › Audit log › Forwarding: empty state; add a test destination only if a receiver exists (otherwise check the form, Test button and validation); Overview shows "Audit forwarding: Off"; Notifications has the "Audit forwarding problems" row.
  • #1043: Administration › Classification: turn on with a preset, label a test folder, check the chip, a public link on it refused/stopped and back when lowered (reason required), the owner bell, Review and Labelled items tabs; then turn off again.
  • #1137: sign-in page normal; Users & access Provider chip OK; (optional) a Keycloak outage shows the sign-in notice and Overview card, local sign-in still works.
  • #936: Administration › Spaces: no unverified rows expected on the dogfoods; Take over asks first.
**v0.6.407 live checks** (Nikola signs in; Claude cannot): - #1152: Administration › Deleted files › Retention: the new "Keep items deleted forever for administrators" row (dogfoods upgraded, so Off until chosen; the offer bell appears). Choose 14 days, Delete forever a test file as a person, find it under Deleted forever, Restore it, then Erase now another with a reason. Turn on "Erase now needs a second administrator" (needs two admins) and check the request card, approval with second step, and that the security mail subject names no person. - #1044: Administration › Audit log › Forwarding: empty state; add a test destination only if a receiver exists (otherwise check the form, Test button and validation); Overview shows "Audit forwarding: Off"; Notifications has the "Audit forwarding problems" row. - #1043: Administration › Classification: turn on with a preset, label a test folder, check the chip, a public link on it refused/stopped and back when lowered (reason required), the owner bell, Review and Labelled items tabs; then turn off again. - #1137: sign-in page normal; Users & access Provider chip OK; (optional) a Keycloak outage shows the sign-in notice and Overview card, local sign-in still works. - #936: Administration › Spaces: no unverified rows expected on the dogfoods; Take over asks first.
Author
Owner

v0.6.408 live checks (#1011; Nikola signs in):

  • Administration › Overview: no state card while storage is fine; Notifications/bell unchanged.
  • Optional outage drill on the enc dogfood only (needs a GitOps/bucket-policy change, so plan it): block read access to the .cairn-state/ prefix in Garage for ~7 minutes, then restore. Expect: amber card at once ("answering from the last copy"), red after 5 minutes listing what stopped, rail dot, app-password/WebDAV clients get 503 (not a password prompt), a delete answers "legal holds cannot be read", emergency sign-in still works; after restore the card goes within ~30 s and the bell shows "State could not be read for N minutes".
  • Sign-in page unchanged in normal operation.
  • GET /api/v1/admin/state-health as admin answers {"stores":[], "tracked": …} when healthy.
**v0.6.408 live checks** (#1011; Nikola signs in): - Administration › Overview: no state card while storage is fine; Notifications/bell unchanged. - Optional outage drill on the enc dogfood only (needs a GitOps/bucket-policy change, so plan it): block read access to the `.cairn-state/` prefix in Garage for ~7 minutes, then restore. Expect: amber card at once ("answering from the last copy"), red after 5 minutes listing what stopped, rail dot, app-password/WebDAV clients get 503 (not a password prompt), a delete answers "legal holds cannot be read", emergency sign-in still works; after restore the card goes within ~30 s and the bell shows "State could not be read for N minutes". - Sign-in page unchanged in normal operation. - `GET /api/v1/admin/state-health` as admin answers `{"stores":[], "tracked": …}` when healthy.
Author
Owner

Live check for v0.6.409 (2026-11-18, about 3 min after the dogfood bump, homelab-config cbcf248):

  • Pods: cairn-enc and cairn-openbao both run cordy/cairn:v0.6.409, are ready and have 0 restarts. Both stay in local mode.
  • Logs: the WARN lines (IdP-managed spaces power mode, no break-glass account on enc, group references to review) are the same as on v0.6.408, so there are no new WARN or ERROR lines.
  • State directory: the start log shows instance state directory (#821) exactly as before.

To check by hand on the dogfoods:

  • the Overview shows the new Coordination: single instance row;
  • /readyz answers like /healthz.

Still pending: the read-only cairnd migrate --to nats --dry-run size measurement on both dogfoods, after Nikola's go-ahead.

Live check for **v0.6.409** (2026-11-18, about 3 min after the dogfood bump, homelab-config cbcf248): - **Pods:** `cairn-enc` and `cairn-openbao` both run `cordy/cairn:v0.6.409`, are ready and have 0 restarts. Both stay in local mode. - **Logs:** the WARN lines (IdP-managed spaces power mode, no break-glass account on enc, group references to review) are the same as on v0.6.408, so there are no new WARN or ERROR lines. - **State directory:** the start log shows `instance state directory (#821)` exactly as before. To check by hand on the dogfoods: - the Overview shows the new **Coordination: single instance** row; - `/readyz` answers like `/healthz`. Still pending: the read-only `cairnd migrate --to nats --dry-run` size measurement on both dogfoods, after Nikola's go-ahead.
Author
Owner

Live check v0.6.410 (homelab-config 6e19bcf), 2026-11-18:

  • cairn-enc: image v0.6.410, ready, 0 restarts.
  • cairn-openbao: image v0.6.410, ready, 0 restarts.
  • WARN/ERROR lines in the first 10 minutes are only the standing ones: IdP power mode, no emergency access (enc), and group references to review (enc 1, openbao 2). Nothing new.
Live check v0.6.410 (homelab-config 6e19bcf), 2026-11-18: - cairn-enc: image v0.6.410, ready, 0 restarts. - cairn-openbao: image v0.6.410, ready, 0 restarts. - WARN/ERROR lines in the first 10 minutes are only the standing ones: IdP power mode, no emergency access (enc), and group references to review (enc 1, openbao 2). Nothing new.
Author
Owner

Live check v0.6.411 (homelab-config 9b16293), 2026-11-19.

Automatic check

  • cairn-enc: image v0.6.411, ready, 0 restarts.
  • cairn-openbao: image v0.6.411, ready, 0 restarts.
  • WARN/ERROR lines in the first minutes are only the standing ones: IdP power mode and no emergency access (enc), and group references to review (enc 1, openbao 2).

To check by hand (Nikola)

  1. Data exports: request a personal data export (Settings › Your data) and an administrator data export on both dogfoods. They should finish and download. Every export with more than one collector failed since #1023; this release fixes it.
  2. Rotation tab: in single-instance mode the Rotation tab behaves as before. The 409 text shows only with several instances.
  3. Not run on the VMs, for a later multi-instance session:
    • the data export takeover under kill -STOP;
    • the trash sweep holder paused;
    • the key back-out (s13 step 7) and cluster forget with a holder;
    • the bell across members, plus a VoiceOver check that the Overview Instances section is not re-announced every 30 s.
Live check v0.6.411 (homelab-config 9b16293), 2026-11-19. **Automatic check** - cairn-enc: image v0.6.411, ready, 0 restarts. - cairn-openbao: image v0.6.411, ready, 0 restarts. - WARN/ERROR lines in the first minutes are only the standing ones: IdP power mode and no emergency access (enc), and group references to review (enc 1, openbao 2). **To check by hand (Nikola)** 1. **Data exports:** request a personal data export (Settings › Your data) and an administrator data export on both dogfoods. They should finish and download. Every export with more than one collector failed since #1023; this release fixes it. 2. **Rotation tab:** in single-instance mode the Rotation tab behaves as before. The 409 text shows only with several instances. 3. **Not run on the VMs, for a later multi-instance session:** - the data export takeover under `kill -STOP`; - the trash sweep holder paused; - the key back-out (s13 step 7) and `cluster forget` with a holder; - the bell across members, plus a VoiceOver check that the Overview Instances section is not re-announced every 30 s.
Author
Owner

Live check v0.6.412 (homelab-config d201778), 2026-11-21.

Automatic check

  • cairn-enc: image v0.6.412, ready, 0 restarts; resumable uploads sealed, spool /data/.cairn/tus (format 3).
  • cairn-openbao: image v0.6.412 (S3), ready, 0 restarts; resumable uploads sealed.
  • WARN/ERROR lines in the first minutes are only the standing ones: IdP power mode and no emergency access (enc), and group references to review (enc 1, openbao 2). No temp-folder (tmpfs) WARN on the S3 dogfood.

To check by hand (Nikola)

  1. Large upload in the web UI: upload a file of a few hundred MB on both dogfoods; it should finish once, without a "(1)" copy and without an earlier version.
  2. Resume: start a large upload, switch the network off for a minute (or close the laptop lid), switch it back on: the upload pauses and goes on by itself (up to 10 minutes).
  3. Restart during an upload: start a large upload, then restart the pod (Argo CD "Restart" on the deployment): the web client resumes after the pod is back; the file is stored once.
  4. Sync client: a desktop sync client keeps working after the update (the change list starts a new epoch once, so the first sync may check everything again).
  5. Still open from v0.6.411: the data exports (personal and administrator) on both dogfoods.
  6. Not run on a dogfood (single instance): uploads across members, the shared change feed across members, migrate with unfinished uploads. These ran on the benchmark VMs (run-u14).
Live check v0.6.412 (homelab-config d201778), 2026-11-21. **Automatic check** - cairn-enc: image v0.6.412, ready, 0 restarts; resumable uploads sealed, spool `/data/.cairn/tus` (format 3). - cairn-openbao: image v0.6.412 (S3), ready, 0 restarts; resumable uploads sealed. - WARN/ERROR lines in the first minutes are only the standing ones: IdP power mode and no emergency access (enc), and group references to review (enc 1, openbao 2). No temp-folder (tmpfs) WARN on the S3 dogfood. **To check by hand (Nikola)** 1. **Large upload in the web UI:** upload a file of a few hundred MB on both dogfoods; it should finish once, without a "(1)" copy and without an earlier version. 2. **Resume:** start a large upload, switch the network off for a minute (or close the laptop lid), switch it back on: the upload pauses and goes on by itself (up to 10 minutes). 3. **Restart during an upload:** start a large upload, then restart the pod (Argo CD "Restart" on the deployment): the web client resumes after the pod is back; the file is stored once. 4. **Sync client:** a desktop sync client keeps working after the update (the change list starts a new epoch once, so the first sync may check everything again). 5. **Still open from v0.6.411:** the data exports (personal and administrator) on both dogfoods. 6. **Not run on a dogfood (single instance):** uploads across members, the shared change feed across members, migrate with unfinished uploads. These ran on the benchmark VMs (run-u14).
Author
Owner

Live check v0.6.413 (homelab-config 4100739), 2026-11-22.

Automatic check

  • cairn-enc and cairn-openbao: image v0.6.413, ready, 0 restarts.
  • WARN/ERROR lines in the first minutes are only the standing ones (IdP power mode and no emergency access on enc, group references to review: enc 1, openbao 2). The old "several instances are a preview" line is gone.

To check by hand (Nikola), enc dogfood with Collabora: editor locks (R8 5.8)
Keep two log windows open: kubectl -n cairn logs -f deploy/cairn-enc --since=1m and the Collabora pod's log. Cairn logs no lock operation name, so check locks on Administration › Locks.

  1. Open a document: the second account's file list shows "In the editor: Nikola".
  2. The second account opens it too: it joins (Join in the editor is first in the row menu), both can type.
  3. The second account writes the file over WebDAV: 423; your own WebDAV client: 423 too (X-Cairn-Code: editing). A manual Lock in the row menu is shown disabled with "Open in the editor".
  4. Leave it idle for more than 15 minutes: a RefreshLock arrives in the Collabora log (note the interval).
  5. Both close: an Unlock; the indicator clears at the next list refresh.
  6. Restart the Collabora pod with the document open, reopen: editable at once.
  7. Lock the file by hand first, then open it yourself: editable; the second account: read-only.
  8. The second account opens as a viewer first, you join as a writer: note whether a Lock arrives.
  9. Administration › Locks shows "Editor" with the people count; End editing: both editors show the end banner and a status message "Editing of … ended for everyone."; no Lock afterwards.
  10. Guest check: while a member edits, a guest opens the document through a link: read-only, and a save by the guest answers 423.
  11. Restart Cairn with the document open in two editors (kubectl -n cairn delete pod <cairn-enc pod>): note what Collabora shows while Cairn is away, and whether the next save after it is back goes through (a single instance keeps no editor sessions across a restart, so the editor may be ended).

Also on both dogfoods

  • Users & access › Access › Two-step sign-in: the "During a coordination outage" block is not shown (single instance).
  • Overview: no change in single-instance mode; Settings › Your data exports still pending from v0.6.411/412.
  • Link passwords: unlocking a password link survives a pod restart now (sealed cookie).

VM-cluster browser steps (signed-in browser, identities): run-r16.md "For Nikola" N1-N4 in .superpowers/sdd/2026-11-21-b12v07/ (bash nik-up.sh / nik-down.sh in scratch/b7e/r16/ start and stop the test cluster).

Live check v0.6.413 (homelab-config 4100739), 2026-11-22. **Automatic check** - cairn-enc and cairn-openbao: image v0.6.413, ready, 0 restarts. - WARN/ERROR lines in the first minutes are only the standing ones (IdP power mode and no emergency access on enc, group references to review: enc 1, openbao 2). The old "several instances are a preview" line is gone. **To check by hand (Nikola), enc dogfood with Collabora: editor locks (R8 5.8)** Keep two log windows open: `kubectl -n cairn logs -f deploy/cairn-enc --since=1m` and the Collabora pod's log. Cairn logs no lock operation name, so check locks on Administration › Locks. 1. Open a document: the second account's file list shows "In the editor: Nikola". 2. The second account opens it too: it joins (Join in the editor is first in the row menu), both can type. 3. The second account writes the file over WebDAV: 423; your own WebDAV client: 423 too (`X-Cairn-Code: editing`). A manual Lock in the row menu is shown disabled with "Open in the editor". 4. Leave it idle for more than 15 minutes: a RefreshLock arrives in the Collabora log (note the interval). 5. Both close: an Unlock; the indicator clears at the next list refresh. 6. Restart the Collabora pod with the document open, reopen: editable at once. 7. Lock the file by hand first, then open it yourself: editable; the second account: read-only. 8. The second account opens as a viewer first, you join as a writer: note whether a Lock arrives. 9. Administration › Locks shows "Editor" with the people count; End editing: both editors show the end banner and a status message "Editing of … ended for everyone."; no Lock afterwards. 10. Guest check: while a member edits, a guest opens the document through a link: read-only, and a save by the guest answers 423. 11. Restart Cairn with the document open in two editors (`kubectl -n cairn delete pod <cairn-enc pod>`): note what Collabora shows while Cairn is away, and whether the next save after it is back goes through (a single instance keeps no editor sessions across a restart, so the editor may be ended). **Also on both dogfoods** - Users & access › Access › Two-step sign-in: the "During a coordination outage" block is not shown (single instance). - Overview: no change in single-instance mode; Settings › Your data exports still pending from v0.6.411/412. - Link passwords: unlocking a password link survives a pod restart now (sealed cookie). **VM-cluster browser steps** (signed-in browser, identities): run-r16.md "For Nikola" N1-N4 in `.superpowers/sdd/2026-11-21-b12v07/` (`bash nik-up.sh` / `nik-down.sh` in `scratch/b7e/r16/` start and stop the test cluster).
Sign in to join this conversation.
No labels
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: Cordy/Cairn#1005
No description provided.