The drawio, excalidraw, and mermaid plugin packages redistribute
third-party material (the draw.io webapp, the Excalidraw editor and its
fonts, mermaid and its dependency tree) without the license texts their
licenses require. Every affected ZIP now carries a licenses/ directory:
- licenses/THIRD-PARTY-NOTICES.txt is generated from the esbuild
metafile (packages/plugins/third-party-licenses.mjs), so the notice
list is derived from what actually lands in plugin.js and cannot
drift the way a hand-maintained list would.
- drawio additionally extracts the upstream LICENSE from the pinned
release tarball (Apache-2.0 requires the text with redistribution);
the extraction guard also heals vendor/ caches from before this
change. The CI fast path (no vendor fetch, no ZIP) is unchanged.
- excalidraw additionally commits curated texts (MIT for Excalidraw,
per-font OFL-1.1/MIT with each font's own copyright statement, plus
a FONT-NOTICES.md attribution table), because neither the npm
package nor upstream ships any license files for them.
The api-side package validator accepts additional ZIP entries, so
installed plugins are unaffected beyond the new files.
Closes#345
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aoPvnakfBP28nAfijgUY9
The paste conversion (issue #30) already handled tables, but any
text/html flavor on the clipboard bypassed it. Code editors (VS Code
with copyWithSyntaxHighlighting) ship the plain text a second time as
styled div/span HTML, so a Markdown table copied there arrived verbatim
while the same text from a plain editor converted fine. Clipboard HTML
without a single structural element (table/list/heading/link/emphasis/
code...) is now treated as equivalent to the plain text; anything from a
rich-text source keeps going through ProseMirror's HTML paste.
Pasting inside a code block never converts anymore -- text is code
there, and the conversion would have split the block around rich nodes.
Hand-typed tables: pressing Enter at the end of a GFM separator row
whose previous sibling is a pipe row replaces the two paragraphs with a
real table (input rules cannot express this -- they see only one
textblock). Conversion is refused inside existing tables; body rows are
then typed cell-wise, with Tab appending rows (#338).
Closes#339
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aoPvnakfBP28nAfijgUY9
Tab used to fall through to the browser's focus navigation everywhere.
Inside tables it now moves cell-wise (Shift-Tab backwards) and appends a
new row from the last cell, Word-style. Outside tables every branch
returns false, so Tab keeps leaving the editor.
Capturing Tab inside tables needs a documented way out (WCAG 2.1.2):
Escape places the cursor after the table -- unlike the arrow keys, which
reach the gap cursor (#335) only from the table's edge cells, it works
from every cell, including from a cell selection. When no textblock
follows the table it falls back to the gap cursor position. The
mechanism is announced to assistive tech via an aria-describedby hint
on the editor surface (visually hidden, de+en).
e2e: cell round trip per Tab/Shift-Tab with typed markers, row append
from the last cell, and the full keyboard-only exit (Escape, then Tab
leaves the editor). The table specs now settle briefly after the insert
-- right after it the collab sync can swallow a click's selection
update, which had the markers landing in stale selections.
Closes#338
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aoPvnakfBP28nAfijgUY9
prosemirror-tables already ships mergeCells/splitCell and the schema
(tableNodes) already carries colspan/rowspan -- only the controls were
missing. Adds the two commands, toolbar buttons whose enabled state
follows the selection (merge needs a multi-cell selection, split a
merged cell), and de+en labels.
Both render paths now carry the spans: docToHtml emits colspan/rowspan
(read mode, exports via the HTML path), and the markdown serializer pads
a colspan with empty cells so every row keeps the table's column count
-- rowspan stays lossy there, GFM cannot express it.
e2e drives merge and split through the toolbar; the cell selection is
made per Shift+Click because a keypress in the same tick as the
preceding click races the editor's post-click rendering (keyboard cell
selection itself works, verified interactively with a settled editor).
Closes#337
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aoPvnakfBP28nAfijgUY9
The delete buttons paired the minus-box with a double arrow (bidirectional
arrows next to the symbol) which reads as "resize/expand", not "delete".
Replace them with axis stripes plus the x delete marker that deleteTable
already established: vertical stripes with x for delete column, horizontal
stripes with x for delete row. Labels/tooltips were correct all along and
stay unchanged.
Closes#336
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aoPvnakfBP28nAfijgUY9
A table (or any other block node without a text position of its own) as
the page's first, last, or only block was unreachable from before/after:
neither mouse nor arrow keys could place the cursor there, so no
paragraph could be created around it.
- add the prosemirror-gapcursor plugin as a TipTap extension (via
@tiptap/pm, no new dependency; schema-neutral, so the editorSchema
drift fence is unaffected)
- style the gap cursor bar in base.css -- the upstream package does not
ship its stylesheet through our import path; the blink animation
honors prefers-reduced-motion
- e2e: keyboard-only round trip that creates paragraphs before and
after a lone table
Closes#335
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aoPvnakfBP28nAfijgUY9
Any authenticated user can invite an e-mail address; the mailed
single-use token lets exactly one signup through even while
registration is closed. Open (pending, unexpired) invitations count
against the new instance setting invitations.maxOpenPerUser (default 5,
0 disables inviting) — plus a 20/day per-user rate limit so a
revoke-and-recreate loop cannot become a mail cannon. Only the SHA-256
token hash is stored (auth-tokens pattern); a failed signup (taken
username) un-redeems the token so the invitee can retry.
Surfaces: invitations section in the user settings (list, invite,
revoke, quota line; wide table in a focusable .table-scroll region),
signup page reads ?invitation=<token> (preview banner, e-mail prefill,
closed-mode gate opens only for a previewed-valid token), admin general
card gets the quota field (flat RHF name per #322; VS-NfD marked and
hideable).
Governance: audit actions invitation.created/revoked/accepted
(catalogue 1.10), VS-NfD profile entry (compliant: 0) + hardening-guide
row, i18n de+en including the invitation mail template.
Tests: api e2e-db (mail link, closed-mode single-use signup with
un-redeem on failure, quota + revoke frees slot, quota 0 = 403, auth
matrix), new web e2e pack invitations.spec.ts (full UI loop through
Mailpit, wired into ci.yml with its own rate-limit reset), a11y scan
waits for the new section. Full api suite (107 files / 607 tests),
auth/admin-settings/a11y packs green against a fresh local stack.
Closes#332
POST /admin/users (Site-Admin guard) creates an account with the same
field rules as self-registration, but active immediately: the admin
vouches for the address, so the e-mail is marked verified and the
personal pond is provisioned exactly like the verify-email path does
(markEmailVerified alone would skip the pond).
The user manager gains a create dialog (useModalFocus/useDismissable,
Field wiring, flat RHF field names per the #322 lesson). New audit
action user.created_by_admin, catalogue bumped to 1.9.
Tests: api e2e-db (create + immediate login + personal pond, duplicate
username 409, non-admin 403), web e2e through the dialog, and the
admin a11y scan now opens the dialog too. Both packs verified locally
against a fresh stack.
Closes#331
Same defect and same fix as the user list's actions cell (#177):
display:flex directly on the override td removed its table-cell
behaviour, so the cell stopped growing to row height and its bottom
border no longer met the row's — visibly uneven separator lines
(Stefan's screenshot from the self-hosting walkthrough). The flex
layout now lives on .quota-row__override-inner.
Measured locally like #177: bottom-delta across all cells of every
quota row was 24–49 px before, 0 px after (override set, so the cell
carries input + two link buttons); admin-quotas e2e pack green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017aviRTgWCcAHUh1SBoxf6P
Two findings from Stefan's manual clean install per the guide, both
ending in an api restart loop that was hard to diagnose:
- #324: the guide recommended `openssl rand -base64 32` for
POSTGRES_PASSWORD, but the compose interpolates the password unescaped
into DATABASE_URL — base64's `/`, `+`, `=` break the URL. Misleadingly,
db stays healthy (it gets the password as a plain env var) while
api/collab/backup crash. Guide and .env.example now recommend
`openssl rand -hex 24` for both secrets and say why; Troubleshooting
gained the symptom line.
- #325: SETUP_ADMIN_PASSWORD's minimum (10 chars,
packages/shared/src/auth.ts) was undocumented, and a violation crashed
the boot with a raw ZodError naming schema fields and i18n keys.
Failing the boot stays — deliberately, no half-seeded instance — but
preseedFromEnv now translates validation errors into operator terms
("Pre-seeding failed: SETUP_ADMIN_PASSWORD must be at least 10
characters. Fix .env and recreate the api container."). Documented in
the guide's first-run section, .env.example, and Troubleshooting; new
test pins the message and that nothing is half-seeded afterwards.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017aviRTgWCcAHUh1SBoxf6P
useDocumentTitle pinned APP_NAME = 'Dorfteich', so every route title —
tab, bookmarks, the window title a screen reader announces (WCAG 2.4.2)
— named the product instead of the operator's instance. The trailing
name now comes from the public branding query, exactly like the TopBar
brand (#306); until the query resolves (or when it cannot, e.g.
maintenance mode) the shipped default keeps the title stable, so an
untouched instance reads exactly as before. The static index.html title
stays the pre-JS placeholder — server-rendering it is #179's territory,
deliberately out of scope (recorded in the issue).
The admin-settings e2e now also asserts the title carries the new name
right after saving, without a reload.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017aviRTgWCcAHUh1SBoxf6P
The pack addresses the registration-mode select by its DOM name
attribute, which react-hook-form derives from the field name — now
`registrationMode` (dot-free, see admin-settings-form.ts). Caught by CI
run 713; the pack needs VS_NFD_MODE stages and was not part of the local
verification set.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017aviRTgWCcAHUh1SBoxf6P
The general and quota cards registered their react-hook-form fields under
the dotted settings keys. RHF treats dots as nested-path separators, so
the form DISPLAYED fine (its getter falls back to the literal flat key)
but typing nested the value ({ instance: { name } }) and the api's strict
PATCH schema rejected the body — none of these fields ever saved through
the UI, on any instance. Found by Stefan on a fresh self-hosted install.
- admin-settings-form.ts: dot-free form model with one explicit mapping
to the dotted settings keys and converters in both directions; the
submit now also carries ONLY the settings these cards edit, so the
internal branding metadata keys never ride along.
- Saving invalidates the branding query too — the TopBar reads the
instance name from it and kept the old name until its staleTime ran out.
- admin-settings.spec.ts (new e2e pack, registered in ci.yml): drives the
rename THROUGH THE FORM — success message, TopBar update without
reload, value survives reload, api returns it. Verified locally to fail
against the unfixed page and pass against the fix. Every existing
admin-settings test patched the api directly, which is why this bug was
invisible to CI.
- admin-settings-form.test.ts pins that no form field name contains a dot
and the mapping round-trips.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017aviRTgWCcAHUh1SBoxf6P
- TAG guidance points at pinned release tags (e.g. v0.14.0) instead of
the pre-release `test` tag; concrete curl commands fetch the three
reference files.
- Backup wording (guide + .env.example) names all four data volumes in
the restore set (uploads, plugins, custom fonts, branding).
- Updating section states the back-up-first step and links the update
runbook.
- Pass the external-authentication variables (OIDC_*, AUTH_LOCAL_ENABLED,
AUTH_PROXY_*) through the reference compose and document them in
.env.example: they were documented in security.md but unreachable from
.env. Empty values count as unset (app-config.service.ts), so the block
is inert until configured.
- New guide section "External authentication (optional)"; neutral
APP_BASE_URL example.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017aviRTgWCcAHUh1SBoxf6P
Under parallel CI load the test repeatedly exceeded the default 5000 ms
per-test timeout (run 685 on main, run 699 on an unrelated PR); the
identical test passed on rerun. Locally it finishes in about 1.3 s, so
30 s is generous headroom, not a mask for a regression.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017aviRTgWCcAHUh1SBoxf6P
Built on #306's storage, serving and crop control — a layer, not a parallel
implementation. `resolveBranding` in shared is the ONE place that answers
"which asset applies here?", and both the sidebar logo and the favicon swap
read it.
The decision most likely to be "fixed" by accident, so it is pinned by name
in `branding.test.ts`: **a logo set belongs to one level and variants are
never mixed across levels.** A pond that uploaded only a light logo shows THAT
logo in dark mode; it does not borrow the instance's dark variant. Decided
2026-08-01 — a logo silently swapping to a different image when the viewer
switches theme is a change nobody ordered, and a design that looks wrong is
more honest than one that is quietly substituted. Only a pond with no logo at
all inherits the instance's set, again as a set. The settings screen warns
about a missing dark variant; it never blocks.
Consequences that fall out of that rule and are easy to get wrong:
- The serving route does NOT fall back when given a pond scope. The caller
already decided which level applies; a "helpful" fallback in the route
would mix variants across levels behind the resolver's back.
- The logo link's accessible name follows the LEVEL: a pond logo is named by
the pond, an instance logo by the instance. It is the link home, and a
link's name has to say where it goes.
- **Charged to the pond's storage quota**, before the write, like attachments.
Without it branding would be a way around the quota, and replacing a logo
repeatedly would consume disk with no ceiling. The replaced asset's bytes
are released FIRST, so re-uploading the same logo costs nothing — and a
refused upload puts the released reservation back, so a rejection cannot
leave the pond with more room than it had.
- **Purge removes the branding files.** The purge standard is absolute: after
it nothing referencing the pond survives, rows or files. Asserted against
the real purge path, not the new code alone.
- Security unchanged from #306 and not relaxed because the uploader is now an
ordinary Pond Admin: SVG refused, magic bytes and IHDR checked server-side,
size caps, content type pinned, no image parsing.
- The favicon swap is driven by the RESOLVED pond, never the raw route
parameter — an unreadable or unknown slug must not leave a stale icon in
the tab. That it happens after first paint is accepted and stated in the
code and the UI: avoiding it would mean server-rendering index.html, which
is #179's territory.
Same audit id as #306 (`branding.changed`) with `scope: 'pond'` — the catalogue
already carries the field, so no version bump.
Verified: api suite 105 files / 592 tests green; 5 pond-branding e2e tests
(pond scope serves the pond's bytes while the instance level still 404s, the
quota is charged and released exactly, SVG refused at pond level, a reader may
read but not change, purge deletes the files); 9 shared unit tests on the
resolution order including both mixing directions.
Deleting a pond already had a strict prompt — typing the pond name, stricter
than a confirm dialog. That was never the gap. The gap is that the person who
deletes it loses access the moment they do: the pond leaves their view, only a
Site Admin can bring it back, and the export is no longer reachable for them.
So the archive is offered INSIDE the deletion flow, before the button.
What it contains, and why it is not the existing export:
- Every page the requester may read, as Markdown, as before.
- **Every attachment of the pond**, not only the embedded ones. An
attachment nobody put on a page would otherwise vanish unnoticed — which
is the whole reason this issue exists.
- `manifest.json`: pond settings (EFFECTIVE, defaults filled in — a
preservation format must not require its reader to know Dorfteich's
defaults), labels, the page hierarchy and sort keys, comments, and
attachment metadata including the #199 hash so a reader can verify bytes.
It extends the #210 manifest rather than adding a second descriptor, and
carries an explicit `formatVersion`.
- `README.txt`, because the manifest is for machines: whoever unpacks a
folder of Markdown a year from now must not believe they hold a one-click
restore.
Decisions worth naming:
- **"Complete" describes the RESULT, not the route.** A pond admin who may
read every page gets `complete: true`; only an archive that actually
leaves pages out is incomplete. The Site-Admin route skips the read filter
(an archive taken before an irreversible purge must not depend on which
ponds the operator happens to be a member of) — those are two different
questions and the first version of this conflated them.
- **The omission is named before the download**, with its number, in the UI
and in the manifest. An archive silently missing content is worse than no
archive, because it ends the search.
- **Not downloading stays allowed.** A pond of test pages should not require
one, and the server cannot tell whether a file arrived anyway — so the
finality is stated in text instead of enforced.
- **A plain link, not fetch-into-a-blob.** The api streams the ZIP; buffering
a whole pond in the tab to draw a progress bar would trade memory for
cosmetics. The browser reports progress and completion; what it cannot say
— that the archive is being BUILT — is announced in a live region.
- Read trail unchanged in kind (ADR 0023): one `export` event per classified
page before any classified byte enters the stream. Attachments never travel
without their page, so the same events cover them.
- New audit action `pond.archived` (catalogue v1.7) with page and attachment
counts, omitted pages, and completeness.
Format documented in `docs/architecture/pond-archive-format.md`, including
what is deliberately NOT in it (history, permissions, trash).
Verified by hand, not only asserted: a real pond's archive downloaded and
unpacked — README, manifest, three page files, the media file; the manifest's
effective settings, per-page classification, the VS-NfD frontmatter and
marking preserved in the classified page's Markdown, and the attachment's
sha256 present. Plus six api tests (including that an unembedded attachment
travels and that a Site Admin gets a complete archive without membership) and
the a11y pack 11/11 in both schemes, which now also scans the pond settings
screen.
Not done, because there is nothing to attach it to: the Site Admin's purge
dialog (#193) exists only as an api endpoint — there is no pond-trash UI in
the web app. The api half is here and tested, so it becomes a link when that
screen is built.
`apps/web/index.html` carries a hard `lang="en"`. The app corrects it at
runtime (#163), but nginx answers every SPA route with that same file, so a
crawler or a no-JS visit — `/public/...` on prod is exactly that — saw `en`
for German content, permanently. WCAG 3.1.1 is about the delivered document,
not the one JavaScript later fixes.
nginx-only, no backend involved: a `map` on `Accept-Language` and a
`sub_filter` in the index.html path. Only the FIRST tag decides, which is
what "the browser's preferred language" means and mirrors #163 — `de-CH`
counts as German, `en-US,de` does not.
`Vary: Accept-Language` is new. The response now genuinely depends on a
request header, and without it a shared cache could hand one language's copy
to the other. Everything else in the location is untouched: same CSP, same
`Cache-Control: no-cache`, same `nosniff`.
The known limit is documented in the config rather than worked around: nginx
cannot read `instance.defaultLocale`, so an unlisted or absent
Accept-Language yields `en` even on a German instance. For public content
that is not the authoritative rendering anyway — the api's server shell
(`/api/v1/public/...`) already renders those with the instance locale.
Verified against a real nginx 1.27 (the image the stage runs) with the
config mounted as-is: `de-DE,de;q=0.9,en;q=0.8` and `de` yield `lang="de"`;
`en-US,en;q=0.9`, `en-US,en;q=0.9,de;q=0.8`, `fr-FR,fr` and a request with no
header yield `lang="en"`; an SPA route (`/public/teich/seite`) negotiates the
same way; the gzipped response is rewritten too, and a JavaScript asset comes
through byte-identical.
An instance had no way to look like itself: the top bar said "Dorfteich"
whatever the operator called their instance, `instance.name` was never
rendered in the running app at all, and there was no favicon anywhere —
`index.html` had no `<link rel="icon">` and `public/` held only fonts and
theme-init.js.
Where the line is drawn, and why:
- **The api never decodes an image.** Cropping, scaling and the conversion
to PNG happen on a canvas in the browser; the api checks the PNG
signature, reads the IHDR dimensions at their fixed offsets and enforces
the caps. An image library would put a decoder in front of
attacker-supplied bytes AND would have to be carried through the
`--network none` offline build. Reading two big-endian integers is not
decoding.
- **SVG is refused**, with its own error message rather than a generic
"not a PNG": it can carry script, and serving it from our own origin
would be a cross-site-scripting vector. An operator who tried one should
learn that it is deliberate.
- **The crop is driven by number inputs, not by dragging.** A drag-only
cropper excludes keyboard and switch users outright; a number input is
arrow-key operable and screen-reader readable without any custom aria.
The resulting pixel size is stated in text, not only drawn as a frame.
- **The variant is chosen by CSS, not JavaScript.** `theme-init.js` has
already resolved `data-theme` before first paint, so the correct logo is
the one painted rather than the one that appears after a flash. Without a
dark variant the LIGHT logo carries both themes — the operator's own
asset shown unchanged beats one they did not choose (the rule #307
extends to ponds). The settings screen warns; it never blocks.
- **The favicon link is static, its resource dynamic.** index.html stays a
static file and the api answers with the uploaded icon or a shipped
default — that route must never 404, or the browser keeps its generic
icon for good. The default is generated by a script from Node's own zlib
(`gen-default-favicon.mjs`), for the same offline-build reason.
- Both favicon sizes are uploaded together: one source, one crop, so the
tab icon and the home-screen icon can never disagree.
- Branding is served WITHOUT a session, because the login screen carries it
and the browser fetches the favicon before anyone signs in. The admin
screen says so — an operator may not expect their logo to be public.
- The metadata is not writable through the settings endpoint: it describes
bytes on disk, and hand-writing it would claim an asset that is not
there.
`./data/branding` follows the three-step rule #303 paid for: env default +
`data-dirs.ts` entry, compose volume (repo AND the stages on ONE), and the
`mkdir`/`chown` line in the api Dockerfile. `data-dirs.test.ts` is new and
closes the hole that made #303's variant invisible: the nightly archive
skips a missing directory WORDLESSLY, so the fence now demands that every
`*_DIR` the backup env declares actually travels in the archive. Verified
against the real defect — removing the line fails it by name.
Audit catalogue v1.7 (`branding.changed`), carrying `scope` from the start
so #307 is the same event with a different scope, not a second id.
Verified: api suite 103 files green (a lone `public-api` ECONNRESET under
local parallel load, green in isolation — the documented local flake);
branding suite 12 tests against a real directory; crop arithmetic unit
tests; a11y pack 11/11 in both schemes; /admin measured at 320px with the
new section (overflow 0); and the whole flow walked in the browser: upload
→ crop 780×180 → stored as 512×118 → logo in the sidebar linking home with
the instance name as its accessible name → topbar wordmark following
`instance.name` → light logo still shown under `data-theme="dark"`.
The font manager's upload live regions made `getByRole('status')` ambiguous
on /admin, and legal.spec.ts — which asserts the legal form's success message
— started failing in the e2e pack. That is the documented trap in CLAUDE.md:
a new label or region makes an existing page-wide locator ambiguous, and the
fix is to scope the SPEC, not to drop the region a screen reader needs.
The section gets a named class for exactly that purpose.
Verified locally against the running stack: legal, fonts, admin-users,
admin-quotas and the a11y pack all pass.
The route-permission fence (#52) failed in CI, not locally: I had run the
fonts and import-export suites, not the full api suite, and that fence needs
a database. `@AuthenticatedOnly()` is the rule the route always meant — a
session, no further permission.
Re-verified with the FULL api suite against a fresh database: 103 files /
575 tests passed.
The backend from #303 could store an operator's font but nothing could
choose one: no list endpoint outside the Site-Admin routes, no @font-face
rules for a family that only exists at runtime, and no management UI.
Found while wiring it up — a real defect in #303, invisible to its tests:
`fontStack` cannot tell an uploaded family from a deleted one, so the PDF
exporter embedded the face and then never named it. Every export of a pond
using an operator font rendered in the system font while the job reported
success. Both `fontStack` call sites now take the uploaded families
(`buildPdfHtml`, `pondFontVariables`); `pdf-html.test.ts` pins the
regression from both sides. Verified against a real Gotenberg: with the
families the PDF embeds PlayfairDisplay-Bold, without them NotoSans-Bold —
that was the whole bug, in one diff of two PDFs.
- `GET /fonts/custom` is readable by any signed-in user, not Site Admins
only: the pickers, the licence page and the injected `@font-face` rules
all need it, and gating it would have forced a second, admin-only UI.
- Bundled and uploaded families are told apart by their `<optgroup>`, not
by a badge — the grouping is then part of the control's semantics, so a
screen reader announces it and the native mobile select keeps it. Within
each source the catalog's category grouping is preserved.
- The delete confirmation names how many ponds use the family and what
happens to them; focus moves to it and back on cancel. Deletion stays
unblocked (the api's decision, #303) — the ponds degrade, they do not
break.
- The licence page grew a second table. That is what makes an attribution
obligation satisfiable: a commercial licence that requires naming the
foundry needs a page to name it on.
Verified in the browser end to end (upload two weights → listed and
rendered in its own font → chosen in a pond → page renders in it → deleted
→ pond falls back): api suite for fonts/export 77 passed, a11y pack 11/11
locally in both schemes, lint/typecheck/i18n:check green.
Found on the real deploy, not in any test: `/data/fonts` in the running
api container was `root:root` and the non-root `node` user could not
write to it. Every upload would have failed with EACCES at runtime while
the api reported ready.
The api Dockerfile already explains the mechanism for uploads and
plugins — Docker copies an image directory's ownership into a fresh named
volume on first mount — and pre-creates them chowned. #303 added
`CUSTOM_FONTS_DIR` to the ENV but not to that mkdir/chown line.
Adds a CI fence so it cannot recur: every `/data/…` path the api image
defaults to must also appear in the mkdir AND the chown. Verified against
the actual defect — removing `/data/fonts` from the chown makes it fail.
An operator holding a font licence could only use it by baking the file
into a custom image, which tied every change to a rebuild and left the
file out of the backup.
ADR 0016 said there is no runtime font management. It also listed this
exact case under "Alternatives considered" — *may become a Site-Admin-
level feature later*. The amendment takes that option and answers the two
objections it raised: licensing risk (Site Admins only, licence recorded
with the family) and file-format attack surface (magic-byte check and a
size cap, never a parse).
- `CUSTOM_FONTS_DIR` (default `./data/fonts`) — a sibling of uploads and
plugins, NOT inside the image-baked `FONTS_DIR`, where a deploy would
overwrite it and no backup would ever see it.
- One list of data directories (`apps/backup/src/data-dirs.ts`) now feeds
both the nightly archive and the restore, so they cannot drift. #306 and
#307 add one line each instead of a second mechanism.
- Both Dockerfiles bake the path. The backup image sets its volume paths
itself ("self-sufficient without compose env" — #71's lesson) and reads
no *_DIR from compose; without the ENV entry the archive would have
skipped the directory silently.
- The PDF path already read WOFF2 from disk at request time, so it only
had to pick the other base directory for a custom family.
- `fontStack`/`fontEntry` take the instance's uploaded families as an
argument — they are runtime data. The catalog is searched first, and a
colliding family name is rejected at upload, so a custom font can never
shadow a catalog one.
- Deletion is never blocked by usage: an unknown family already falls back
to the system stack, so affected ponds degrade instead of breaking. The
count of affected ponds travels into the audit entry.
- Audit catalogue v1.6 (`font.uploaded`, `font.deleted`).
Verified: api full suite against a fresh database, 102 files / 571 tests.
The upload suite writes into a real temp directory and reads the bytes
back off disk, so the storage layer is exercised rather than mocked.
CI 665: the reflow guard itself passed; the run died two packs later on
`fixture login for fixture-admin failed: 429`.
The a11y pack costs one more login since this branch added the reflow
test, and that was enough to exhaust the budget before the VS-NfD packs.
Same trap the workflow already documents for the content and collab
packs — it just needed one more reset, in the place the extra login
pushed it over.
The sorted report finally named it: `table.api-tokens__table` at 833px
wide, with its `.visually-hidden` heading reaching right=737 — exactly
the document's scrollWidth. Same mechanism as the sessions table, a
second table I had not wrapped.
Locally the API-tokens table was empty and therefore narrow, which is why
this only ever appeared in CI. With a token present it reproduces:
without the wrapper 345px of page overflow, with it none.
The feed-token table gets the same treatment — it is built the same way
and would fail as soon as someone holds a feed token with a long name.
The "[in fitting scroller]" marker in the report is misleading for these:
`main.main` is a scroller, but it is `position: static`, so it never
clipped the absolutely positioned heading. Only a positioned ancestor
does — which is what `.table-scroll` now is.
Verified locally against a real stack, with a wide token table present:
reflow guard green, whole a11y pack green in both colour schemes.
CI still reports 737 while the local stack is now clean, and the box list
was capped at 15 entries — all of them nav links clipped by their own
scroller. Whatever pushes the page in CI sits past that cap.
The list is now sorted by reach, marks each entry as either clipped by a
fitting scroller or actually pushing the page, and shows 40.
Found by standing up the local stack instead of guessing through CI.
The DOM tree under `.app-body` shows it in one line:
span.visually-hidden rect=[342,343] pos=absolute
Its right edge is 343, and `.app-body` reports scrollWidth 343 against a
320 client. The table's actions column carries a `.visually-hidden`
heading, which is `position: absolute`. `.table-scroll` was `position:
static`, so it was NOT that span's containing block — the span escaped
the scroller's clipping, kept its static position out at the table's
right edge, and pushed the page.
`position: relative` on the wrapper makes it the containing block, and
the span is clipped like the rest of the table.
This is one cause behind both numbers: 23px locally, matching the
original report, and 417px in CI, where different font metrics make the
table wider and carry the span further out. Chasing them as separate
problems is what cost three CI rounds.
Verified locally against a real stack: the reflow guard passes and the
whole a11y pack is green, 11 tests in both colour schemes.
Two rounds now reported no element past the viewport edge while the
document still claimed 417px of overflow — a combination that rules out
every hypothesis I had, including my own filter.
So stop inferring. The guard now prints the html/body metrics, every
element whose own content is wider than its box (with its overflow-x, so
the intentional scrollers are distinguishable), and every box reaching
past the edge with no filtering at all. Diagnostics ride in the assertion
message, not the compared value, so they show up even when they match.
The previous run came back with an empty offender list and an unchanged
417px overflow: the filter treated everything under a scroll container as
innocent, including the container that was itself too wide. A scroller
only absolves its children when the scroller fits.
It now reports the chain from body down to the widest offender with each
box's width, so the first element wider than the viewport is visible
instead of inferred.
The first attempt fixed plausible suspects. CI measured the actual page
and named something else: six `.settings-nav__link` buttons, 417px of
page-level overflow at 320px.
`.settings-nav` already had `overflow-x: auto`, but as a flex child it
also had the default `min-width: auto` — the min-content width of the
whole jump strip. That forced the column wider than the viewport, so its
own overflow rule never had anything to scroll. `min-width: 0` is exactly
the case CLAUDE.md warns about under Reflow.
The guard now ignores elements that sit inside a scroll container. Such
content is *meant* to be wider than the viewport — reporting it buried
the one finding that mattered under twelve lines of noise, and the cap
truncated the list before it could show anything else.
The table wrapper and the wrapping settings rows from the first commit
stay. Neither was the cause here, but a table cannot shrink below its
min-content width and those rows cannot wrap on their own, so both are
hardening that holds regardless of content.
WCAG 2.1 SC 1.4.10 asks for no two-dimensional scrolling down to 320px,
which is also what 400% zoom on a 1280px screen produces. The layout
skeleton was already hardened for this in #165; the overflow came from
content inside the sections.
- The sessions table cannot shrink below its min-content width — four
columns, one of them the full user-agent string. It now scrolls inside
its own container rather than pushing the page. The container is
focusable with a role and a name, because a scroll area that only a
mouse can reach trades one barrier for another.
- `.settings-checkbox` rows may wrap. The accent swatches have a fixed
size and cannot shrink, so an unwrappable row set a floor for the whole
page width.
Adds a reflow guard to the a11y pack. axe does not cover 1.4.10 — the
criterion is not derivable from the DOM — so this is a separate check,
and it names the overflowing elements when it trips instead of only
reporting that something overflows.
The Obsidian fixture vault contains a note called "Startseite", and the
pond now creates one too — the seeded fixtures use locale `de`. Two
consequences, and the second is the one that mattered:
- the unscoped title locator matched two sidebar entries;
- `/p/<pond>/startseite` no longer belongs to the imported note. The
pond's own start page took that slug, so the import landed on a
suffixed one and the test was about to assert against the wrong page.
Both are fixed by scoping to the mount page and navigating through the
sidebar instead of guessing a slug. The test stays meaningful: it then
clicks a wikilink inside the page content, which the empty auto-created
start page would not have.
CI caught this; the local run passed it. Worth remembering that a
title-based locator can go green by luck.
Every pond created through the api now carries one, and the matrix pond
is created that way. The start page is an ordinary page with no grant of
its own, so it follows the pond-wide permissions: the three member
subjects each see one more, the label-restricted editor too, and the
outsider — who reaches only the explicitly public page — still sees one.
The 429 in the same run was the login rate limit, reached through the
retries of this failure rather than on its own.
Opening a pond landed on whatever sorted first in the sidebar — stable,
but a rule nobody could see, and one whose target moved as soon as
someone added a page ahead of it. New ponds landed on the empty-pond hint
instead of anything useful.
- `startPageId` joins the pond settings. No migration: `Pond.settings` is
already jsonb. It stores an id, not a slug, so renaming or moving the
page keeps it working.
- `PondHomePage` prefers it, but only when the page is in this user's
page list. That list already holds just what they may see, so a start
page hidden by a page-scoped grant — or trashed — falls back silently
instead of landing them on a 404, and it costs no extra request.
- Both creation paths give the pond a start page, titled from the
creator's stored locale. It happens after the creating transaction
commits: the owner's grant is written inside it and permissions cache
per pond, so creating the page any earlier would ask about rights the
grant has not published yet. A failure is logged, not fatal — a pond
without a start page still works.
`PagesModule` imported `PondsModule` without using it. Removing that
vestigial edge let PondsModule depend on PagesModule in the honest
direction instead of tying the two together with forwardRef.
Every pond created through the api now owns a page, which broke eight
suites whose teardown deleted ponds directly — `Page.pond` deliberately
has no cascade, because a real purge removes contents explicitly and
audits it. A shared `deletePondsWhere` helper deletes pages first. Two
tests that counted pages now account for the start page rather than
pretending the pond began empty.
The notification bell sat higher and larger than search and the theme
toggle next to it. The cause was not the glyph: `.notifications-bell__button`
carried its own rules with neither flex centring nor an icon size, so the
svg was laid out inline on the text baseline and rendered at lucide's
24px default instead of the 1.15rem the shared `.icon-button` enforces.
Route every icon-only control through the shared components instead:
- `IconLink` joins `IconButton`, sharing one class helper. Three controls
navigate (pond settings, graph, trash) and are links, not buttons —
without a link twin they would have stayed the one group gluing the
class on by hand.
- 17 hand-applied `className="icon-button …"` usages across nine files
now go through the components, which is what enforces the accessible
name on a control that shows only an icon.
- The bell's unread count reaches assistive technology. The badge sits
inside the control, so `aria-label` hid it and a screen reader
announced "Notifications" without ever saying how many.
An ESLint rule keeps it that way: `icon-button` on a raw button, anchor
or Link is now an error, in both string and template-literal form.
The plugin uninstall button keeps a title that differs from its name (it
explains why a required plugin is locked); IconButton spreads rest last,
so the explicit title still wins.
Also drops the graphify block from CLAUDE.md — it duplicates the
workspace-level instructions.
Operator decision at the ADR 0020 acceptance: verification is
subkey-only now instead of waiting for the stated 2026-11-01 expiry.
Links in digest mails sent before the #188 key separation stop working;
recipients use the in-app notification settings. A regression test pins
that the legacy derivation (root key + purpose prefix) can never verify
again; security.md records the removal.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
The install path records the SHA-256 of the delivered bundle ZIP
(plugins.bundle_hash; pre-#232 installs show it as unknown until
reinstalled). plugins.allowlist in instance_settings names permitted
ids with their pinned hashes: empty (default) = not enforced, existing
instances unchanged; non-empty = installs of unlisted or deviating
bundles are rejected (plugin_not_pinned / plugin_hash_mismatch, 403),
and an installed plugin outside the list or with a deviating hash does
not load — absent from pond mount lists, frame/assets 404. Every
rejection is audited (plugin.rejected, catalogue v1.5). A version bump
changes the hash and therefore requires an explicit re-pin — the
intended friction (ADR 0025). Admin UI shows observed vs pinned hash
per plugin with pin/re-pin/unpin. Scope stated honestly in
plugin-architecture.md: the pin answers "is this the reviewed bundle";
post-install disk tampering is platform integrity (ADR 0019), sandbox
containment stays the sandbox's job. Hardening guide row + catalog
advisory triage; residual risk R-03 resolved. e2e: empty-allowlist
compatibility, pinned load, unpinned and tampered installs rejected and
audited, pin drift blocks loading while the admin still sees the
mismatch, version bump needs re-pin. Full api suite 101 files / 561
green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Stefan reviewed and accepted all nine VS-NfD ADRs one by one. Two
adjustments from the review: ADR 0021 decision 3 now states the #216
refinement in the decision itself (PAT/feed-token issuance stays
available to IdP-authenticated sessions — API authorization under its
own switches, not interactive sign-in) instead of contradicting the
later Decisions section; and the ADR 0020 dual-verify window will be
removed early (issue #296) rather than waiting for its stated expiry.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
In enforced mode the ONE settings write path every caller uses rejects
catalog-violating values with the stable code vs_nfd_profile_violation
(403 — the request is well-formed, the policy says no). Existing
violating values are reported at startup (log line, database-less boots
must not fail) and on the admin card, never auto-changed. The UI
renders as in hidden (#245 already keys on hidden|enforced). The
hardening guide now names enforced as the recommended mode for VS-NfD
reference operation. Tests: violating write rejected with the stable
code and nothing stored; compliant writes pass; the same violating
write passes in marked and hidden (own app boots); pre-existing
violation reported and untouched. Full api suite 100 files / 555 green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
In hidden (and later enforced) mode, catalog-listed controls whose only
purpose is enabling a violation are not rendered while their saved value
is compliant (the four master switches, the Nextcloud backup block);
value-listed selects keep only their compliant choices (registration
mode, new-page classification, upload policy, SVG policy). Every
affected section shows one accessible policy note (i18n de+en) so
policy is distinguishable from missing features. A value that was
already violating is surfaced exactly like in marked — never silently
hidden. The API stays unchanged; enforcement is #246. e2e: hidden half
of the marking pack (rows disappear, note visible, already-violating
row stays marked, axe WCAG A/AA clean) — verified live locally; CI runs
it against a second api (VS_NFD_MODE=hidden, same database) behind its
own static server.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Every catalog-listed control on the admin surfaces carries an accessible
deviation marking in mode marked: text + icon under the control (never
colour alone), part of the control's accessible description
(aria-describedby), i18n de+en. The check runs against the CURRENT
control value, so a violating choice is marked before saving. Covered
controls: registration mode, new-page classification, upload policy,
SVG policy, the four master switches (api/mcp/feeds/plugins), the legal
texts (violating while empty), and the Nextcloud backup toggle on the
system panel. The profile card (#243) gains the warning summary and the
hardening-guide reference. e2e: new vs-nfd-marking pack (marked half in
CI — the e2e api now runs VS_NFD_MODE=marked, which also puts the
marked state into the a11y admin scan; off half in local default runs;
both halves verified live). hidden/enforced follow in #245/#246.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
The deployment declares through VS_NFD_MODE (off | marked | hidden |
enforced, default off) how the application treats configuration that
violates the VS-NfD reference profile — deploy-level like
BACKUP_ALLOWED_TARGETS, so a compromised Site Admin cannot widen it.
The catalog in shared (vs-nfd-profile.ts) is the single source of
truth: every profile-relevant setting with a decidable compliant value,
judgement calls in an explicit advisory list, and a fence test parsing
the hardening guide's reference tables so neither can drift (pattern
#201). The api evaluates the catalog against the typed settings
registry and validated env and exposes mode + verdict on
GET /admin/system/vs-nfd-profile; the admin settings view shows the
card whenever the mode is not off. Display only — the treatments land
with #244–#246 (ADR 0027, proposed).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Adds docs/operations/update-runbook.md (obtain, verify by digest, back
up, apply, verify, roll back) with the migration behaviour stated
explicitly: a failed migration rolls back its own transaction but is
recorded in _prisma_migrations and blocks every further migrate deploy
(P3009) — including a re-deployed old image — until migrate resolve
--rolled-back; semantically irreversible migrations have exactly one way
back, the pre-update backup set. No rolling updates on a compose stage.
Rehearsed in the isolated environment of #220: regular update to a v2
image set, then a deliberate failed-update (P3018 division by zero,
schema change proven rolled back) with image-rollback-alone shown
insufficient and the documented recovery executed. Protocol:
docs/vs-nfd/98-update-rollback-protokoll.md. ADR 0024 decisions 5+6
recorded as executed; operations handbook and restore runbook updated;
plan checkbox P1-3 ticked.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Full deployment exercised in a compose stack whose networks are all
internal: true — setup, login, live collaboration, search, upload, all
export formats, backup and restore. tcpdump full capture on both
bridges: zero packets leave the isolated subnets; the only outbound
attempt the application makes is SMTP, which fails contained in the
outbox (5 retries, then FAILED) while the instance stays fully
functional. The restore finding became #288, fixed earlier in this
chain and re-verified in the same stack. Plan checkbox P1-3 and the
I-28 open question ticked; operations handbook airgap section updated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Since #224 read_events is partitioned; the dump carries per-partition
primary keys as own entries, and pg_restore --clean emitted DROP
CONSTRAINT against inherited constraints, which PostgreSQL refuses. The
restore then reported FAILED although the content was restored. Dropping
and recreating the public schema first makes every --clean drop a no-op
and the restore faithful: objects created after the backup no longer
survive. Verified in the isolated environment of #220 (set
20260731-132200, exit 0, readyz green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
The ADR 0024 §4 decision, taken explicitly and both ways: customers
OPERATE prebuilt digest-pinned images (no customer-side build), and
ADDITIONALLY the workspace build is verified to work with networking
disabled - so site-local patching stays possible without internet.
Evidence (docs/vs-nfd/96-offline-build-protokoll.md): pnpm install
--offline --frozen-lockfile plus pnpm build under docker run
--network none (node:22.15.1-alpine + pnpm 11.9.0, the pinned
toolchain), reproduced twice from clean checkouts with identical
results. The offline kit is the pnpm store (~870 MB) plus the build
user's ~/.cache (~460 MB - the prisma engines live there; without the
cache the prisma postinstall fails offline).
The one network dependency found and bounded: the drawio plugin's
installable ZIP fetches its pinned vendor tarball on first build.
Deploy images contain no plugin ZIPs, so the delivery-relevant build is
fully offline (CI=1 skips the fetch, as in CI); an offline ZIP build
pre-seeds the tarball into packages/plugins/drawio/vendor/.
Also catches up the operations manual's scheduler-job table to 10
(read-trail-maintenance was added in #224 without the row here).
Refs #219.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Airgapped sites pull from their own registry (ADR 0024). The image list
is GENERATED (deploy/scripts/list-images.sh resolves the compose file
incl. the caddy profile) so a mirror can never silently miss a service;
third-party images gain a configurable ${REGISTRY_PREFIX:-} in the
compose file (digest pins unchanged - Docker verifies the same sha256
regardless of which registry serves it), own images keep IMAGE_PREFIX;
no image reference is ever edited per site.
Step-by-step procedure in deploy/stages.md 5b: generate list, copy
digest-preservingly (docker buildx imagetools create; plain
pull/tag/push as the documented fallback - the digest comparison closes
the loop either way), verify the digest in the mirror against the pin,
point the deployment via REGISTRY_PREFIX/IMAGE_PREFIX.
Executed once end-to-end and recorded as assessor-facing evidence
(docs/vs-nfd/95-mirror-protokoll.md): all four third-party images
mirrored digest-identically into a local registry:2, plus
dorfteich-api:v0.12.0 (sha256:576f1646... identical on both sides; the
imagetools stall against the Gitea registry is recorded with its
workaround). Operations manual's airgap section now lists the mirror
part as available.
Refs #218.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Declarative instance setting idpMapping.rules turns ID-token claims into
pond roles and the site-admin flag on every OIDC login — configuration,
not code. Mapped grants travel through the SAME GrantsService path as
manual ones (permission cache invalidated, collab access notify fires so
live sessions revalidate — asserted by test), never raw rows.
Ownership makes precedence explicit: role_grants.origin marks mapped
rows, users.is_site_admin_managed marks a mapping-set admin flag. The
mapping only creates and revokes what it owns — manual wins: hand-made
grants and hand-promoted admins are never revoked by a missing claim (a
manual toggle clears the marker and takes ownership). Removal of a claim
revokes the mapped grant and the managed flag on the next login. Every
mapping-driven change is audited with origin idp_mapping.
Failure containment: unknown pond slugs and the last-Pond-Admin
protection log-and-skip — a mapping problem must never become a login
lockout. Tests drive real OIDC logins against the fake IdP with group
claims: grant + working access, revocation incl. notify, manual-wins,
managed site-admin promote/demote/hands-off.
Documented in permissions.md (own section), ADR 0021, data-model.md and
the hardening guide (care rule: same PR).
Refs #217.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
The deploy-level realization of auth.local.enabled (ADR 0021): FALSE
answers 404 on every local credential flow — login, signup, e-mail
verification, resend, password forgot/reset/change — enforced centrally
in the auth guard via the @LocalCredentialFlow() marker before any
session or CSRF logic runs. Deploy-level on purpose: a compromised Site
Admin cannot reopen the local path, so the runtime-flip residual risk
from ADR 0021 does not materialize (R-02 closed in the risk list).
An enumeration fence fails when an auth route is neither marked nor on
the reviewed allowlist, so a new credential flow cannot ship unswitched.
Stated decisions, each tested: sessions/logout keep working for
externally authenticated users; PAT and feed-token issuance stays
available (API authorization under its own switches, not interactive
sign-in). Bootstrap: complete setup (or SETUP_ADMIN_* pre-seed) before
flipping; the api warns at boot when local auth is off with neither OIDC
nor proxy auth configured. GET /auth/methods reports local:false and the
login page hides the local form and credential links.
Hardening guide: the planned auth.local.enabled row moves from 1.3 into
the live deploy table with the bootstrap ordering, and the verification
checklist gains the login-404 probe.
Refs #216.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
For perimeters that authenticate before the application (ADR 0021 §4).
Off unless BOTH AUTH_PROXY_HEADER and AUTH_PROXY_TRUSTED_PEERS are set —
nothing about the header is guessed. The peer check runs against the TCP
peer address only (a forwarded header is attacker-influenced): a request
carrying the header from any other peer is rejected outright and audited
as auth.proxy_rejected (catalogue v1.4) — that is a spoof attempt, not a
misconfiguration — even when a valid session cookie rides along. From a
trusted peer the header IS the identity; a session cookie never
escalates beyond it; with the feature off the header is inert.
Mapping is explicit (AUTH_PROXY_MAP: username or e-mail); deliberately
no just-in-time creation — the header carries no verified address. The
mTLS variant (AUTH_PROXY_MODE=mtls-dn) maps the configured attribute
(default CN) out of the certificate subject DN the TLS terminator
forwards, under the same peer rules. Session-less proxy requests key the
read trail per user (user:<id>).
The trust boundary is stated in security.md (the section an assessor
reads closest), the VS-NfD security documentation and the hardening
guide's deploy table. Tests cover all four decisions: off = inert,
trusted peer authenticates (username and DN mapping), untrusted peer
rejected + audited, no escalation past a session cookie.
Refs #215.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
External authentication (ADR 0021) built on jose (#188's vetted library)
plus fetch — no new dependency enters the supply chain for a security
base function. Discovery-configured; ID tokens validate against the
IdP's JWKS under an explicit RS256/ES256 allowlist with issuer,
audience, expiry and nonce binding. State, nonce and the PKCE verifier
travel in a signed HttpOnly Lax cookie keyed by a dedicated HKDF
purpose (oidc-state, ADR 0020).
Deploy-level configuration (OIDC_ISSUER/CLIENT_ID/CLIENT_SECRET/SCOPES/
PROVIDER_LABEL): who authenticates users is a platform decision. The
login page discovers the provider via GET /auth/methods and renders the
SSO button (i18n de+en).
Identities use the existing slot (provider oidc:<issuer>, subject from
the token). First login creates the account just-in-time — ACTIVE and
mail-verified only when the IdP asserts a verified address. An existing
local account is NEVER adopted silently by e-mail (account-takeover
path): login refuses with oidc_link_required and the owner links
explicitly via GET /auth/oidc/link (audited auth.identity_linked,
catalogue v1.3). Sessions come from the one existing session service.
Tests run the full flow against a protocol-faithful fake IdP: PKCE
verifier at the token endpoint, JIT creation incl. personal pond,
invalid state/nonce/signature/issuer/audience/expiry each rejected, the
linking refusal and the explicit link flow. Verified end-to-end against
a real Keycloak 26.0 (repeatable procedure documented in security.md
§External authentication).
Refs #214.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
New instance switch readTrail.enabled, default OFF: read logging is
employee monitoring in a works council's eyes — an ordinary instance
must not surveil reads. Off means no event is written ANYWHERE (no row,
no stdout line, verified by test); the api announces the switch position
once per boot, so an eventless trail is never ambiguous — a gap reads
as "was off", never "was lost".
The written purpose limitation ships as section 7 of the VS-NfD
security documentation (#228): what is recorded (no content, no titles,
no IPs, no fingerprinting), why (evidence for reads of marked content
only — variant A is the technical anchor of the promise), who may read
it (Site Admin, API-only), for how long (readTrail.retentionDays,
audited pruning), and what it may NOT be used for (no performance or
behaviour monitoring). The hardening guide's reference configuration
turns the trail on (reference value true) and points to that text; the
existing trail suites now enable the switch explicitly.
Refs #225.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Convert read_events to monthly RANGE partitions on occurred_at, with a
DEFAULT partition as safety net: a lagging maintenance job must never
turn the trail's hard-failure semantics into an outage for classified
reads. The dedup unique pair (#223) moves to per-partition indexes
(PostgreSQL cannot carry it on the parent); a bucket spanning a month
boundary may record one duplicate — over-recording is acceptable, gaps
are not.
New daily job read-trail-maintenance (job-count fence 9 -> 10) creates
months ahead — each with its dedup index — and applies the trail's own
retention readTrail.retentionDays (default 365, deliberately independent
of audit.retentionDays): whole expired months are DROPped without
scanning, remainders deleted by range, every run audited as
read_trail.pruned (catalogue v1.2; the fence regex now admits an
underscore namespace).
Site-Admin query path GET /admin/system/read-events answers "who read
page X" and "what did user Y read" within a period — API-only by
design, documented. Growth measured and documented in data-model.md:
~1 MB per 1000 events including indexes.
Tests: retention pruning + audited deletion + admin queries on the
shared database; the partitioned shape, per-partition P2002 dedup,
months-ahead creation and DROP-based pruning against a fresh database
built by the real migration chain.
Refs #224.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
One event per (session, page, channel) within an aligned window of
readTrail.dedupWindowMinutes (default 5): buckets are
floor(epoch / windowSeconds), and a unique (dedup_key, window_bucket)
pair collapses concurrent duplicates race-free at insert time — the
first access in a window is always recorded, a later duplicate lands on
the unique violation and is skipped quietly (a skipped duplicate is not
a gap; only real write failures still abort the read). Each row carries
windowSeconds, so the evidence states it represents a window, never a
request count.
Reconnects within a window stay one event; a new session records again
even for the same user; channels never collapse into each other; the
page-less attachment key uses the documented `-` placeholder. Load
evidence: 30 collab-token renewals inside one window produce exactly one
event (test), bounding a live editing session at ~12 events/hour/page.
Window semantics documented in ADR 0023, the VS-NfD security
documentation (#228) and as a hardening-guide line for the new setting
(care rule: same PR).
Refs #223.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
Instrument every full-content read channel for pages with
classification = vs_nfd (ADR 0023, variant A): SPA state fetch and read
rendering, public JSON content, no-JS shell, expanded embeds, public API
GET (incl. the MCP read_page path and write echoes), attachment download
under the #212 effective classification, all export shapes (markdown,
pond ZIP, account data export, queued docx/odt/pdf at enqueue), and
collab-token issuance as the api-side proxy for the WS join.
Events land in the new read_events table (no FKs — evidence survives
page purges and hard user deletions) with actor, session key
(session:/token:/job:/anon), page, pond, channel and the classification
at read time. Recording failures are NOT swallowed: a failed write
aborts the read (hard failure, the deliberate contrast to AuditService —
decision recorded in ADR 0023 and security.md §Logging, together with
the recorded residuals: content fragments and feeds).
One e2e test per channel proves both the event and its absence for
unclassified pages, plus the hard-failure semantics.
Refs #222.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUtYMxwTCMHG9mVHnwbFg8
docs/vs-nfd/80-grundschutz-mapping.md against the Edition 2023 texts
of both building blocks (fetched from the BSI single PDFs, edition and
retrieval date stated; the dropped requirements of APP.3.1 are listed
as such, CON.11.1's 18 requirements are all Basis). Every requirement
classified as product / operator / n.a.: product rows point at code,
configuration and tests (auth+rate limits, upload controls, security
headers with the honest HSTS-at-the-proxy split, marking = the whole of
M26 under CON.11.1.A7 incl. the answered does-the-marking-carry-a-
security-function question); operator rows say what we hand over
(copy list, procedures, network plan, SBOMs); n.a. rows are argued via
the delimitation statement (no §52 security functions, no built-in
remote maintenance). Open requirements point at their closing issues
(M27/M28/M29/M32), so the document doubles as the gap list; the
never-scheduled external pentest is stated honestly.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
docs/vs-nfd/90-restrisiken.md: nine entries, each with risk, why it is
accepted, compensating control and decider — unmarked attachment
content (#212), local auth not yet switchable incl. the open runtime-
flippability question (#216), deferred plugin hash pinning (#232), the
one-time git-history secret check with its pattern caveat (#198),
digest-mail titles (I-23, revisit M32), page_links slug residue (I-24),
the IndexedDB endpoint copy (I-25), deliberately unscheduled features,
and the Site-Admin read bypass. Binding same-PR maintenance rule
stated; referenced from the delimitation statement and consumed by the
Grundschutz mapping.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
docs/vs-nfd/50-haertungsleitfaden.md: one adoptable profile — every
entry with the exact switch name, value, default and the reason, split
into instance settings (registration closed, api/mcp off, feeds off,
plugins off, classification defaults vs_nfd + upload block, svg reject,
minimal extension list) and deploy-level configuration (empty
BACKUP_ALLOWED_TARGETS enforces backup-local-only outside Site-Admin
reach; tightened session hours; SMTP deliberately unconfigured with the
consequence stated honestly). auth.local.enabled is listed as the one
pending row (#216) with its compensation until then; the guide states
the binding updated-in-same-PR rule for every future switch. Includes
an operator verification checklist (four unauthenticated 404 curls +
readyz + admin spot checks). Cross-referenced from the delimitation
statement (file names made concrete) and consumed by the Grundschutz
mapping (#230).
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
docs/vs-nfd/70-betriebshandbuch.md: installation as run on the real
stages (airgap variant explicitly pending #218-#221 with what already
exists as groundwork), update/rollback incl. the no-down-migrations
caveat, backup/restore with the ADR-0026 target allowlist and the
rehearsed monthly restore drill (evidence: logs on #98), the full
scheduler-job table (cadences verified against code), the deletion-and-
destruction chapter built on the #228 copy list (per content type:
what deletion reaches, what remains, immediate-destruction path,
decommissioning), and role separation incl. the deliberate limits of a
Site Admin and the honest note that Site Admin read-bypass makes the
content/platform split non-absolute app-side. Every procedure carries
its evidence level (erprobt / nicht geprobt / offen) — nothing claimed
above what was actually executed.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
docs/vs-nfd/60-sicherheitsdokumentation.md: component diagram with per-
service purpose and privileges, network plan digit-exact against the
deploy compose (127.0.0.1-only app bindings, internal-only data zone),
data-flow diagrams (auth, realtime editing incl. LISTEN/NOTIFY and the
60s collab token, export via the pinned sidecars, backup incl. the
ADR-0026 allowlist, and every read channel), named trust boundaries
(reverse proxy, plugin sandbox, outbound SMTP/mirror), and the complete
list of content copies — in-database, on-volume and outside the
instance — that the deletion concept in #229 builds on. Mermaid only,
German (assessor audience), with the maintained-in-same-PR rule stated.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
The attachments panel of a classified page shows a persistent notice
naming the consequence (de+en): the file inherits the page's
classification but its content carries no marking (#212). The new
instance setting classification.uploadPolicy (default warn, documented;
the VS-NfD reference configuration blocks, #227) hardens the warning
into a server-side rejection (403 classified_upload_blocked) — enforced
in the upload service, not only in the UI. Tests: warning visible in the
local attachments pack; block enforced server-side with warn/block both
ways and open pages unaffected.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
Downloads whose effective classification is vs_nfd carry the documented
VS-NfD_ filename prefix (single source classificationFilenamePrefix() in
shared; ADR 0022 records the short form for file names). Effective
classification: the linked page's level; an attachment with unset pageId
(paste-then-insert, pond-level) FAILS CLOSED to the highest level of any
live page in its pond. The pond export ZIP adds a sibling
<file>.classification.txt companion with the full marking for classified
media, next to the manifest entry (#210). Documented in operations.md,
incl. the deliberate residual risk: the file's own content carries no
marking (recorded on #231, not hidden). Tests: prefixed classified
download, unchanged open download, fail-closed orphan both ways, ZIP
companion + manifest level.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
Feeds: classified entries carry a standard Atom <category>
(term=level, scheme=urn:dorfteich:classification, label=the fixed
wording); the feed document states the highest contained level once;
all-open feeds carry none. Public API: page representations (list+get)
gain the classification field, OpenAPI + public-api.md documented.
Search: every hit carries the level and the palette renders the marking
with the snippet (compact form of the banner, text token only). No-JS
shell: banner above and below the content, own markup for the separate
render path; unclassified pages unchanged everywhere. One test per
channel (feed categories + count, public API list/get with the switch
on, search hit levels, shell top+bottom).
Also: fidelity CI sidecars get per-job container names — the fixed
names collided across parallel runs on the shared host (run 547's red
fidelity job; a fixed-name cleanup could even kill a sibling's live
sidecars).
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
A classified page's .md carries the level in YAML frontmatter AND the
marking line at top and bottom; unclassified files are byte-identical to
before. Every pond archive (incl. the per-pond folders of the account
data export) ships a manifest.json listing each file with its level and
stating the highest level once at archive level — media inherits the
highest classification among the readable pages referencing it
(fail-closed). Round trip: the importer recognizes exactly our
frontmatter block, strips it plus the imprint lines, and creates the
page at least at the imported level (content must not escape its marking
by traveling through a ZIP) — pinned by unit and e2e round-trip tests.
Foreign frontmatter passes through unchanged; the Obsidian vault import
keeps its own frontmatter modes.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
reference-vs-nfd.docx/.odt ship as derived binaries: the pinned pandoc's
default reference documents plus a header and footer with the marking —
part of the document's page setup, so it repeats on every page in Word
and LibreOffice and is not deletable body text. Source of truth is
scripts/gen-classified-reference-docs.mjs (wording from shared
classificationMarking(); maintenance documented in assets/README.md).
The converter passes reference docs to pandoc-server via in-request
files + reference-doc; the worker attaches them for marked docx/odt jobs
(job option {marking}, as in #208). Unclassified exports pass nothing
and are unchanged (pinned by fake-converter test). Fidelity suite
asserts against real pandoc 3.6 that marked outputs carry the
header/footer parts and unmarked ones do not; per-page repetition
verified via LibreOffice 25.8 headless PDF (5/5 pages, 2 markings each,
both formats). Word: quick manual look pending (sample files in the
workspace), procedure documented in assets/README.md.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
A classified page's PDF export carries its marking as a job option; the
renderer hands it to Gotenberg's Chromium header/footer templates, so it
repeats on every page — bold centered in the running header and next to
the existing page numbers in the footer. Unclassified pages send exactly
the pre-#208 forms (unchanged PDF, asserted by the fidelity smoke and a
lastMarking=null check). New real-Gotenberg fidelity test asserts the
marking appears twice on EVERY page of a multi-page render while the
document-level header keeps working.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
First @media print support at all: page size/margins, navigation and
interactive chrome suppressed, break behaviour for headings, tables,
code blocks, figures and plugin blocks. The VS-NfD marking runs as
header AND footer on every sheet via a real-table PrintFrame whose
thead/tfoot browsers repeat per page — @page margin boxes are
unimplemented and position:fixed places unreliably in both engines
(verified empirically); on screen the table chain renders as plain
blocks, so nothing changes visually. Verified as PDF-from-browser in
Chromium 140 and Firefox 153 (2 markings on every page of a multi-page
document); the repeatable procedure is documented in
apps/web/e2e/README.md. Unclassified pages print without a marking.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
ClassificationBanner renders the fixed ADR-0022 wording above and below
the content in reading view, editor and public page view; unclassified
pages show nothing. Announced to assistive tech via a localized hidden
prefix (de+en); styled from the plain text token only, so contrast holds
in both themes and under every accent with no new color pair. Public
content endpoint now carries the classification. New seed fixture
classified-note; a11y pack asserts banner top+bottom and axe-clean in
light and dark.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
New pages take max(instance default, parent level); moving a subtree
under a higher-classified parent raises every member below that level.
No move-like path (reposition, trash-promote, purge-promote) lowers a
level as a side effect — pinned by test. Raising is ordinary editorial
work; lowering requires the dedicated capability canLowerClassification
(pond-wide Pond Admin) in the central permission model. Both directions
are audited (page.classification_raised/_lowered, catalogue v1.1) with
old value, new value, actor and page.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
Enum field on Page (UNCLASSIFIED default, VS_NFD), migration backfills
existing pages. New pages take the instance-wide default from
classification.newPageDefault (admin-visible, de+en). The value rides in
every PageView, so no channel needs an extra request. The field is a
marking, not a protection mechanism: a test pins that permission
decisions are unchanged by it. The marking wording is fixed in ADR 0022
and sourced solely from classificationMarking() in @dorfteich/shared.
Co-Authored-By: Claude Fable 5 (1M context) <noreply@anthropic.com>
The four third-party images in the deploy compose (postgres, pandoc,
gotenberg — previously a floating MAJOR tag —, caddy) are now
name:tag@sha256 pins; the tag stays for readability, the digest decides
what runs. The pinned digests are exactly what the stages already run
(verified against the live containers' RepoDigests on ONE), so the next
recreation is byte-identical. A new early CI step fails on any
third-party compose image without a digest; compose.dev.yml is a local
convenience and deliberately exempt (its node helpers now follow the
#236 pin). Update + rollout procedure in deploy/stages.md — CD does not
sync stage composes, so the hand rollout to test/int/prod is part of
this issue's definition of done.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
The 36 audit action ids become a typed union (AUDIT_EVENTS in
audit-actions.ts) — an uncatalogued id is now a compile error; every
existing id keeps its name. The published, versioned catalogue
(docs/architecture/audit-events.md, v1.0) documents per event: trigger,
severity, actor and target semantics, and every field, plus the
compatibility promise (ids are never repurposed; retiring keeps the row
forever) and the stable stdout field set. audit-catalogue.test.ts is
the fence: it parses the document's event tables and fails when ids or
severities drift from the code (negative case verified). Audit stdout
lines now carry the catalogue severity as a routing hint — pino level
stays 30 so transport is unaffected; no DB migration.
Forwarding path documented: container stdout -> operator's collector;
deliberately no application-side syslog client.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
plugins.enabled (instance setting, default on — plugins predate the
switch; the VS-NfD reference configuration turns it off) makes every
plugin surface answer 404 via a shared guard: Site-Admin
install/list/mode, pond activation and plugin list, the sandbox frame
and asset routes. The dropzone watcher quarantines drops instead of
installing. Deliberately NOT guarded: the authenticated
fallback-metadata route — it serves no plugin code and existing
plugin_block nodes need it to render their declared fallback (an image
fallback degrades to the neutral placeholder while off, because its
bytes live on the disabled asset surface). The editor offers no plugin
blocks because the pond plugin list is one of the 404ing surfaces.
Admin settings panel gets the toggle (i18n de+en) with the documented
api-restart note (in-process settings cache).
Answers "code execution inside the zone?" with one verifiable
off-switch instead of per-plugin trust machinery (#232, ADR 0025).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Every upload stores the SHA-256 of its bytes, computed from the
in-memory buffer that is written — never by re-reading disk. Every
download re-hashes the stored object BEFORE the first byte leaves
(memory bounded by the max_file_bytes quota that gated the upload) and
fails closed on mismatch with attachment_integrity_failure; the
mismatch lands in the audit trail as file.integrity_failed with both
hashes. Detection of payload manipulation is the one integrity duty
par. 52 VSA leaves with the application — only it knows what the file
should be.
Pre-#199 rows are hashed by a bounded, idempotent backfill that rides
the existing nightly orphan-file-sweep job (no new scheduler job, job
fence untouched); unreadable files are logged and retried, never
silently skipped, and null-hash rows are served unverified only until
the backfill reaches them. Operator runbook note in security.md
(restore from backup, re-download, audit entry carries both hashes).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
The release run now generates CycloneDX 1.6 SBOMs with a pinned
anchore/syft container — one per released image (scanned from the
freshly built image tar, OS packages included) and one for the pnpm
workspace (from the lockfile) — plus the full pnpm licenses report, and
attaches everything as build artefacts BEFORE publishing the release,
so a red gate stops the release. Runner constraints dictated the
mechanics (documented in the workflow): the job talks to the HOST
daemon, so files travel into the syft container via docker cp and
images via docker save to a tar copied the same way (syft cannot read
a tar from stdin — verified).
scripts/check-licenses.mjs is the documented license policy: permissive
allowlist, MPL-2.0/CC-BY-4.0 with recorded reasoning, per-package
exception table (khroma: MIT text shipped, metadata missing). CI runs
the gate on every PR (pnpm licenses:check); positive and negative case
tested locally, both SBOM paths tested against real images/lockfile.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
release.yml and drill.yml ran throwaway `docker run node:22.15-alpine`
helpers outside the pin; the drift check now also fails on any
node:<other>-alpine reference in .gitea/workflows.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
.node-version (22.15.1) becomes the single authoritative Node version:
CI/CD select Node only via node-version-file, every Dockerfile pins
node:22.15.1-alpine, and the engines floor in package.json states the
same version (open-ended upwards so a newer local Node keeps working —
reproducibility rests on images and CI). An early CI step fails on any
drift between those places; update procedure in operations.md
(Update strategy). Precondition for the reproducibility claim in #219.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
The row is only the index of a wikilink whose text (slug = title)
remains visible in the linking page's own content either way; deleting
the index would remove nothing the system still shows while breaking
phantom-link re-resolution. Kept as an accepted residue, reasoning in
operations.md (deletion/purge section) and recorded on #231.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Sent mails were kept forever, and digest bodies name page titles and
actors — an unbounded copy of content-adjacent data. A new daily
mail-outbox-retention job deletes SENT rows (by sentAt) and permanently
FAILED rows (by nextAttemptAt, the last attempt's stamp) once they pass
mail.outboxRetentionDays (instance setting, default 30). PENDING rows —
including failed-but-retryable ones — stay the retry loop's alone.
Decision recorded (security.md §Privacy, residual-risk note for #231):
digest mails keep carrying page titles for now — there is no per-page
classification marking yet to key a suppression on (ADR 0022 / M32
revisits), and a VS-NfD reference configuration can leave SMTP
unconfigured entirely.
Job-count fence in system.spec: 8 -> 9.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
The raw input/result bytes of import/export conversion jobs were kept
forever; a deleted classified page could live on inside its last export.
A new daily conversion-payload-prune job nulls both once a finished
(succeeded or failed) job passes conversion.payloadRetentionDays
(instance setting, default 30) — the row survives for status/audit.
PENDING and RUNNING rows keep their payload, so the worker's stale-lock
recovery path is untouched; a hand-requeued pruned job fails finally
via conversionInputOf instead of crashing the worker.
The input column becomes nullable; the migration backfills by clearing
payloads of jobs already finished longer ago than the default period
(recent results stay downloadable until they age out).
Job-count fence in system.spec: 7 -> 8 (new scheduler registration).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Verification result: only deploy/compose/.env.example was ever tracked
(full-history check), zero hits for obvious secret patterns across all
added lines in history — recorded on issue #231 (residual-risk list).
The new CI step in the checks job fails if any .env other than
.env.example is tracked or a tracked file matches an obvious secret
pattern (private key blocks, AWS/GitHub/GitLab/Slack token shapes).
.env.example already documents every variable the compose files
reference (verified: comm of compose ${VAR} refs vs example keys is
empty). README states the example as the authoritative reference.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Hand-rolled middleware instead of helmet: the header set is small enough
to own, every value is a deliberate decision, and the api gains no
transitive dependency. HSTS (no includeSubDomains — the api cannot speak
for sibling subdomains), nosniff, Referrer-Policy no-referrer,
X-Frame-Options SAMEORIGIN (not DENY: the plugin sandbox frame embeds
same-origin and its CSP has no frame-ancestors, so this header governs),
and a minimal deny-all Permissions-Policy.
CORS grants no foreign origin anything; only the APP_BASE_URL origin is
ever echoed (where browsers do not consult CORS anyway), with
Vary: Origin on every response. No preflight handling — same-origin
requests never preflight, and cross-origin API access is cookie-less by
design (PAT/Bearer).
Wired via the AppModule MiddlewareConsumer so createTestApp boots the
identical middleware. Fences: security-headers.e2e.test.ts (header set,
foreign origin gets no ACAO) and a frame assertion in
plugins.e2e.db.test.ts (framing stays possible). Rationale table in
security.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
audit.retentionDays (instance setting, default 365) bounds the audit_log:
the daily audit-retention job deletes entries past the period and records
the deletion itself (audit.pruned with count, cutoff and period) so a gap
in the trail is always explainable. Lives in its own AuditRetentionService
because the settings service audits its writes - folding retention into
AuditService would close a constructor cycle. The read-access trail
(#222-#225) is deliberately not covered; it gets its own period.
security.md gains the Logging section the schema has cited for a while;
the maintenance-job fence moves 6 -> 7 (the deliberate new row).
Refs #196
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Trashing a page (promote and subtree modes) clears the affected search
vectors, restoring rebuilds them; pond trash clears every page vector of
the pond, pond restore reindexes only the live pages (pages trashed
inside stay out); the GDPR pseudonymization's personal-pond trash does
the same. reindexAll now converges to the invariant (clears trashed,
rebuilds live), and a one-off migration backfills vectors of
already-trashed content.
The query-side deleted_at guards stay untouched as the independent
second layer - the test proves both layers separately, including writing
a vector back onto a trashed page (simulating a future path that forgot
the clear) and asserting the query still hides it. New provider methods
removePond/reindexPond behind the SearchProvider seam.
Refs #195
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
The system panel spec pins the registered-job count on purpose; the
orphan-file-sweep registration (#194) is the deliberate sixth row (CI
run 493 caught exactly this, 14x resolved to 6).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Nightly sweep with two directions: attachments still unclaimed (pageId
null) after a 24 h grace period - claimed by no collab persist, page
upload, or import - are reclaimed (row, file, quota released); files on
the uploads volume without a database row (drift after a crashed
upload) are removed once older than the grace period. The grace period
protects the paste-then-insert window.
Deliberate deviation from the issue's content-reference idea, documented
in schema comment and operations.md: claimed attachments whose page
content no longer embeds them are NOT auto-deleted. The page attachments
panel lists claimed files as user-managed objects (inserting into the
document is optional there), so 'not embedded' is not 'unused' - an
auto-delete would destroy panel assets. Humans clean those up in the
panel or the pond file manager, which flags orphans already.
Attachment.deletedAt is removed by migration - deletion is hard
everywhere (sweep, purge, manual), there is no soft-delete state; the
never-true deletedAt:null filters in files/export queries went with it.
Refs #194
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Deletion now actually deletes: a trashed pond past the trash retention
(same clock as pages, extended trash-purge job) or purged manually via
DELETE /ponds/:id/purge (Site-Admin-only, like pond restore) is removed
with everything it holds. Files go first (idempotent rm, resumable on a
crash), then one transaction ordered around the FK actions: attachments
and labels (Restrict) precede the pond; the page delete cascades
versions, comments, content cache incl. the search vector, update log,
mentions, label assignments, favorites, outgoing links and open collab
sessions; the pond delete cascades grants, usage counters (that is the
quota correction), pond-plugin opt-ins and conversion jobs; polymorphic
watches and pond quota overrides are deleted explicitly. A purge racing
a restore or another purge is a no-op; both paths record a pond.purged
audit event.
Known residues by design, documented in operations.md: target_slug in
other ponds' page links (#235) and backups within their retention.
Refs #193
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
BACKUP_ALLOWED_TARGETS (comma-separated destination hosts) constrains
where backups may go, enforced twice: the api rejects settings writes
and connection tests towards non-allowlisted hosts with admin-visible
error codes and resolves a non-allowlisted configured target to null,
and the sidecar enforces the same policy at the point of egress for the
WebDAV upload and the rsync mirror alike (shared policy helpers in
packages/shared/src/backup-target-policy.ts).
BREAKING: the empty default disables every remote target - backups stay
local only, the VS-NfD reference configuration (ADR 0026). Existing
deployments with a remote target must list its host or uploads and
mirror stop. The admin UI distinguishes unavailable-by-policy from
unconfigured (i18n de+en) and shows the permitted hosts.
Refs #192 (ADR 0026)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Chosen path: an instance master switch following the api.enabled/
mcp.enabled pattern — while off, both feed routes AND the feed-token
management answer 404 (existence hidden). Default ON: feeds predate the
switch, existing instances and their subscribed readers keep working;
the VS-NfD reference configuration (#227) turns it off. Admin UI gets
the toggle next to the API/MCP switches (i18n de+en).
Moving the token out of the query string is documented as rejected: a
path segment lands in the same proxy and request logs, and feed readers
cannot send headers — that is why the credential is in the URL at all.
What DID leak was our own request log (pino logs req.url): the req
serializer now masks ?token= values (common/mask-token-param.ts), so no
code path logs the credential.
Refs #191
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
SESSION_ABSOLUTE_HOURS (default 168 h) caps a session's total lifetime
from login: expiresAt is set once at creation and never extended — the
old sliding 30-day renewal is gone. SESSION_IDLE_HOURS (default 72 h)
ends sessions unused for that long, enforced server-side against
lastSeenAt with a write throttle scaled to the idle bound so short idle
windows still renew. Expired rows are removed on validation and the
session list applies both bounds, so idle-dead sessions never show as
active. The cookie maxAge follows the configured absolute bound.
Documented in .env.example (with the VS-NfD reference values for the
upcoming hardening guide #227), compose passes the variables through,
security.md and ADR 0007 record the amendment.
Refs #190
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
The tampered-token case flipped the LAST base64url character of the
signature. Its low bits are padding that decoders ignore, so whenever a
signature ends in 'A' (~1/16 of tokens) the flip to 'B' decodes to the
same bytes and the token verifies — jose compares decoded bytes, unlike
the pre-#188 homegrown code that compared encoded strings. Reproduced
deterministically (20/20 A-ending signatures accepted the flip); CI run
477 and one local full-suite failure were this, not load. Flipping the
first character makes the tamper always significant.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
A cookie-carrying mutation without Origin and Referer (or with an
unparsable one) is now rejected with 403 csrf_origin_mismatch instead
of passing unchecked. The exception for non-browser clients stays
structural: PAT/bearer requests carry no session cookie and never reach
the check, and a request that does carry the cookie is always checked.
The test harness injects the matching Origin (supertest simulates a
browser page of this instance) with an explicit suppression header for
the negative cases; the Playwright fixture contexts send the header on
their manual seeding calls; release-qa.sh pins APP_BASE_URL and sends
the matching Origin. Dedicated spec covers: missing headers 403,
mismatch 403, unparsable 403, match passes, GETs untouched, PAT
mutation without headers passes, cookie+bearer still checked.
Refs #189
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
COLLAB_TOKEN_SECRET becomes a root key: every purpose derives its own
HKDF-SHA-256 subkey (deriveTokenKey), and no code path signs with the
root key directly. Collaboration tokens are signed and verified by jose
with HS256 as an explicit allowlist; the sign/verify API turns async at
its three call sites. Unsubscribe tokens move from a purpose-prefix
string to the structural subkey, with a documented dual-verify window
(legacy derivation accepted until 2026-11-01, covering the 90-day TTL
of links in already-sent mail).
The cross-runtime property that justified the homegrown implementation
is now proven by a test: the built CJS and ESM dist artefacts round-trip
tokens in both directions in child processes (jose v6 reaches CJS via
Node's require(esm), pinned Node 22 images). Negative tests cover
cross-purpose subkeys, root-key-signed tokens, alg:none and RS256.
Refs #188 (ADR 0020)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Reviewer-facing document derived from ADR 0019: per base function
(encryption, media protection, network termination, authentication,
integrity) what the application does, what it deliberately does not,
and which party provides it — every claim traceable to code via the
ist-aufnahme. Includes the operator-duty handover (with the IndexedDB
endpoint copy named explicitly, I-25) and the delta list mapping every
divergence from the target state to its closing issue.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
Add docs/vs-nfd/: the analysis brief, the as-is assessment (42 findings,
all verified against the code), the prioritized action plan rev. 2 with
issue references written back to every checkbox, the two-stage issue/ADR
brief, and the full reviewed draft used to create the forge state.
Add eight proposed ADRs 0019-0026 covering the VS-NfD architecture
decisions: no security base functions (par. 52 VSA anchor), HKDF token
key separation, external authentication, page classification, read-access
audit trail (variant A), reproducible offline deployment, plugin trust
model, and backup target restriction.
Forge state created alongside this commit: 11 labels, milestones M24-M31,
issues #188-#236 (docs-only change, no code touched).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0168Ph5uBmHm8X28CSVpbpnJ
pondSettingsSchema gains theme = { accent: '#rrggbb' | null } (null =
inherit the viewer's theme), exposed as a top-level key of the flat
updatePondInputSchema and included in the PondsService settings merge
(the known silent-no-op pitfall). The server validates only the hex;
conformance arises at render time: PondThemeScope (mounted around the
page content next to PondFontScope) derives the accent pair for the
EFFECTIVE mode via useEffectiveTheme and sets it as inline custom
properties — inline beats both tokens.css and the user-theme <style>,
which IS the cascade precedence pond > user > default.
Pond settings get a PondThemeSection (inherit | presets | custom color
with per-mode preview swatches, explicit save like the font manager);
AccentSwatches extracted for reuse; i18n de+en. The no-JS public shell
stays deliberately un-themed (ADR 0018 amendment).
Tests: pond DB test (theme merge keeps fonts, invalid hex 400), e2e
pond-theme.spec (scope boundary content vs. chrome, per-mode
re-derivation, axe on the pond settings page; resets the fixture pond).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
apply-theme.ts derives BOTH modes' accent tokens from the stored choice
(ui.theme.accent: preset id or {custom:'#hex'}) and writes them as
<style id="user-theme"> with :root:root + :root:root[data-theme='dark']
blocks — the doubled :root beats tokens.css regardless of document
order, since theme-init.js injects the ui.theme.css cache during <head>
parsing, before the bundle styles. The default preset means NO override
(hand-tuned tokens.css values stay). main.tsx re-derives from the
choice at startup, healing stale caches after app updates.
Settings: accent radiogroup inside the Appearance section (visible
names, color never the only cue) with per-mode preview swatches on
each mode's canonical background, plus a custom color input; i18n
de+en. The second fieldset made bare .settings-fieldset locators
ambiguous — theme specs now scope via input[name] (fence stays).
Tests: apply-theme unit pack, BASE_PALETTE<->tokens.css drift fence in
theme-contrast.test.ts, e2e theme-accent.spec (instant apply, pre-paint
persistence, default removes override, axe smoke with garish yellow in
both modes). ADR 0018 amendment documents the stage-B details.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
Dependency-free packages/shared/src/theme.ts: relativeLuminance /
contrastRatio (WCAG 2.1), deriveAccentTokens(hex, mode) keeps hue and
saturation and binary-searches lightness until the accent clears 4.5:1
against the mode's bg, bg-subtle AND surface (a passing hex is kept
verbatim; accent-contrast follows by symmetry). THEME_PRESETS (pond
green = default), BASE_PALETTE as the canonical backgrounds. A sweep
test (36 hues x 3 saturations x 3 lightnesses x both modes) fences the
by-construction guarantee for arbitrary input.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
An IconButton between the notifications bell and the user menu cycles
the theme mode in radio order (sun/moon/monitor mirror the CURRENT
choice). New useThemeMode() hook is the single write path (persist +
apply + same-document event), so the settings radios and the toggle
stay in sync; AppearanceSection now uses it too. Also rendered for
signed-out visitors — the mode is a device-local preference. i18n de+en;
unit tests for cycle/setter, theme.spec covers cycling, radio sync,
persistence, and the signed-out top bar.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
CI run 453 caught it: settings-nav.spec.ts pins the user-settings section
count, which #180's Appearance section raised from 8 to 9. Verified
locally against a fresh e2e environment (both pack tests green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
theme-contrast.test.ts parses tokens.css and asserts every real UI colour
pairing (4.5:1 text, 3:1 UI) for BOTH palettes, so palette drift fails
unit tests instead of review. theme.test.ts covers resolve/apply logic
(Node >= 22 ships a shadowing undefined localStorage global — the test
brings its own in-memory storage). The a11y pack now runs its four scans
in light AND dark via emulateMedia; the new theme pack exercises the
three-way switch end to end (instant apply, reload persistence, live OS
follow in system mode, override beats OS). ADR 0018 records the theming
model broadly: modes now, accent themes by derivation later.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
The dark palette lives as a single :root[data-theme='dark'] block in
tokens.css; theme.ts and the pre-paint public/theme-init.js (external file
because the prod CSP forbids inline scripts) always resolve the stored
ui.theme.mode to a concrete data-theme, so 'system' needs no @media
duplicate and follows live OS changes via matchMedia. color-scheme flips
per theme (native controls/scrollbars), paired theme-color metas track the
effective theme, and the new Appearance settings section offers the
three-way choice as native radios (device-local, like #170). Label chips
gain a chip-outline ring so arbitrary user colors stay separated on the
dark canvas; useEffectiveTheme() is exported for the later pond-scoped
theming stage (ADR 0018).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
New tokens (light values, dark arrives with the theme block): surface-muted
(was fallback-only), danger-contrast, danger-strong, badge ok/error/warn
pairs, chip-outline. Replace the phantom --color-primary with the real
--color-accent and unify the three danger reds (#a02818/#b91c1c/#ab091e)
on --color-danger. attachments-panel maps to bg-subtle, matching its old
6%-grey fallback. Light rendering is visually unchanged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRtCnB3uLdQtFmvp9HXcRX
Die Trennlinie unter der Aktionen-Spalte endete auf Höhe der Icon-Reihe
statt am Zeilenende: display:flex direkt auf dem td nahm der Zelle ihr
table-cell-Verhalten, sie wuchs nicht mehr auf Zeilenhöhe. Das Flex-
Layout liegt jetzt auf einem Innen-Wrapper (.user-row__actions-inner);
gemessen: 0 px Bottom-Delta über alle Zellen jeder Zeile.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Die Zeilen-Aktionen der Personenverwaltung sind jetzt IconButtons
(lucide): MailCheck (Bestätigung erneut senden, nur bei Ausstehend),
ShieldPlus/ShieldMinus (Zum Admin machen / Admin entfernen),
UserX/UserCheck (Deaktivieren/Aktivieren), Trash2 (Löschen) — in dieser
Reihenfolge. Das zweistufige Löschen bleibt: die Bestätigung ist
weiterhin ein roter Text-Button und erhält beim Umschalten den Fokus
(kein Fokusverlust, ADR 0017). Lokalisierte Namen kommen unverändert
aus users.json via IconButton (aria-label+title), Icons aria-hidden.
Der Admin-Bereich ist neu im a11y-CI-Pack (axe WCAG A/AA auf /admin).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Jede künftige UI-Änderung entwickelt Barrierefreiheit direkt mit
(Stefans Vorgabe nach Abschluss des WCAG-2.1-AA-Audits): verbindliche
Checkliste als ADR, Kurzfassung in CLAUDE.md für jede Dev-Session,
Eintrag im ADR-Index.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
@axe-core/playwright als devDependency (exakt +2 Lockfile-Pakete,
axe-core hat null Runtime-Dependencies; Freigabe durch Stefan im Chat).
Neuer e2e-Pack a11y.spec.ts scannt Login, Lesemodus, aktiven Editor und
Nutzer-Einstellungen gegen WCAG 2.1 A/AA — jede neue Verletzung bricht
den Build (Allowlist bewusst leer, nur mit Begründung erweiterbar);
Best-Practice-Regeln bleiben außen vor. In ci.yml als eigener Schritt
mit Rate-Limit-Reset nach dem Muster der übrigen Packs verdrahtet.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Toast-Standzeit 2,5s auf 6s (WCAG 2.2.1 — für Screenreader-/Zoom-Nutzer
kaum erfassbar). Neue Einstellungs-Sektion Bedienung mit dem Schalter
Einzeltasten-Kürzel deaktivieren (lokale Geräte-Einstellung); die
Handler von e und / prüfen sie beim Tastendruck (WCAG 2.1.4).
prefers-reduced-motion: CSS-Transitions kollabieren auf instant, die
Graph-Simulation rechnet ihr Layout synchron zu Ende statt zu animieren
(WCAG 2.2.2). settings-nav-Spec auf 8 Sektionen nachgeführt. Bewusst
KEIN zusätzliches role=status (legal.spec-Locator-Falle).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Task-Checkboxen tragen in beiden Renderpfaden einen Namen: docToHtml
setzt aria-label aus dem Aufgabentext, die Editor-NodeView ebenso. Die
NodeView rendert ihr Host-Element jetzt selbst als li (ReactNodeView-
Renderer as/attrs) — TipTaps zusätzliches div-Host-Element zwischen ul
und li brach die Listensemantik; der Wrapper flacht per display:contents
ab, die #137-Pixel-Abstimmung bleibt erhalten (Selektor auf die neue
Tiefe nachgeführt, Ausrichtung nachgemessen: 1px-Versatz unverändert).
Der Wissensgraph-SVG bekommt ein beschreibendes aria-label inklusive
Verweis auf die Backlinks als gleichwertige Listenform. Der
Bild-Alt-Editor existierte bereits (Bild-Controls bei Auswahl) — kein
Änderungsbedarf. Hinweis: gecachte Seiten übernehmen das
Checkbox-Label wie bei jeder docToHtml-Änderung erst mit dem nächsten
Persist ihrer Inhalte.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Der Field-Baustein verdrahtet Hinweis/Fehler jetzt per aria-describedby
und aria-invalid mit dem Eingabefeld (cloneElement auf das einzelne
Kind; Fragmente bleiben unangetastet) — Screenreader nennen den Fehler
damit auch beim Feld-Fokus. Quota-Typ-Select mit Namen; die leeren
Aktions-/Erledigt-Spaltenköpfe in API-Tokens, Feed-Tokens, Sitzungen
und der Aufgabenübersicht (NodeView UND Server-Renderpfad) tragen
visually-hidden-Beschriftungen.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Die öffentliche Server-Shell bekommt AA-geprüfte Dark-Mode-Farben
(color-scheme: light dark hatte den UA dunkel rendern lassen, Links
fielen durch 1.4.3; Text 14,8:1, Links 10,1:1, Muted 8,5:1). Wikilinks
tragen eine permanente Unterstreichung — Farbe allein war das einzige
Link-Merkmal bei nur 2,5:1 Abstand zum Fließtext (1.4.1). Neues Token
--color-border-input (#7d8a97, 3,5:1/3,3:1) für Eingabefeld-Ränder
(1.4.11); Wächter-Kommentar am Favoriten-Gold.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Skip-Link als erster Tab-Stopp springt auf #main; die angemeldete
Seitenansicht bekommt ein visually-hidden h1 (der sichtbare Titel ist
ein Input, der jetzt auch ein aria-label trägt); der Sidebar-Resizer
wandert in die nav-Landmarke (absolut an der Kante positioniert), damit
kein Inhalt außerhalb von Landmarken liegt. Zwei e2e-Locator auf das
Sidebar-Formular gescoped — das Editor-Titelfeld matcht seit dem neuen
Label ebenfalls auf /title|titel/i.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Vier Ursachen des seitenweiten Horizontal-Scrollens behoben: die Topbar
saß mit min-content-Breite in der Grid-Spalte (min-width: 0 nach dem
#100-Muster) und wickelt auf schmalen Viewports auf eine zweite Zeile
(Grid-Zeile minmax, Suchtext wird zum Icon); die aufgeklappte Sidebar
liegt unter 40rem als Overlay über dem Inhalt statt ihn auf einen
Streifen zu quetschen; Footer wickelt; Titel-Input und Settings-Spalte
schrumpfen (min-width bzw. align-items: stretch im Schmal-Layout).
Der Haupt-Scrollbereich ist per tabindex=0 tastatur-scrollbar — auf den
Rechtstext-Seiten gab es sonst keinen Weg, den Inhalt zu scrollen.
Gemessen: 10 Ansichten bei 320 px ohne Dokument-Überlauf (vorher 892 px
Inhaltsbreite); e2e content/settings-nav/sidebar/legal/page-tree/search
grün.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Die Editorfläche bekommt einen lokalisierten zugänglichen Namen und ist
im Lesemodus role=document statt eines unbenannten Textfelds (setOptions
im selben Layout-Effekt wie setEditable). Eingeklappte Sidebar zusätzlich
inert (aria-hidden allein ließ fokussierbare Kinder im Tab-Weg). Die
li-Zwischenknoten der Listboxen (Wikilink-/Mention-Autocomplete,
Suchergebnisse) sind role=presentation, damit listbox→option wieder eine
gültige Eltern-Kind-Beziehung ist. Toolbar: Pfeiltasten-Navigation über
die Controls (native Selects behalten ihre Pfeiltasten) und ein
sprechendes Toolbar-Label statt des Absatz-Buttons-Labels.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
i18n spiegelt die aktive Sprache auf <html lang> (Init + languageChanged;
der User-Locale-Wechsel in auth-context läuft über dasselbe Event). Neuer
useDocumentTitle-Hook setzt je Route einen sprechenden Titel
(Seite — Teich — Dorfteich), verdrahtet in allen Routen-Komponenten;
dynamische Titel folgen den geladenen Daten.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Gemeinsamer useModalFocus-Hook: Initialfokus in den Dialog, Tab/Shift-Tab
zyklisch gefangen, Fokus-Rückgabe an den Auslöser (bzw. returnFocusRef,
wenn der öffnende Menüpunkt mit dem Menü unmountet). Dialoge tragen jetzt
aria-labelledby auf ihre Überschrift und tabindex=-1 als Fokus-Fallback.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AGM8jo3hwoV9wsCVGfy8iq
Die NodeView las editor.isEditable nur beim Mount. Die Seite mountet
immer im Lesemodus, und der Moduswechsel läuft über setEditable() —
das emittiert in TipTap nur ein update-Event, aber keine Transaction,
weshalb React-NodeViews nie neu rendern (geprüft in @tiptap/react
3.27.1: updateProps feuert nur bei Node-Änderung und Selektions-
Wechsel). Folge: die Block-Leiste blieb ohne Bearbeiten-Knopf, für
alle Block-Plugins (ChordPro, Mermaid, Excalidraw, draw.io).
Fix: useEditorEditable abonniert das update-Event und liest
isEditable reaktiv; verliert die Seite die Editierbarkeit, während
die Editier-UI des Plugins offen ist, fällt der Block auf render
zurück (der Lesemodus blendet die Leiste aus, es gäbe sonst keinen
Weg mehr heraus). Damit stimmt auch die setData-Schreibrecht-Prüfung
(editableRef) wieder.
Regressionstest im plugin-blocks-Pack: Block existiert bereits,
Seite lädt im Lesemodus, Wechsel in den Edit-Modus zeigt den Knopf
(fiel ohne Fix reproduzierbar durch); Rückweg Lesemodus→render
mitgeprüft. Die bisherigen Tests fügten Blöcke immer erst nach dem
Moduswechsel ein und konnten den Fall nicht sehen.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Neues Referenz-Plugin packages/plugins/chordpro nach dem
mermaid-Muster: bewusst ohne Fremdbibliothek (Supply-Chain-Lehre aus
#136) — eigener minimaler ChordPro-Parser (Direktiven title/subtitle/
artist/key/capo/tempo/comment, Chorus-Fences, [Akkord]-Marker,
#-Kommentare) plus SVG-Formatter mit Monospace-Raster: Akkorde über dem
Text, Titelkopf, Chorus-Einrückung, XML-escaped. Persistenz {source,
svg} — der Snapshot bedient Lesemodus, Public-Ansicht und Exporte über
den PluginFallbackRenderer. Edit-Modus: Textarea + debounced
Live-Preview. ZIP 20 KB (Limits 64/256 MiB), 4 Parser-/Formatter-Tests,
i18n de+en, Doku-Listen (site-admin en+de, plugin-architecture)
ergänzt. Prod-Installation wie üblich per Site-Admin.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Der animierte scrollIntoView landete auf einer veralteten Zielposition,
wenn Query-Sektionen (Sessions/Tokens) während der Animation noch
wuchsen — auf dem CI-Runner deterministisch rot. Jetzt springt die
Navigation sofort; der Spec lässt die asynchronen Inhalte vor dem Klick
settlen (networkidle) und lief lokal 10× ohne Retry grün.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Das Block-Menü ist seit dem eingebauten Aufgabenübersicht-Eintrag
(#154) immer sichtbar und teilt die Styling-Klasse
editor-toolbar__section-select — der section-styles-Pack adressiert
das Abschnitts-Select jetzt per :not(.editor-toolbar__block-select).
settings-nav: toBeInViewport bekommt 10 s für Smooth-Scroll auf
langsamen Runnern.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neuer Notification-Typ mentioned; abgeleitete Tabelle page_mentions
(Migration), vom Collab-Persist transaktional neu geschrieben — der
Diff gegen den Vorzustand wird als pg_notify
(page_mentions_changed) emittiert, nur NEU Erwähnte lösen aus (kein
Spam bei Folge-Saves). Der api-Listener (erweitert um den zweiten
Kanal) ruft NotificationsService.fanoutMentions: Zustellung nur nach
canAccessPage-Recheck, die Autoren (pending contributors) benachrich-
tigen sich nie selbst; Payload wie gehabt mit Actor-Namen. API-seitig
erzeugte Seiten seeden page_mentions aus deriveContent. Glocken-Text
de+en; DB-Test (Leser ja / Outsider nein / Autor nein); kompletter
Loop live verifiziert (Tippen → Persist → NOTIFY → Notification).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neuer Block-Atom task_overview (Markdown-Fence dorfteich-tasks,
HTML-Placeholder). Shared extractTaskRows liest Task-Zeilen mit Text,
Mentions (#150) und Start-/Zieldaten (#152); TasksService sammelt zur
Lesezeit den Teilbaum (rekursiv via collectSubtreeIds, canAccessPage-
Filter je Quellseite) aus Basis-State + page_updates-Log — KEINE
abgeleitete Tabelle nötig (Teilbäume sind klein, kein Drift). Neuer
auth-Endpoint GET /read/:pond/:slug/tasks; die öffentliche Ansicht
expandiert den Placeholder serverseitig zur statischen Tabelle
(Instanz-Sprache). NodeView mit Live-Tabelle und Rückschreib-Checkboxen
(optimistisch, Override bis der debounced Collab-Persist nachzieht);
Einfügen über die Block-Auswahl (eingebauter Eintrag). Unit- + DB-Tests,
neuer CI-Pack tasks.spec (voller Loop inkl. Rückschreiben end-to-end),
User-Guide-Doku en+de.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
task_item bekommt ein optionales id-Attr (default null — Bestandsdocs
bleiben gültig), durchgereicht in toDOM/parseDOM und dem Lese-HTML;
der Editor vergibt/entdoppelt IDs lazy per appendTransaction
(TaskItemIds-Extension, auch gegen Copy/Paste). Neuer Kanal
TASK_TOGGLE_CHANNEL; POST /pages/:id/tasks/:taskId {checked} prüft
Schreibrecht, registriert den Toggler als pending contributor und
feuert pg_notify; neuer collab task-toggle-listener (Struktur =
restore-listener) öffnet eine DirectConnection und flippt das
checked-Attribut in einer Transaktion — offene Editoren konvergieren,
unbekannte taskId = geloggter No-op. DB-Test (NOTIFY-Payload,
Attribution, 403/404/400).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neuer Inline-Atom date_marker {kind: due|start, date: ISO}. Markdown
kanonisch ISO (>>2026-12-31), Eingabe-Kulanz dd.mm.yyyy; Block-Guard
vor blockquote hält zeilenführende >>Daten aus dem Zitat-Parser;
ungültige Kalenderdaten bleiben Text. Editor: InputRule beim Tippen
(+Leerzeichen), Anzeige per Intl.DateTimeFormat in Nutzersprache,
Überfällig-Färbung. Die User.locale-Verdrahtung existierte bereits
(auth-context, #17) — keine Änderung nötig. 6 Unit-Tests inkl.
Task-Listen-Zeile mit Marker und Mention.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neuer Inline-Atom mention {userId, username}: Markdown-Regel @username
(E-Mail-sicher über Wortgrenzen), Serializer, HTML-Span dt-mention,
Plain-Text für die Suche, Extraktor extractMentionUserIds. Neue
Endpoints GET /users/search (auth, min. 2 Zeichen, Limit 10,
Rate-Limit) und GET /users/brief (Batch-Auflösung für live
Anzeigenamen; gelöschte Nutzer → toter Chip). Editor: MentionView mit
Live-displayName, MentionAutocomplete (Klon des Wikilink-Musters),
Chip-CSS. 5 Unit-Tests.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
GET /public/:pond/feed.xml (zuletzt geänderte Seiten) und
GET /public/:pond/:page/feed.xml (Versions-Historie), @Public mit
404-Semantik; öffentliche Teiche anonym, nicht-öffentliche über neues
read-only Feed-Token je Nutzer als ?token=dt_feed_… (neue Tabelle
feed_tokens + Migration, Verwaltung in den Nutzer-Einstellungen,
FeedTokensSection). Öffentliche HTML-Seiten annoncieren den Teich-Feed
per link rel=alternate. DB-Tests (anonym/privat/Token-Lifecycle) und
User-Guide-Doku en+de.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
GET /api/public/v1/me existiert bereits — OpenAPI-Summary nennt jetzt
ausdrücklich die User-ID, api-guide (en+de) ebenso. MCP war bereits
paritätisch (list_ponds + Token-Identität); kein neuer Endpoint nötig.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neues pageListQuerySchema (ISO 8601, Kulanz für Datum ohne Zeit),
Query-Parameter auf interner und Public-API-Seitenliste, Prisma-where
mit gte; neue Indizes (pondId, createdAt)/(pondId, updatedAt) als
Migration. OpenAPI-Parameter, MCP-Parität (list_pages
created_since/updated_since), Doku (api-guide, mcp-guide,
public-api.md), DB-Test inkl. 400 bei ungültigem Datum.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neue SettingsLayout-Komponente leitet die Navigation per
MutationObserver aus den section>h2-Blöcken ab (erfasst konditionale
und komponenten-eigene Sektionen ohne Verdrahtung), sticky Leiste
neben dem Inhalt, auf schmalen Viewports horizontale Chip-Leiste;
aktive Sektion über Scroll-Position, am Seitenende gewinnt die letzte.
Auf allen vier Einstellungsseiten verdrahtet; die Admin-Grundeinstel-
lungen bekommen dafür eine eigene Überschrift. Neuer CI-Pack
settings-nav.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neues bare-Attr am transclusion-Node; $-Präfix in Markdown-Regel,
Serializer und Autocomplete; HTML-Placeholder trägt
data-transclusion-bare, Server-Expansion und NodeView lassen bei bare
Rahmen und Titel weg. Gleiche Tiefen-/Zyklen-/Permission-Regeln,
zählt weiter als Link. Unit- und DB-Tests ergänzt.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Aufgenommen auf Prod (v0.9.1, Handschrift-Fonts nach dem CSP-Fix aktiv)
auf der öffentlichen Seite konzerte-live — vervollständigt die neue
Excalidraw-Sektion analog zu den Mermaid-/drawio-Beispielbildern.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Vom Fallback-Renderer inline eingesetzte SVGs (z. B. Excalidraw-
Skizzen) tragen feste Pixelmaße und liefen auf schmalen Viewports über
den Rand — die bestehende img-Regel greift für inline-<svg> nicht.
Neue Regel .dt-plugin-fallback svg { max-width:100%; height:auto }.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Excalidraw bettet beim Speichern die verwendeten Schrift-Subsets als
data:-URIs ins Snapshot-SVG ein. Die Seiten-CSP (nginx) und die
Plugin-Frame-CSP erlaubten aber nur `font-src 'self'` bzw. den
Asset-Pfad — die Handschrift fiel in der öffentlichen Ansicht und im
Snapshot-Render auf Serifen zurück (auf Prod an der ersten Demo-Skizze
sichtbar). Fix: `data:` in beiden font-src-Direktiven. data:-Fonts
lösen keinerlei Netzwerk-Request aus — die Zero-Third-Party-Garantie
(security.md) bleibt unberührt; fonts.ts-Kommentar entsprechend
präzisiert.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Der Spec klickte nach page.reload() den Mode-Toggle „um den Edit-Modus zu
verlassen" — nach dem Reload ist die Seite aber schon im Lesemodus (React-
State resettet), der Klick wechselte also HINEIN. Das passierte jahrelang
folgenlos, weil der Fokus auf dem Toggle-Button blieb und „/" die Suche
öffnete. Seit dem Editor-Auto-Fokus (PR #140) landet der Fokus im Editor
und „/" wird dort zu Text — je nach Ausgang des Rennens gegen den rAF-
verzögerten Fokus mal grün (Läufe 381/385), mal rot (383/387/389/390).
Fix: den Klick streichen; „/" läuft im Lesemodus als globaler Shortcut.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Stefans Feedback nach dem Nachfahren-Selektor-Fix: „noch 3px weiter
runter". Die Zeilenmetrik des Editor-/Auth-NodeViews (label-Wrapper)
setzt die Checkbox ~3px tiefer als im öffentlichen docToHtml-Markup —
daher NUR für den label-Pfad `margin-top: calc(0.25em - 3px)`; die
öffentliche Ansicht (bare input, war korrekt) bleibt bei 0.25em.
Live per Injektion auf Test vermessen: Versatz −9 → −6px, Checkbox
mittig auf der Textzeile (Zoom-Screenshot).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Stefan sah die Checkbox-Verschiebung weiterhin — in der ANGEMELDETEN
Lese-/Bearbeiten-Ansicht. Dort rendert der TipTap-ReactNodeView das <p>
ZWEI Wrapper tief (`li > div[data-node-view-content] > div > p`), die
bisherige Kind-Kette `li > div > p` griff also nur im flachen
docToHtml-Markup der öffentlichen Ansicht. Fix: Nachfahren-Selektoren
(`li p:first-of-type` / `li p:last-of-type`) — robust gegen die
Wrapper-Tiefe beider Renderpfade.
Live am echten NodeView-DOM verifiziert (Injektion auf Test:
p-marginTop 16px→0, Checkbox bündig; öffentlicher Pfad unverändert ok).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Beim Umschalten in den Bearbeiten-Modus (Stift-Icon oder Shortcut „e")
landet der Cursor jetzt automatisch im Editor — bisher brauchte es einen
zusätzlichen Klick, der auf neuen/leeren Seiten zudem pixelgenau den
schmalen Inhaltsbereich treffen musste.
Effekt feuert, sobald der Editor editierbar wird (nach setEditable),
und überspringt den Fokus-Klau, wenn gerade ein Textfeld (z. B. der
Seitentitel) den Fokus hält — der Moduswechsel darf den Caret nicht aus
dem Titel reißen. TipTaps focus() stellt die letzte Auswahl wieder her
bzw. setzt den Caret an den Anfang einer leeren Seite.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Zwei im Bearbeiten-Modus gemeldete Layout-Bugs:
1. Die sticky Editor-Toolbar klebte an der Padding-Kante des Scroll-
Containers `.main` (padding-top: --space-6), also mit sichtbarer Lücke
unter der Navigation, durch die die scrollende Seite schien. Fix:
`top: calc(-1 * var(--space-6))` → die Toolbar pinnt bündig an die Nav.
2. Das „…"-Overflow-Menü der Navigation lag HINTER der Toolbar: `.topbar`
steht im DOM vor `.app-body`, hatte aber keinen Stacking-Kontext, also
malte die z-index-20-Toolbar in `.main` darüber und verdeckte
Menüeinträge. Fix: `.topbar { position: relative; z-index: 30 }` (> 20;
Modals mit 1100+ gewinnen weiterhin).
Beide am echten Test-Editor mit langer, scrollbarer Seite verifiziert
(gap_px=0 nach dem Scrollen; Menü vollständig über der Toolbar).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Im Lesemodus-Markup ist das <input> das erste Kind des <li>, also ist das
<p> nie :first-child — die Regel `li > p:first-child { margin-top: 0 }`
griff daher NICHT, das <p> behielt seine ~1em-Obermarge und der Text saß
deutlich tiefer als die Checkbox (auf Test/Int sichtbar, kein Cache-Bug).
Fix: :first-of-type/:last-of-type treffen den ersten/letzten <p>
unabhängig vom vorangehenden <input>. Beide Renderpfade abgedeckt.
Verifiziert per Harness mit exaktem <li><input><p>-DOM (alt vs. neu).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
pondGraph-Test sortierte die EMPFANGENEN Phantom-Slugs, verglich aber
gegen ein UNsortiertes Literal. Da beide Slugs (ghost-<sfx>,
ghost-secret-<sfx>) den Zufalls-Suffix teilen, kippt ihre Sortierreihen-
folge auf ~1/5 der Suffixe (wenn sfx[0] > 's') → nicht-deterministischer
Fehlschlag. In CI-Lauf 372 traf es zu (Suffix „vpxwe…"). Fix: beide
Seiten sortieren. Vorbestehender Bug, unabhängig von M17–M19.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
route-permissions.e2e.db.test.ts (#52) verlangt, dass JEDE Route ihre
Zugriffsregel explizit deklariert (PERMISSION_KEY, @Public oder
SiteAdminGuard). Der neue GET /read/:pond/:slug hatte keinen Decorator
(verließ sich auf den Default-Guard) → Coverage-Test rot in CI.
@AuthenticatedOnly() ergänzt (Session erforderlich; per-Page-Recht prüft
weiterhin der Service via resolve→canAccessPage→404).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neues Referenz-Block-Plugin „Excalidraw" (handgezeichnete Whiteboard-
Skizzen), analog zum draw.io-Plugin. Anders als draw.io (vendored Webapp)
ist Excalidraw eine React-npm-Lib: esbuild bündelt Controller + React +
Excalidraw in plugin.js, die Font-/Locale-/Data-Assets werden aus
node_modules in den ZIP-Root kopiert und zur Laufzeit über
EXCALIDRAW_ASSET_PATH (Plugin-Asset-Basis) geladen — nichts spricht mit
excalidraw.com, die Sandbox-CSP pinnt jede Anfrage auf self.
- manifest.json: kind=code, Block-Extension-Point diagram,
permissions blockData+ui, fallback "[Excalidraw]".
- src/plugin.tsx: Render-Modus zeigt gespeichertes SVG; Edit-Modus zeigt
Snapshot + Bearbeiten-Knopf (leerer Block öffnet direkt); Vollbild via
host.ui.enterFullscreen mountet <Excalidraw> (React), „Speichern &
Beenden" exportiert per exportToSvg, persistiert {scene, svg} über
host.blockData.setData → Fallback-Renderer bedient Lese-/Public-Ansicht
+ Exporte ohne Backend-Änderung.
- build.mjs: esbuild (jsx automatic, css→text, production-conditions) +
fflate-ZIP. Build erzeugt excalidraw-1.0.0.zip: 15,5 MiB zip /
22,3 MiB unpacked (Limits 64/256 MiB — passt).
- i18n de+en, globals.d.ts (CSS-Modul-Deklaration).
pnpm-Override @floating-ui/react-dom@2.1.2: Excalidraw 0.18.1 zieht sonst
@floating-ui/dom@^1.8.0, das (noch) nicht im Registry ist und `pnpm
install` repo-weit bricht (dokumentiert in pnpm-workspace.yaml).
VERIFIZIERT: typecheck/lint, Manifest-Validierung (SDK), Build+ZIP-Größe.
NICHT lokal verifiziert (braucht Preview/Test-Stage): Laufzeit —
Excalidraw-Rendering + Speichern unter Sandbox-CSP, Font-Laden vom
Asset-Pfad. Prod-Installation macht Stefan als Site-Admin.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Obsidian-Syntax `![[slug]]` (optional `![[slug|Anzeige]]`) als Seiten-
Einbettung. Im Lese- und öffentlichen Modus wird der Inhalt der Zielseite
inline gerendert; im Editier-Modus zeigt die NodeView eine Platzhalter-
Karte (Titel + Öffnen-Link).
Shared (Vorbild plugin_block):
- Neuer Block-Atom-Node `transclusion` (targetSlug + optional displayText).
- Markdown: Block-Regel für eine reine `![[…]]`-Zeile (vor `paragraph`
registriert; mitten im Absatz greift sie bewusst nicht), Token→Node-
Mapping, Serializer — Round-Trip stabil.
- html.ts: Platzhalter `<div class="dt-transclusion" data-transclusion>`.
- extractWikilinkSlugs erfasst jetzt auch Transklusionen → Einbettung
zählt als Backlink/Graph-Kante.
Backend (zentraler Render-Pfad):
- PublicService expandiert Platzhalter zur gerenderten Body-HTML der
Zielseite: SELBER Pond, read-permission-geprüft, Tiefe ≤2 + Zyklen-
Guard (visited); Fehlend/unlesbar/zyklisch → neutraler Wikilink. Medien
werden EINMAL über den ganzen Baum aufgelöst (kein Doppel-Processing).
- Neuer authentifizierter Endpoint GET /read/:pondSlug/:pageSlug (nicht
@Public) liefert dieselbe gerenderte HTML — für die NodeView im
authentifizierten Lesemodus, auch bei nicht-öffentlichen Seiten.
Web:
- NodeView `transclusion.tsx`: Editier-Modus → Karte; Lesemodus → holt
/read/:pond/:slug und rendert den (server-sanitisierten) Inhalt inline.
- WikilinkAutocomplete unterstützt `![[` → fügt einen Transklusions-Block
ein (statt Wikilink).
- CSS für Karte (.dt-transclusion-card) und Embed (.dt-embed), i18n de+en.
Tests: shared Round-Trip-Unit (5), public-DB-Test um Embed-Expansion
(zyklus-sicher, Fehlend→Link) erweitert — grün. typecheck/lint/i18n grün.
Visuelle Editor-Verifikation folgt auf dem Test-Stage.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
body.threads[0] ist unter noUncheckedIndexedAccess möglicherweise
undefined; per Destrukturierung + Non-null-Assertion nach dem
toHaveLength(1)-Check geglättet. (Der Fehler rutschte durch, weil der
#133-Commit nach dem Nachtragen des Tests nicht erneut getypecheckt
wurde.)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Kommentare erscheinen jetzt fest im Lesefluss zwischen Backlinks und
lokalem Graph statt in einem ein-/ausblendbaren Panel. Der
Kopfleisten-Toggle (Icon + Unread-Badge) entfällt.
Frontend:
- CommentsPanel → CommentsSection (Inline-Sektion, ohne Panel-Chrome/
Close-Knopf; markiert beim Sichtbarwerden als gelesen). Neue
Read-only-Variante PublicComments für die anonyme öffentliche Ansicht.
- Umzug auf die äußere Ebene in PageEditorPage (view-Modus, zwischen
BacklinksPanel und LocalGraphPanel). Das Schreibrecht (collab rw) wird
per onWriteAccess aus dem inneren PageEditor hochgereicht, damit die
äußere Ebene den Composer bei commentPolicy=editors korrekt zeigt/
verbirgt.
- Deep-Link ?comments=1 scrollt jetzt zur Inline-Sektion statt ein Panel
zu öffnen. Resolve/Unresolve-Knöpfe zusätzlich an mayComment gekoppelt
(früher nur an isRoot) — Leser sehen keine 403-Knöpfe mehr; Read-only
blendet alle Aktions-Controls aus.
- CSS comments-panel* → comments-section*; tote Unread-Badge-Regeln raus.
Backend:
- GET /public/:pondSlug/:pageSlug/comments (@Public), read-only. Nutzt den
vorhandenen resolve()-Pfad (erzwingt ggf. anonymen Lesezugriff → nicht
öffentliche Seiten 404en) und CommentsService.list. PublicModule
importiert CommentsModule.
Tests: public.e2e.db.test.ts um anonymen Kommentar-Lesezugriff + 404-Fälle
ergänzt (grün gegen frische Test-DB); comments.spec.ts auf die Inline-UI
umgestellt. typecheck/lint/i18n:check grün.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Neue schlanke Statuszeile (letzte Aktualisierung · Wortzahl · geschätzte
Lesezeit) zwischen Seitenkopf und Artikel — im authentifizierten
Lesemodus und in der öffentlichen Ansicht.
- Geteilte Komponente `PageStatusBar` (Datum via Intl in der aktiven
Sprache, Lesezeit = ceil(Wörter/200), Singular/Plural, Lesezeit
ausgeblendet bei 0 Wörtern).
- `countWords`/`htmlToText`-Helfer in lib/word-count.ts.
- Authentifiziert (`PageEditorPage`, nur Lesemodus): Wortzahl aus dem
vorhandenen Markdown-Export (geteilter Query-Key ['page-markdown']),
`updatedAt` direkt von `page.data`.
- Öffentlich (`PublicPageView`): Wortzahl aus dem server-gerenderten HTML
per DOMParser — kein Editor-Bundle nötig; kein Backend-Change.
- i18n common.statusbar (de+en), CSS `.page-statusbar` (middot-getrennt,
gedämpft). Gates grün (typecheck/lint/i18n:check); visuell verifiziert.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
`graphify claude install` erzeugte CLAUDE.md und .claude/settings.json,
die nicht Prettier-konform sind und `pnpm lint`/CI rotmachen würden.
Diese Dateien sind tool-generiert (bei Re-Install neu geschrieben), daher
per .prettierignore ausgenommen statt von Hand formatiert.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Die Task-List-CSS war auf `.editor-content` gescoped und griff daher in
keinem Lese-Container (public/comment/legal/home/history-preview), wo
docToHtml sein `<li><input><p>`-Markup einspeist — dort blieb der Bullet
sichtbar, die Checkbox lag inline und das Block-`<p>` mit Default-
`margin: 1em 0` versetzte den Text in die nächste Zeile.
Fix: Task-List-Regeln über das eindeutige `data-type='task_list'`-Attribut
(nur von docToHtml und der Editor-NodeView erzeugt) entscopen, sodass sie
im Editor UND in allen Lese-Containern greifen; Checkbox per kleinem
margin-top auf die erste Textzeile ausrichten und Ober-/Untermarge des
Item-Absatzes neutralisieren. Deckt beide DOM-Formen ab: Lesemodus
`li > input` + `li > p`, Editor `li > label > input` + `li > div > p`.
Visuell verifiziert (Vorher/Nachher, beide Pfade).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
`graphify claude install` im Monorepo: graphify-Abschnitt an (neue)
CLAUDE.md angehängt und PreToolUse-Hooks in .claude/settings.json
registriert (Graph-Check vor Such-/Lesetools, Auto-Rebuild nach
Code-Änderungen). Der generierte graphify-out/ (AST-Graph über 741
Code-Dateien, ~10 MB) wird nicht versioniert.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Captured on prod with a second collaborator account (Zoraya Shahin):
chapter 12 shows Nadia's bell after a watched-page comment, chapter 13
shows the presence strip + Zoraya's named caret in Nadia's editor.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
Before/after pair for chapter 4 via temporary version rollback; the two
remaining placeholders (notification inbox, live cursors) need a second
account and are deferred.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
The manifests show only mermaid/drawio are insertable blocks; toc and
page-index are page tools in the top bar, section-styles extends the
toolbar. Chapters now describe each plugin where it actually appears.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
21 step-by-step chapters for non-technical users, using the public demo
pond nadia-morgenstern as the running example. German is the original
for the tutorial (unlike the manuals); screenshots are pending and
marked with visible placeholders. Manual READMEs (de+en) link to it.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155v2aT8AG1kZDQEZiCLBWC
The always-mounted toast container carried role="status", so every
page-scoped getByRole('status') locator suddenly resolved to two
elements — legal.spec failed CI with a strict-mode violation. The
region keeps aria-live="polite" (announcements work the same); the
status role stays with the per-page elements that had it before.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
Semantics changed from the issue during planning (documented there,
comment 1192): favorites are PERSONAL per user, not pond-wide — the
sys-fav label approach is dropped entirely. Storage is a page_favorites
table (userId+pageId, FK cascade); PUT/DELETE /pages/:id/favorite
toggles idempotently and needs read access only (#60 404 semantics —
a star is a note-to-self, not a page modification), GET
/ponds/:id/favorites lists the account's stars sliced to still-readable
pages. Trashed pages keep their rows, so restore keeps the star; purge
cascades it away.
Web: one shared ['favorites', pondId] query feeds the TopBar star
(between labels and history, golden when set), the golden tree icons in
the sidebar, and a latching "Favorites" filter button next to the view
switch that narrows either view (combinable with the label filter).
No public-API/MCP exposure — with the label approach gone, that parity
is no longer free; favorites stay UI-only for now.
New favorites e2e pack (star toggle, golden icon, filter, per-user
isolation) wired into CI; DB suite covers the round-trip, read gating,
and the trash/restore/purge lifecycle.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
Each trash row gets a checkbox, a toolbar above the list offers
"select all" (native indeterminate for partial selections) and the two
bulk actions; bulk purge confirms with the selection count (pluralized).
Processing is sequential on purpose — purge promotes leftover children
(#107), so concurrent tree mutations would race. Failures don't strand
the rest: the loop keeps going, failed pages stay selected for a retry,
and an alert banner reports the count. Single-row actions run through
the same path, which also fixes their previously unhandled rejection.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
Cmd/Ctrl+S used to snapshot silently. A new app-wide ToastProvider
(components/Toast.tsx) owns a bottom-center stack — permanent polite
live region, auto-dismiss after 2.5 s, click to dismiss early, error
variant. Both snapshot paths (the keyboard chords in PageEditorPage and
the save-version TopBar button) now confirm with the version name when
there is one, and their failure alert becomes an error toast.
The plugin host capability ui.toast (declared since #74, wired
nowhere) connects to the same stack: PluginBlockScope carries the
showToast handle, plugin-block passes it into the sandbox context.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
The SVG scaled to container width with height following the fixed
800×560 viewBox ratio — on wide windows the graph grew taller than the
viewport, the page got a scrollbar, and wheel-zoom scrolled along.
The graph page is now a flex column filling the main column; the canvas
takes the remaining height (flex: 1, min-height: 0), a ResizeObserver
feeds its measured size to ForceGraph as width/height, and the SVG
fills it exactly. LocalGraphPanel keeps its fixed defaults. Scope
deliberately layout-only (issue comment 1187): with no scrollbar there
is nothing for the wheel to scroll, so no non-passive listener needed.
The graph pack now asserts the main column does not overflow
vertically on the graph route.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
.sidebar is itself the scroll container, so the footer's margin-top:
auto only pinned the icon row to the end of the CONTENT — behind the
fold on long page lists. The row is now position: sticky with negative
bottom/side margins undoing the sidebar padding, a background, and a
top border, so the list scrolls away beneath it and the four actions
stay visible at any list length. The import progress list keeps
floating above the row (absolute within the sticky footer).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
Compressing and inflating the 257-MiB zero buffer is CPU-bound and
exceeded vitest's 5 s default on a busy CI runner (5.4 s — the M15 push
ran CI and CD concurrently). Flaky tests are defects (ADR 0014): the
test gets an explicit 30 s timeout.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
Real vault ZIPs broke umlauts in page titles ("Fußball zum Götzen" →
mojibake, slug fua-ball…goi-tzen): fflate honors only the ZIP UTF-8
flag, which common archivers omit, and decodes unflagged names as
Latin-1. That decoding is byte-lossless, so parseVaultZip now re-reads
any name whose chars all fit one byte as UTF-8 (a strict decoder —
genuine Latin-1 and flag-decoded precomposed chars fall back
unchanged), then NFC-normalizes: macOS zips store umlauts decomposed,
which silently broke slugify's ä→ae digraphs, wikilink matching, and
duplicate-basename detection. slugify itself also precomposes first as
defense in depth for NFD input from other paths.
Unit tests pin both cases: a hand-patched ZIP whose UTF-8 name bytes
carry no UTF-8 flag, and an NFD-named note that must come out
precomposed with an ueber- slug.
Pages already imported with garbled titles stay as they are — delete
the imported subtree and re-import after this lands (or rename by
hand).
Fixes#127
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
The Dockerfile has passed the release tag into the web build as
VITE_APP_VERSION since the beginning, but nothing consumed it — the
footer now renders it right of the live/offline icon; dev builds show
the 0.0.0-dev placeholder.
Fixes#126
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
In reading mode a plain "e" (guarded against typing targets) switches
to edit mode. In edit mode the platform's native chord — Cmd on macOS,
Ctrl elsewhere — +S saves an unnamed manual snapshot in place, and
+Shift+S asks for a name and returns to reading mode; both always
swallow the browser's save dialog. The shared isTypingTarget guard
moves from TopBar into lib/keyboard.ts next to the new modifier helper.
Unnamed snapshots needed the API to accept them: the version label is
optional now (trigger stays MANUAL, label null), and the history list's
existing null-label fallback text becomes "Manueller Schnappschuss" /
"Manual snapshot" — it only ever shows for exactly those. DB test for
the label-less path, e2e coverage in the CI content pack.
Fixes#125
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
Graph, new page, import, and trash collapse from scattered text links
into one icon row pinned to the sidebar's bottom edge, in that order,
each with a hover hint (the trash reads "Papierkorb anzeigen"). The
new-page button now toggles the inline form, which still renders above
the footer with the same classes; the import trigger becomes an icon
whose progress list floats above the row so the icons stay put. All
e2e class hooks (.sidebar__new-page, .sidebar__graph-link,
.sidebar__import-*) are unchanged.
Fixes#124
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
The force layout used to run once (tick(250)) and freeze; dragging
moved a single node with no reaction from its neighbors. The simulation
now stays alive: React renders the SVG structure (testids, edge/ring
classes — the e2e contract is unchanged) while each tick writes
positions imperatively into the element refs, and it settles to rest
via alpha decay, which also keeps Playwright's stability wait happy.
Dragging pins the node (fx/fy) and reheats the physics, so the
neighborhood gets pulled along; a plain click still just opens the
page. Surviving nodes keep their positions across data refreshes, e.g.
when a phantom becomes a real page.
The graph page gains four sliders — attraction, repulsion, node size,
font size — persisted per pond (ui.graph.settings.<pondId>) with a
reset; the inner view is keyed by pond id because usePersistentState
reads its key only on mount (#108 trap). The local panel adopts those
settings (no second set of sliders) and swaps the two fixed hop
buttons for a 1–5 depth slider.
Fixes#123
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
The tree reused .sidebar__page-item's flex ROW from the flat list, so
the nested children <ul> sat BESIDE the parent row and align-items:
center made the parent float vertically centered next to its subtree.
Stack the two in tree view instead: parent first, children indented
below (about two characters per level).
Rows get compacter (smaller font, tighter padding) so a deep imported
vault fits on screen, and folder pages now read differently from leaf
pages at a glance: lucide Folder/FolderOpen vs FileText, with the caret
glyph upgraded to a ChevronRight that keeps the existing CSS rotation.
Caret button, classes, and aria-labels stay untouched for the e2e
contract (#101 convention: no page title in the caret label).
Fixes#122
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
The label multiselect was gated on the pond already having labels — but
before a first import that is the common case, so the section silently
vanished and no import-wide label could be chosen. Render the fieldset
unconditionally (with a hint when empty) and add an inline create
field: POST the new label directly to get its id back, refresh the
shared label query, and tick it right away. Same pond_admin permission
as the dialog itself.
The e2e pack now creates its label through the dialog instead of the
API, covering exactly the empty-pond path that slipped through.
Fixes#121
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
.modal (and three other rules) referenced --color-surface without a
fallback, but the token was never defined — the background declaration
was silently dropped and every modal panel rendered see-through over
the page. Define the token in tokens.css (white, matching the #fff
fallbacks other rules already carried) and pin the vault-import
dialog's opaque background in the e2e pack.
Fixes#120
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fb2VzvcoBPHkjh8bZ6PzQn
CI runs the two new packs after the graph pack (chained, each preceded
by the login rate-limit reset): create-missing-page.spec.ts (#115) and
import-vault.spec.ts (#117/#118).
Docs: the pond-admin guide gains a full 'Import an Obsidian vault'
chapter — the three dialog choices, and what happens to folders, links
(including the duplicate-name rule: the alphabetically first vault path
wins), tags, images, and embeds, plus the limits and the all-or-nothing
semantics. The user guide explains following a link to a page that does
not exist yet. features.md gets both bullets. German mirrors updated
throughout (English stays authoritative).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
An admin-only 'Import an Obsidian vault' section on the pond settings
page opens a dialog with everything the #117 endpoint expects: the ZIP,
an indented mount-parent picker over the page tree (the MovePageDialog
pattern), a multi-select over the pond's label tree, and the
frontmatter radio (strip / keep as code block). Submit uploads and
polls the job with a vault-sized budget (600 x 1 s), then invalidates
pages, graph, phantom-links, and labels so the sidebar tree, graph, and
pickers show the import without a reload — and links to the mount page.
apiUploadFile now takes extra multipart fields (the options JSON);
existing callers are unchanged.
e2e import-vault.spec.ts: an admin imports the fixture vault through
the dialog and the app shows the folder tree under the mount page, a
rewritten Obsidian link navigates to the right page, the embedded image
renders, and the nested tag labels exist next to the dialog's extra
label; a plain editor gets no section at all. 3x flake-free locally
(CI wiring lands with #119).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
POST /ponds/:pondId/import/vault (pond-admin-gated; a vault import
creates a subtree, uploads files, and creates labels — administration,
not everyday editing) takes the ZIP plus a JSON options field
{parentPageId?, labelIds?, frontmatterMode}. The archive is parsed at
enqueue for fast 400s; the job (new kind import_vault, riding the
existing isImportKind worker routing) re-parses and runs the #116
transform, then: containers top-down → notes (asset placeholders →
uploaded pond files; non-images become page attachments) → tags to
labels (nested tags build a label hierarchy via LabelsService, so
locking and cache invalidation apply) plus the dialog labels.
All-or-nothing: any failure hard-deletes the created pages (children
first) and removes the stored files (quota restored), then surfaces as
import_vault_invalid_zip / import_vault_too_large / quota_exceeded /
conversion_failed — and makes the worker's retry policy safe.
Supporting changes:
- conversion_jobs gains a nullable options jsonb column; enqueue takes
kind-specific options and a maxInputBytes override (the 25 MiB
default protects the pandoc sidecar, which a vault never touches —
vaults use the 64 MiB upload limit).
- insertPage accepts a pre-reserved slug (the batch reserves all slugs
up front against pond ∪ batch).
- NEW: pages born with content seed their outgoing page_links rows
(deriveContent now returns wikilinkSlugs) — imported pages would
otherwise stay invisible to backlinks and the graph until their
first collab save. Collab still rewrites the rows on every save, and
the existing phantom resolution heals batch creation order.
import-vault.e2e.db.test.ts (4 tests, real worker drained): gating +
input rejection, the full fixture import (tree under a mount page,
collision suffixes, link rows incl. phantom, nested tag labels, extra
label everywhere, frontmatter stripped, image embedded + PDF attached),
complete quota rollback, and a clean re-import with fresh suffixes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The asset type lost its name field during the split into path+extension
(tsc error), and the fixture's .obsidian/app.json needed Prettier's
formatting. Lesson from the /srv move repeated within one day: the gate
chain must gate — never '; echo' past a failing typecheck.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pure functions from vault ZIP to import plan — no DB, no DI:
- parseVaultZip: fflate unzip with the plugin-package protections
(zip-slip rejection, incremental unpacked ceiling 256 MiB,
parameterized for tests); dot-directories like .obsidian/ skipped;
deterministic ordering.
- extractFrontmatter: leading --- block, tags:/tag: in scalar, inline-
array, and block-list forms; strip mode drops the block, preserve
re-emits it as a fenced yaml code block.
- extractInlineTags: fence- and inline-code-aware #tag / #nested/tag
extraction and removal (headings and pure numbers untouched).
- rewriteLinks: [[Name]], [[Name|Display]], [[Name#Heading]] (fragment
stripped), [[folder/Name]] (path match beats basename) → the FINAL
slug with the human name as display; unresolvable → slugified
phantom; ![[img]] and relative  → vault-asset: placeholders
the uploader resolves (#117); non-image embeds → italic filename +
page attachment; SVG deliberately stays an attachment (never inline,
security.md); note embeds degrade to plain wikilinks.
- planFolders: folder chains merged at the deepest levels to fit
MAX_PAGE_DEPTH below the mount page (merged titles read c/d).
- planSlugs: -n suffixing against existing ∪ batch; duplicate
basenames resolve to the lexicographically first vault path.
- planVaultImport ties it together into containers + notes + the
referenced-asset set.
Fixture vault under fixtures/import/obsidian-vault/ (umlauts,
duplicate basenames, nested tags, deep folders, embeds, code traps);
13 unit tests colocated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Following a phantom wikilink now ends with a way out instead of a dead
end: when the pond resolved and the page 404s as plain not_found, the
error screen offers creating the page in place. Title = the URL slug
(the PhantomPagesView mechanic), so every wikilink pointing at the
address resolves; the invalidated page query then mounts the editor on
the same URL. The affordance is deliberately ungated like the sidebar's
new-page button — the client cannot tell 'never existed' from 'not
readable' (#60), and a reader's POST surfaces as the regular 403
banner. The page_trashed branch (#31) is untouched.
Rides along: PhantomPagesView now also invalidates ['pond-links'] —
the graph views kept showing a just-created target as a phantom.
e2e pack create-missing-page.spec.ts (CI wiring lands with #119):
author a phantom link, follow it, create, backlink proves resolution;
reader path asserts the 403 banner and no editor mount.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Create-from-link on the not-found screen and the Obsidian vault import
(#115-#119), planned and filed today.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
prod-deploy.yml reused the test-stage deploy key since go-live (the
checklist's optional-hygiene item). A third keypair now completes the
one-key-per-stage picture: DEPLOY_SSH_KEY_PROD secret, pubkey
dorfteich-deploy-prod in the deploy user's authorized_keys. This
separates rotation and audit per stage — not privileges: every key
lands in the same docker-group deploy user on the single host.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Prettier keeps the markdown table columns padded; the /srv/DOCKER
replace left them one character short. (Lesson re-learned: the lint
gate runs before every push, even for doc-only commits.)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ONE consolidated every Docker stack under /srv/DOCKER (BASEL
convention, tracked in stwaidele/infrastructure-one); the three
Dorfteich stages follow. Deploy targets in cd.yml (test/int) and
prod-deploy.yml plus the docs now point at /srv/DOCKER/dorfteich-<stage>.
Data lives in named volumes keyed by the unchanged compose project
name, so the directory move carries no data migration. The go-live
checklist's uptime-kuma path already lives under /srv/DOCKER — the doc
just catches up.
Deliberately committed together with the host-side move: this commit
must not deploy before the directories exist at the new path.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two new CI-wired Playwright packs, each in its own shared pond so the
fixture ponds stay untouched:
- page-tree.spec.ts — create-as-child with the form hint, collapsible
folder view (collapse state survives reload), label view grouping,
the local view override vs the owner-set pond default (fresh context
without localStorage sees the new default), the Move-to dialog with
the own subtree disabled, promote vs subtree delete, and a restored
orphan re-attaching at the root.
- graph.spec.ts — pond graph nodes/edges/legend, node click-through,
the phantom-create flow (dashed node turns solid), the local panel
with hop toggle and highlight ring, and the permission slice: a
label-denied reader sees neither the hidden node nor its edge.
Both packs 3× flake-free locally. Manuals: user guide (page tree,
moving/deleting with subpages, knowledge graph + local graph), pond
admin guide (sidebar view default), features.md (knowledge graph
bullet) — with the docs/de mirrors updated (English authoritative).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A collapsible 'Local graph' panel joins the backlinks below the page
content in read mode: the current page (highlight ring) with its
wikilink neighbors in both directions, switchable between direct
neighbors and two hops. Computed client-side by BFS over the cached
pond-wide graph response — no second endpoint; the TanStack query is
shared with the pond graph view. Phantom targets render dashed; a
click navigates to the neighbor; pages without any links show no
panel at all. Reuses the ForceGraph renderer from #112 unchanged.
Verified live: hop toggle reveals the second-hop page, ring on the
current page, click-through, and the panel's absence on a lonely page.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
/p/:pondSlug/graph (static segment ranked above :pageSlug, same
documented reserved-slug gap as trash/settings) renders the pond's
readable wikilink graph from GET /ponds/:id/links: pages as nodes
colored by their first label (legend included, DEFAULT_LABEL_COLOR for
unlabeled), resolved links as edges, phantom targets as dashed nodes —
clicking one offers to create the page, which resolves its links.
Rendering is a self-contained SVG force graph: only d3-force is
bundled (no d3 DOM/zoom modules, zero external requests); the layout
runs synchronously to rest, zoom/pan/node-drag are plain pointer math.
SVG over canvas deliberately — every node carries a data-testid the
e2e packs can click. Ponds beyond 500 pages get a capped-view notice.
Sidebar footer links every member to the graph (trash stays
owner-only). New i18n namespace graph (de+en).
Verified live: nodes/edges/legend render, node click opens the page,
phantom click creates it and the node turns solid.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The slug-based machine surfaces now see and shape the hierarchy:
- REST: page list/detail carry parent (the parent page's slug, nulled
when the token's user may not read it — same no-leak rule as the
internal list); create accepts parent; PATCH accepts parent
(slug nests, null moves to the top level, appended at the end of the
new sibling group via the new PagesService.moveToEnd). Cycle/depth
refusals keep their regular error codes. OpenAPI updated.
- MCP: list_pages returns parent, create_page takes an optional parent
slug, update_page moves with parent (slug|null); tool errors carry
the api code (page_cycle covered in the e2e pack).
- ZIP export deliberately stays flat — noted in features.md; the
hierarchy is organizational only.
e2e: REST pack covers nested create, list shape, move/root-move, 409
page_cycle, 404 unknown parent; MCP pack covers nested create, list
parent, and the cycle tool error.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sidebar folder view: a row now has three drop bands — the edges keep
the within-group reorder, the middle band nests the dragged page under
the row (appended to its new sibling group, with a drop-into outline
cue). Cycle/depth refusals surface as a translated banner; successful
moves are announced for screen readers.
The overflow menu gains 'Move to…': a modal parent picker over the
page tree (top level first, the page's own subtree disabled) that works
in every sort mode. Delete now decides per case: childless pages keep
the plain confirm; pages with subpages open a dialog offering promote
(default wording: move subpages up) or subtree delete.
The children lookup reads the CACHED pages list on purpose: an async
fetch before window.confirm broke the click→confirm→DELETE rhythm the
content pack (and users) rely on, and a stale childless read errs
toward promote — never toward a silent subtree delete. Sidebar caret
labels deliberately exclude the page title: accessible names are
matched by substring in the specs (#101), and a title like 'Editor…'
collided with the edit-mode toggle.
Verified live: move dialog (subtree option disabled), promote and
subtree delete flows; content/trash/export packs green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GET /ponds/:pondId/links returns the caller's readable slice of the
wikilink graph in one read: nodes (id, title, slug, labelIds for the
coloring), resolved edges deduplicated per direction (a rename can
leave several slugs pointing at one target), and phantom targets with
their referrer ids. An edge survives only when both endpoints are
readable; a phantom disappears entirely once its last readable referrer
is filtered — a hidden page's existence never leaks through any of the
three collections. Trashed pages and their links are excluded.
Shared PondGraphView types feed the knowledge-graph views (#112/#113).
DB tests cover the owner's full graph, the label-DENY reader slice,
edge dedup, trash exclusion, and labelIds on nodes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The sidebar now presents pages as a collapsible tree built from parentId
(folder view) or grouped under the hierarchical label tree (label view,
read-only; multi-label pages appear under each label, untagged ones in
an 'unlabeled' group). The pond owner sets the default via a new
sidebarView pond setting (PATCH-merged like the other keys); every user
can override it locally (ui.sidebar.view.<pondId>), and the toggle sits
above the page list. Collapse state persists per pond.
New pages created while a page is open become its children — the inline
form says so and sends parentId. Reordering (buttons and drag-between)
now operates within one sibling group; the label filter stays a
folder-view feature and falls back to the flat list while active, so
the filtered order is never mistaken for a partial tree.
SidebarContent is keyed by pond id so the per-pond localStorage hooks
mount with the right key. e2e hooks (.sidebar__pages, .sidebar__page,
reorder buttons) kept; reorder/labels/content packs green locally, plus
a live smoke of nesting, collapse persistence, and both views.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
DELETE /pages/:id?mode=promote|subtree — promote (the default) moves
the page's live children up to its parent; subtree trashes every live
descendant with one timestamp and requires write permission on all of
them (no partial deletes; trash access is write capability, ADR 0013).
Trashed pages keep their parentId. Restore re-attaches to the nearest
live ancestor (else root), which makes restore order-independent:
restoring a parent afterwards never re-claims an already-restored
child. Purge promotes any remaining children to the purged page's
parent; the FK's SetNull stays as backstop only.
tree-trash.e2e.db.test.ts covers promote, subtree + one-timestamp,
the 403 descendant gate (label-DENY editor), order-independent
restore, and child promotion on purge.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pages form a tree via a nullable parent_id self-relation (SetNull
backstop; the real trash/purge semantics follow with #107). Slugs and
URLs stay flat and pond-unique, so moving a page never breaks links.
- Shared: generic parent-id tree helpers in tree.ts (labels re-export
them; buildLabelTree keeps its name-sorted behavior), MAX_PAGE_DEPTH=6,
parentId on PageView, createPageInputSchema.parentId (nullish),
repositionPageInputSchema.parentId (optional; absent = keep parent).
- API: create validates the parent (same pond, live, depth);
PATCH /pages/:id/position reparents atomically with the placement,
rejecting cycles (page_cycle) and depth violations
(page_depth_exceeded); GET /ponds/:id/pages nulls parentId when the
caller may not read the parent, so hidden page ids never leak.
- New error codes translated de+en; hierarchy.db.test.ts covers create,
404s, depth, cycle, atomic reparent, and the permission nulling.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
M10 (UI polish) and M11 (public API & MCP) existed as Gitea milestones
but were missing here; M12 (page hierarchy & knowledge graph, #106–#114)
is newly planned. The backlog heading no longer claims to sit after M9.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The landing form reused LegalTextField, whose .legal-editor wrapper the
legal e2e selects by index (nth(1) = privacy policy). Placed before the
legal form it shifted those indices, so the test drove the imprint field
and the published privacy text never appeared. Generalize the component
to MarkdownTextField with a wrapperClass prop: legal keeps .legal-editor,
the landing editor uses .markdown-field. Verified locally: legal,
admin-users, admin-quotas packs green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
The landing form reused the legal namespace's "Save legal pages" label,
so two buttons shared that text and the legal e2e's page-wide button
lookup hit a strict-mode violation. Use a dedicated settings-namespace
"Save landing page" label instead.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
The public home page (/) now renders Markdown the Site Admin stores in
the new home.content instance setting, through the same sanitizing
pipeline as the legal pages; empty falls back to the built-in welcome
text. New public GET /home/content, an Admin → Settings editor with
live preview, and an e2e test covering default/configured/escaping/
admin-only.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
The content-cache renderer (docToHtml) emitted wikilinks without an
href, so on the read-only public page view they rendered as styled but
unclickable text. Emit href="<slug>" — relative to the current page URL,
it resolves to the sibling page under both /public/<pond>/… and the
in-app /p/<pond>/… version-history preview, without the renderer needing
pond context. Slugs are [a-z0-9-], safe as a bare path segment.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
Prettier normalizes *after* to _after_; the check was masked in the
previous commit because eslint failed first and short-circuited the
&& before prettier ran.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
no-console is not enforced for apps/api, so the directive on the
boot-time restore-wait log line was flagged as unused and failed the
lint gate (--report-unused-disable-directives). Keep the console.log
and its rationale as a plain comment.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
Uptime-Kuma stood up on ONE with the four Prod monitors green and
Matrix alerting into the existing Trinity room (test alert confirmed);
the restore drill, now pointed at dorfteich-prod_backups, restored the
fresh set 20260712-183629 green (2 users, 16 pages). Only the optional
DEPLOY_SSH_KEY_PROD hygiene item remains open.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
The runner on ONE holds dorfteich-prod_backups; since go-live the drill
that proves a backup restores should target the live instance, not Test.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
Mirrors the English tree (docs/de/{features.md,manual/*,developer/
extending.md}) so relative links between translated guides resolve
within the German set; links into untranslated areas (self-hosting,
architecture, deploy) point at the English files and say so. Every
quoted UI label matches the actual German interface strings. Each
pair of files cross-links the other language; English stays
authoritative when the two diverge.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
The pond switcher grows a "+ New pond" entry with an inline form
(name + optional description, quota errors surfaced translated); the
pond settings of shared ponds end in a danger section that moves the
pond to the site-level trash after typing its name to confirm.
Personal ponds keep hiding the section. .button--danger is now a
solid red button (also fixes the admin restore button, which showed
red text on the accent-green background). Manuals no longer call
these actions API-only; covered by a members-pack e2e test.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
- deploy/legal/dorfteich-online-{impressum,datenschutz}.md: ready-to-
paste Markdown for the flagship instance (Admin → Legal pages after
the wizard) — operator standard texts (§5 DDG, §18(2) MStV, VSBG
no-participation, UGC/liability/copyright notices) plus a privacy
policy grounded in what THIS instance actually does: Hetzner hosting
with DPA, session cookie only (no banner), rate-limit IPs, proxy logs
≤30d, transactional + digest mail with unsubscribe, plaintext content
with versions/comments and public pages, hashed API tokens with
audited writes, nightly backups incl. the encrypted-tunnel mirror to
the operator's private server, self-service export,
deletion/pseudonymization, LfDI BaWü.
- docs/self-hosting/legal-template.md: review checklist and both
language templates extended for the processing added since #82 —
comments/version history, notifications/digest mails, API tokens,
off-host backup copies.
No legal advice; texts follow the operator's standard building blocks
and should get a final human read before publishing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
Seven audience-targeted documents (English first, German translation to
follow), linked from the README and a new docs/manual/ index:
- docs/features.md — public-facing feature overview: what Dorfteich
can do and why that matters
- docs/manual/user-guide.md — everyday use: editor, wikilinks, labels,
search, comments, watches/digests, import/export, settings
- docs/manual/pond-admin-guide.md — pond configuration: members/roles,
access rules incl. label scoping and public pages, labels, comment
policy, plugins, API/MCP opt-ins, files, export
- docs/manual/site-admin-guide.md — instance administration: wizard,
settings, quotas, uploads, API/MCP switches, legal pages, plugins,
users, and the system panel (jobs/backups/audit/storage)
- docs/manual/api-guide.md — example-driven public-API walkthrough
(tokens, reading, writing through the collab-safe path, labels,
comments, error semantics)
- docs/manual/mcp-guide.md — connecting AI assistants: switches, token
scopes, Claude Code one-liner, mcp-remote bridge, tool table, audit
and safety properties
- docs/developer/extending.md — plugin development (sandbox contract,
SDK, block plugins, bundled apps/fullscreen, shipping) and core
contributions (stack, dev environment, gates, house rules)
README: documentation index, repository-layout rows for docs/manual and
docs/developer, and the stale "architecture phase" status brought up to
reality. All relative links verified.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.