Supersedes #1353, per the discussion in #1352: one `API_KEY` environment
variable; when set, a request carrying it is authenticated as the first
admin, and cookie sessions keep working exactly as before (cookie **or**
key). Nothing else changes. No table, no UI, no settings-tab switch, no
per-user keys.
## What
- `API_KEY` (env). Unset → the feature is off and none of the new code
runs.
- `Authorization: Bearer <key>` or `X-Api-Key: <key>` on any existing
`/api/*` route authenticates that request as the first admin in
`users.db` (`ORDER BY id`), or as a bare admin identity
(`user_id="api"`, `is_admin=True`, no local user row) if the install has
no admin yet. Per request only; nothing is persisted; the admin's role
is read live, so deleting or demoting that user takes effect on the next
request.
- Both headers are checked and either may match. That is what makes the
key usable behind a reverse proxy that injects its own `Authorization`
header (oauth2-proxy, Authelia, forwardAuth): send the key in
`X-Api-Key`.
- A credential that is **not** the key is ignored and the request
continues on the normal session path, so proxy-forwarded tokens are
unaffected. Without a valid session such a request gets the usual `401
{"error": "Unauthorized"}`, identical to a request with no credential,
so there is nothing to probe.
## How
- `shelfmark/config/env.py`: `API_KEY = os.getenv("API_KEY",
"").strip()`.
- `shelfmark/core/api_key.py`: `extract_api_key_candidates()` (Bearer
token if the scheme is Bearer, then `X-Api-Key`) and `matches_api_key()`
using `hmac.compare_digest` on bytes.
- `shelfmark/core/user_db.py`: `UserDB.get_first_admin()`.
- `shelfmark/main.py`: `api_key_auth_middleware` (`before_request`,
registered before `proxy_auth_middleware`, which early-returns for keyed
requests). Only `/api/` paths; `/api/health` and `/api/auth/*` exempt;
no-op when `API_KEY` is unset or the auth mode is `none`. On a match it
mirrors the proxy-auth pattern: `session.clear()` then populate
`user_id` / `is_admin` / `db_user_id` for this request, `permanent =
False`, `modified = False`, `g.api_key_auth = True`. An `after_request`
hook guarantees no `Set-Cookie` is written for a keyed request even if a
handler dirties the session.
- `docs/api-access.md` (new), the `API_KEY` entry in
`docs/environment-variables.md`, and a README link.
## Security
- Constant-time compare; the key is never logged or echoed.
- Keyed requests never mint or refresh a session cookie and ignore any
cookie sent with them (a non-admin cookie plus the key yields admin for
that request; the browser's own session is left untouched and usable).
- The mismatch path touches neither the session nor `g`, so a stray
bearer on a browser request can neither log the user out nor change how
their cookie is refreshed.
- Store errors during the admin lookup fail closed (`500 {"error":
"Authentication error"}`), never to anonymous.
- Verified against Flask's `save_session` / `should_set_cookie`
ordering, and under auth modes `none`, `builtin`, `proxy`.
## Tests
`tests/core/test_api_key_env.py` (36): extraction and matching;
first-admin lookup; middleware behaviour on a guarded route and an admin
route, with and without a user_db, `X-Api-Key`, both-headers
combinations, no `Set-Cookie` when a handler dirties the session,
incoming non-admin cookie ignored, browser cookie still usable after a
keyed request, security headers, store error → 500, mismatch → guard's
401 / cookie path / permanent cookie untouched, unset → off, exempt
paths and path probes, `none` and `proxy` modes, deleted and demoted
first admin, a keyed write passing the guard. Existing auth suites
unchanged. All CI gates green on the fork:
https://github.com/gavinmcfall/shelfmark/pull/2 (CI-only draft).
Also exercised against a running instance: 47 scripted checks including
150 concurrent requests, proxy-mode switching through the key, an
unset-key restart, and a log scan for the key.
## Naming
`API_KEY` as discussed. If you'd rather namespace it
(`SHELFMARK_API_KEY`) to avoid clashing with other tools' env vars in
shared compose files, it is a one-line change; say the word.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
With `AUTH_METHOD=proxy` and no admin group configured, every user the
proxy authenticates for the first time is provisioned as an admin
(`is_admin = True` unless the user already exists in `users.db`). The
intent to never lock an instance out makes sense, but the effect is that
anyone the SSO gate lets through becomes an administrator. On an
instance shared with family or a small community that is a footgun; I
hit it when the first invited reader landed as an admin.
This keeps the guarantee and removes the footgun: the first account is
still provisioned as an admin while the instance has no admin at all,
and later first-time users follow a new `PROXY_AUTH_DEFAULT_ROLE`
setting (Security tab / env), default `user`. Known users keep their
stored role; the `PROXY_AUTH_ADMIN_GROUP_NAME` path is unchanged and
still takes precedence. I couldn't find a way with Cloudflare access to
pass this along.
Changes: `UserDB.has_admin()`, `_proxy_default_is_admin()` in the proxy
middleware, the new `SelectField` beside the other proxy settings, the
regenerated `docs/environment-variables.md` entry and a row in
`docs/reverse-proxy.md`.
Compatibility: the default moves from "everyone admin" to "first admin,
then users". Accounts already in `users.db` are unaffected; new SSO
users on an existing instance become regular users unless
`PROXY_AUTH_DEFAULT_ROLE=admin` is set. If you would rather ship this
purely opt-in I can flip the default to `admin`.
## Verification
- `tests/core/test_auth_api.py::TestProxyProvisioningRole`: first user
admin / second user not; `PROXY_AUTH_DEFAULT_ROLE=admin` restores the
old behaviour; an admin from another auth source counts as "an admin
exists"; a known user keeps their role whatever the default.
- Full suite (3094), ruff, ruff format, basedpyright, vulture green.
- Running on my own instance since 2026-09-19.
Add **TorBox** as a torrent download client for Prowlarr releases.
Users can select `TorBox` in the download client settings, configure it
with the new `TORBOX_API_KEY` environment variable, and verify their
credentials with the connection test button.
The integration supports both magnet links and `.torrent` files. It
tracks the torrent lifecycle through TorBox, downloads supported book
and audiobook files from the TorBox CDN, preserves safe nested file
paths, and cleans up remote and local download state.
Important: Shared HTTP download logs omit full download URLs and
URL-bearing exception text to avoid exposing credentials, following best
practices. This applies to all clients that use the shared
`download_url()` path; URLs remain available to the HTTP operations
themselves.
---
There is already related work in progress in #1173, which includes both
torrent and direct-download support for TorBox.
This PR is not intended to replace or compete with that contribution. It
offers the tested torrent client functionality as a smaller, focused
change that can make TorBox available to the community sooner. The
direct-download integration proposed in #1173 remains valuable and could
be reviewed or introduced separately.
Automated tests cover configuration, connection validation, magnet and
torrent-file submission, API errors, status and progress handling, file
retrieval, path traversal protection, cancellation, cleanup, and
sensitive URL redaction.
I also validated the complete flow locally with several magnet links and
`.torrent` downloads. TorBox processed the torrents and Shelfmark
downloaded the resulting files as expected.
AI was used to help with the implementation, with human validation. This
PR and long description? Took me some good minutes at night after work,
but gives me joy to open this PR to share with the community this
improvement.
Closes#1230
## What
Adds a "Word Separator" setting (Space / Dot / Underscore / Hyphen /
Custom) that replaces internal whitespace in each naming-template
placeholder's rendered value — e.g. `{Author}` renders
"Arthur.Conan.Doyle" instead of "Arthur Conan Doyle" when Dot is
selected.
This follows option 2 from the issue rather than inventing new
dotted-keyword template syntax (`{Author.}`), since it's a smaller
surface: one setting applies uniformly across all four templates
(books/audiobooks × rename/organize) instead of needing a parallel token
for every existing one.
## How it works
- Literal characters typed into the template itself (e.g. the `.` in
`{Author}.-.{Title}`) are never touched — only whitespace *inside* a
placeholder's resolved value is affected.
- Default is "Space", which is a no-op: existing templates produce
byte-identical output after this change (verified via the existing test
suite, unmodified, still passing).
## Where
- `shelfmark/core/naming.py` — `word_separator` param on
`parse_naming_template` / `build_library_path`.
- `shelfmark/download/postprocess/policy.py` — `get_word_separator()`,
mirroring the existing `get_file_organization()` accessor.
- `shelfmark/download/postprocess/transfer.py` — wires the resolved
separator through the four existing template-rendering call sites.
- `shelfmark/config/settings.py` — new `Word Separator` / `Custom Word
Separator` fields next to the existing naming-template fields.
- `src/frontend/.../namingTemplatePreview.ts` +
`NamingTemplateField.tsx` — the settings UI has its own TS mirror of the
Python renderer for the live preview; updated it in lockstep so the
preview doesn't lie about what the separator will actually do.
- Tests added on both sides (pytest + vitest).
## Testing
- `uv run pytest tests/core/test_naming.py
tests/core/test_destination_file_organization.py` — all pass, including
new cases.
- `uv run pytest` (full suite) — same pre-existing failures as on `main`
before this change (browser/network-dependent bypass & e2e tests
unrelated to this diff), everything else green.
- `uv run ruff check` / `ruff format --check` / `basedpyright` — clean.
- `npm run lint` / `format:check` / `typecheck` / `test:unit` (196
tests) — clean.
Closes#930.
## What
New `{FirstAuthor}` naming-template token. It renders only the first
author when metadata lists several ("Author1, Author2, Author3"), so
multi-author books can be filed alongside the rest of that author's work
instead of getting their own "Author1, Author2, ..." folder.
```
{Author} -> Terry Pratchett, Neil Gaiman
{FirstAuthor} -> Terry Pratchett
```
## How
- Added to `KNOWN_TOKENS` in `shelfmark/core/naming.py`, positioned
before `author` so `{FirstAuthor}` isn't parsed as literal `First` +
`{Author}`.
- Derived inside `parse_naming_template` from the existing `Author`
value (split on `,` / `;`), so every caller — folder transfer, rename,
the settings preview — picks it up with no extra wiring. An explicit
`FirstAuthor` key in the metadata still wins if one is ever passed.
- `{Author}` behaviour is unchanged.
- Frontend `namingTemplatePreview.ts` token list + `KNOWN_TOKENS` kept
in lockstep (there's a test enforcing that), with a matching
`firstAuthor` helper.
- Settings field descriptions + `docs/environment-variables.md` list the
new token.
## Known limitation
A lone author written `Last, First` is split on the comma too and
renders as `Last` — the source metadata doesn't mark which form it is.
Called out in the token help text and covered by a test. `{Author}`
remains available for anyone who wants the raw string.
## Checks
- `make python-test` — 2963 passed
- `make python-lint` / `make python-format` / `make python-typecheck` /
vulture — clean
- `make frontend-test` — 187 passed · `frontend-lint` /
`frontend-format` / `frontend-typecheck` — clean
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Follow-up to #1276 with a measurement from the instance I reported
there. v1.3.13 solves the challenge again, but on my setup the solve was
being thrown away immediately afterwards:
```
19:26:08 Bypass successful using _bypass_method_cdp_gui_click
19:26:16 Bypass failed (attempt 1/10): TimeoutError: Time ran out while waiting for: {html}
```
`_get()` ends with `return await page.get_page_source()`, which is
`find("html", timeout=1)` in SeleniumBase. One second is enough for a
page that is already sitting on its content, but Anna's Archive answers
a cleared check with a redirect to the real page, so the document is not
there yet. The solve is discarded, the whole attempt restarts, and the
extra requests are what earn the 429 that `note_rate_limited()` then
parks the host for — 120 s, then 300 s.
## Change
`_read_page_source()` waits for the document itself, with a
`BYPASS_PAGE_SOURCE_TIMEOUT` setting (default 20 s, min 1, max 120) in
Direct Download → Cloudflare Bypass, next to the existing bypasser
timeouts.
## Measured on a live instance
I patched the wait in the running container (`find("html", timeout=1)` →
`timeout=20` in the installed seleniumbase, which is the same effect as
this PR) and re-ran the same searches on the same host, k3s behind a
Surfshark WireGuard exit, internal bypasser, v1.3.13:
| | 1 s wait | 20 s wait |
|---|---|---|
| `Time ran out while waiting for: {html}` | one per solve | none |
| 429 backoffs | 2 (120 s, then 300 s) | none |
| Search for a book AA has | 199 s and 200 s, both errored | 61 s, 2
epub releases |
A download after that took 5 s from LibGen, so the search was the whole
cost.
## Tests
Two tests in `tests/bypass/test_bypass_budgets.py`, the file already
covering #1276: a page that needs longer than a second still yields its
HTML, and `BYPASS_PAGE_SOURCE_TIMEOUT` overrides the default.
`uv run pytest tests/ --ignore=tests/e2e`: 2848 passed, 47 skipped. Ruff
check and format clean. The docs table is auto-generated, but running
`scripts/generate_env_docs.py` here rewrote unrelated entries (Newznab,
BOOK_LANGUAGE), so I added only the new entry by hand in the generator's
format rather than commit that churn.
One thing I could not judge from outside: whether 20 s is the right
default for hosts other than AA. It only costs anything when a solve
would otherwise be discarded, but I have measured it on one site.
## Summary
- add a named Newznab indexer table with per-indexer URL and API key
settings
- search every configured indexer and retain the originating indexer
name on each result
- namespace cached release IDs across connections and isolate individual
indexer failures
- preserve the legacy single-indexer settings as a fallback
- support masked API-key cells and trusted SABnzbd prefetching for named
indexers
## Validation
- 121 Newznab and SABnzbd backend tests passed on Python 3.14
- Ruff passed for all changed Python files
- frontend TypeScript and strict lint checks passed
- all 134 frontend unit tests passed
- frontend formatting check passed
## Compatibility
Existing `NEWZNAB_URL` and `NEWZNAB_API_KEY` configurations continue to
work whenever `NEWZNAB_INDEXERS` is empty.
Co-authored-by: Ryan <zab1996@users.noreply.github.com>
## Why
`BOOK_LANGUAGE` is a per-reader property, not a per-instance one. On a
shared install one household member searches in German while another
wants English and German — today whoever changes the setting changes it
for everyone, and the only escape is re-picking languages in the filter
on every single search.
The per-user override machinery already carries `SEARCH_MODE`, the
metadata providers and the default release sources, so the language
default mostly had to opt into it.
## What changed
**The field.** `BOOK_LANGUAGE` becomes `user_overridable` and moves from
the **General** tab to **Search Mode**, next to the other
user-overridable search defaults (per
[review](https://github.com/calibrain/shelfmark/pull/1255#issuecomment-5391189094)
— the first version had the Search section span two tabs, this one
doesn't). Admins set it per user in the user editor, users set it in
**My Account → Search Preferences**, and the Search Mode tab carries the
usual "N users override this" summary.
**No migration for the move.** `general` and `search_mode` both persist
into `settings.json`, and a field's value is resolved through
`load_config_file(tab)` for the tab it's declared on — so an install
that already stores `BOOK_LANGUAGE` keeps its value. Checked against a
`settings.json` written while the field still lived on General: the
stored value resolves unchanged, a fresh install still gets `["en"]`,
and `BOOK_LANGUAGE` in the environment still overrides both.
**The two places the default is read.**
- `/api/config` seeds the frontend's language filter, so it now resolves
`BOOK_LANGUAGE` for the session user.
- `build_release_search_plan` falls back to the default whenever a
request carries no language filter — which is exactly what the filter's
"Default" option sends. It takes an optional `user_id`, passed by
`/api/releases` from the session and by the Prowlarr retry path from
`task.user_id`, so a retry re-searches in the languages of whoever
queued the download.
**Validation.** Overrides go through `normalize_language()`, so
`"German"`, `"ger"` and `"de"` all store as `de`, and an unknown
language is rejected with a message naming it instead of being silently
searched for. An empty list stays an empty list (a deliberate "no
default filter"), `null` clears the override as everywhere else, and ENV
still wins: with `BOOK_LANGUAGE` set in the environment the field
reports `fromEnv` and overrides are ignored.
**Scope.** Only the language default becomes overridable. The two format
lists left behind under "Default Search Filters" stay admin-only — they
describe what the library and its post-processing accept, not what a
reader wants to read. There's a test pinning that.
## Verification
- 2681 unit tests pass (2670 before, 11 added)
- `ruff check`, `ruff format`, `basedpyright` over backend and tests,
and `vulture` all clean; frontend lint, format, typecheck and 126 unit
tests clean
- `docs/environment-variables.md` regenerated via
`scripts/generate_env_docs.py` (the `BOOK_LANGUAGE` row follows the
field into the Search Mode section)
- Manually against a two-user instance with builtin auth (first round,
before the tab move): with user A on German and user B on
English+German, `/api/config` returns each reader their own
`default_language` and an unfiltered `/api/releases` plans the matching
languages; an admin can set and read the same override for another user;
clearing it falls back to the global value; a stray `"klingon"` is
rejected; and `BOOK_LANGUAGE` in the environment overrides both users
with the field marked `fromEnv`
- After the tab move I re-ran the suites above plus the
stored-value/fresh-install/ENV check described under "No migration for
the move"; the behaviour it exercises is what the move could have broken
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: CaliBrain <calibrain@l4n.xyz>
## Problem
Some trackers — MyAnonamouse in particular — distribute AAC audiobooks
as per-chapter `.mp4` files. That's the same ISO-BMFF container as
`.m4a`/`.m4b`, just with the generic extension (`ftyp isom`,
audio-only).
Today those releases:
1. show up in Prowlarr search results with **no format chip** — only the
generic "Audiobook" icon, because no format could be inferred;
2. download successfully; then
3. fail post-processing with **"No book files found in download"**,
because `.mp4` isn't in `AUDIOBOOK_FORMATS` (`shelfmark/core/utils.py`).
Real example: MAM #627978, *The Martian* (Andy Weir, 2020 edition) — 142
files `0001 … 0142 Andy Weir (2020) The Martian.mp4` + `cover.jpg`, 305
MB. Every file is a valid AAC-in-MP4 chapter.
Adding `mp4` to `SUPPORTED_AUDIOBOOK_FORMATS` in `settings.json` doesn't
help since the hard-coded tuple is what post-processing scans against.
## Change
- Add `"mp4"` to `AUDIOBOOK_FORMATS` (single source of truth — settings
UI, Prowlarr parsing, IRC parser, archive extraction and post-download
scan all derive from it), with a comment explaining why.
- Add `".mp4"` to the two hand-maintained debrid `_BOOK_EXTENSIONS`
lists (AllDebrid / Real-Debrid) so file selection matches.
- Slot `mp4` into the IRC `AUDIOBOOK_FORMAT_PRIORITY` table right after
`m4a` (same container family).
- Update the documented default in `docs/environment-variables.md`.
- New regression test
`test_audiobook_multifile_mp4_chapters_are_book_files` modelled on the
existing multi-file usenet test.
### Note for existing installs
The legacy-default migration only widens configs that still hold the old
`m4b,mp3` list, so users on the current widened default won't pick up
`mp4` automatically — they'll need to tick it in Settings → Audiobook
formats. New installs get it by default. Happy to extend the migration
if you'd rather it be automatic.
## Testing
- `ruff check` / `ruff format --check`: clean
- `pytest tests/core tests/config tests/irc tests/prowlarr
tests/download -m "not integration and not e2e"`: 2296 passed, new test
+ `test_audiobook_format_consistency.py` all green. The 10 failures in
`test_entrypoint_permissions.py` / `test_orchestrator_stall.py`
reproduce identically on untouched `main` on macOS and are unrelated.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Follow-up to #1237. \`rename_and_group\` only grouped when the source
root was a directory, so a multi-file audiobook delivered as a single
archive fell through to the flat path: a \`Book.zip\` of twelve chapters
landed loose in the destination root with its original chapter names —
the layout #1181 is about.
The \`is_dir()\` guard was there to keep \`Book.zip/\` from becoming the
folder name, but skipping the file case gives up the grouping instead of
naming it. A non-directory source can only produce several book files by
having been extracted (\`collect_staged_files\` returns a single-element
list for every other file shape), so the archive stem is the release
name and the suffix is packaging: group under \`Book/\`.
Also regenerates the env docs for the new option and gives it the same
\"do not use with ingest folders\" caveat Rename and Organize carries,
since both now create directories in the destination.
Tested: reverting only the source fix makes both new tests fail and the
\`rename\` control case pass, so grouping stays opt-in. Full non-e2e
suite green (2653 passed).
Two independent bugs, both from an indexer that Prowlarr proxies rather
than answers for itself: the search never reported that it had failed,
and the grab never resolved what it was handed.
Search. A Torznab search is Prowlarr proxying a live request out to the
tracker, so for a Cloudflare-fronted indexer it waits on FlareSolverr.
The client gave it the 30s budget sized for Prowlarr's own JSON
endpoints, then swallowed every failure -- the timeout, the 429 Prowlarr
returns once it has disabled an indexer, a parse error -- into the same
empty list that means "this indexer has nothing". A cold challenge
routinely runs past a minute, so the UI said "No releases found for this
book" while FlareSolverr was still solving. That empty list also drove
the auto-expand retry, which fires on "no results with the category
filter". A timeout satisfies it, so Shelfmark sent a second search to an
indexer still busy with the first -- two Chromes at once, enough to take
FlareSolverr's down on a small host.
torznab_search now raises ProwlarrSearchError, and an empty list
strictly
means the indexer answered with no matches. The source records which
indexer searches failed: one dead indexer no longer sinks the others,
auto-expand runs only when every indexer genuinely answered, and zero
results with at least one failure raises SourceUnavailableError, which
the releases endpoint already turns into a 503 carrying a real message.
Prowlarr being unreachable was the same lie by another route -- the
indexer list came back empty, leaving nothing to query -- and now says
so.
Indexer searches also get their own timeout, PROWLARR_INDEXER_TIMEOUT,
defaulting to 90s and clamped to 5-300. Prowlarr's status and indexer
list keep 30s so Test Connection stays responsive, and the connect
timeout is split out at 10s so an unreachable Prowlarr fails fast rather
than hanging for the whole read budget. The overall per-request search
budget now scales to twice the indexer timeout, capped at 240s, so
raising the setting is not undone by the cap one level up while staying
under the 300s gunicorn worker timeout.
Grab. Prowlarr hands out a proxy download URL, with no magnetUrl and no
infoHash, for any indexer that only publishes torrent files. The native
Real-Debrid client built its magnet as "if not
url.startswith('magnet:') and expected_hash", so with no hash to work
from it left the URL alone and POSTed it to /torrents/addMagnet as the
magnet field. Real-Debrid answered 404 and the grab died on a raw HTTP
error. AllDebrid carried the same line and the same bug.
Both now resolve the URL first, through the extract_torrent_info path
the
torrent clients have used since #1108: pass a magnet through untouched,
follow a redirect or a response body that turns out to be a magnet,
otherwise upload the fetched .torrent, and fall back to a magnet built
from the infoHash only when the fetch failed. The file is preferred over
a synthesized urn:btih: magnet because it carries the tracker list; a
bare hash leaves the service to find the swarm on DHT alone. Fetches are
shared with the rest of the add path through the torrent fetch cache, so
resolving costs at most one request. Real-Debrid takes the file on PUT
/torrents/addTorrent with the raw bytes as the request body, AllDebrid
on
POST /magnet/upload/file as multipart files[]. A URL that resolves to
neither form now raises before any request reaches the service, so the
user reads why instead of a 404. Neither debrid client had any test
coverage; both have some now.
Fixes#1249Fixes#1250
Every protected request spawns a fresh helper subprocess, paying
interpreter start and imports before any work begins. Measured inside
the container, five consecutive runs of `python -c "import
shelfmark.bypass.internal_bypasser"`:
```
3.53s 3.45s 3.55s 3.54s 3.46s
```
A single search issues several protected requests, so that is paid
several times over per search.
## What changed
The helper now serves one JSON request per line of stdin until the
parent closes the pipe, and an idle timer
(`BYPASS_BROWSER_IDLE_TIMEOUT`, default 180s) shuts it down once
searching stops.
Answers still travel by result file, but the file is now written to a
`.part` path and renamed into place — the parent treats the file's
existence as the answer, so it must never observe a half-written one.
stdout and stderr stay attached to the parent's, so helper logs keep
appearing in `docker logs` exactly as before.
Failure handling, since a warm helper is exposed to more of it than a
per-request one ever was:
| Situation | Handling |
| --- | --- |
| Helper died between requests | Detected via `poll()`, respawned |
| Pipe broken at write time (`poll()` can miss this) | One retry on a
fresh process; a fresh one failing there is a real failure |
| Helper exits without writing a result | `RuntimeError` naming the exit
code |
| Wedged past the timeout, or cancelled mid-bypass | Helper killed, then
`_cleanup_orphan_processes` because a killed helper never got to close
Chrome |
| Idle reaper racing an arriving request | Re-checks the deadline under
the lock and re-arms instead of killing a helper that just did work |
The DNS config now travels with every request rather than only at spawn:
a warm helper outlives changes the parent makes to its provider.
## `BYPASS_REUSE_BROWSER`, off by default
This parks the CDP driver between bypasses. A driver's websockets are
bound to the loop that opened them and cannot outlive their process, so
the persistent helper is what makes this possible at all — and the warm
path runs on `_CDP_WORKER`'s long-lived loop rather than `asyncio.run`
for the same reason.
The mechanism works. With it on, the browser start disappears from the
second request onward: 0.7s from `Reusing warm Chrome browser` to the
first bypass attempt, against roughly 16s cold.
**It still ships off, because a matched-pair test shows it is a net loss
against DDoS-Guard.** Each round primed with one cold bypass, waited
10s, then measured a second — identical timing in both arms, only the
browser strategy differing, order balanced (fresh, warm, warm, fresh) so
drift over the session cannot masquerade as an effect:
| Arm | Measured request |
| --- | --- |
| fresh browser | 42.8s, 40.6s |
| warm browser | 57.1s, 59.6s |
Spread within each arm is 2.2s and 2.5s, against 16.7s between them.
Reuse removes the ~15s browser start and then gives back roughly twice
that in solving: a returning browser draws a harder challenge. Where the
cold browser is through on the second bypass method, the warm one fails
the first three and only `_bypass_method_humanlike` gets it, at ~30s for
that method alone.
Worth separating from a second effect I ran into while measuring: five
back-to-back searches slow from ~32s to 51–98s with reuse **disabled**
as well, so DDoS-Guard escalates on request rate independently of any of
this. That is why the pairs above are timed identically rather than
simply run in sequence. It is the larger of the two effects, but not
something this project can patch around.
Reuse is left available rather than dropped because Cloudflare sites may
not respond the same way, and because the two concerns are independent:
the helper start is pure overhead and always worth removing, the browser
is not.
## Verification
- 2559 unit tests pass (2542 before, 17 added in
`tests/bypass/test_warm_browser.py`)
- `ruff check`, `ruff format`, `basedpyright` over backend and tests,
and `vulture` all clean
- `docs/environment-variables.md` regenerated via
`scripts/generate_env_docs.py`
- Live against Anna's Archive on a warm helper: searches return their
usual ~760KB and 667 results, the app's own search warm-up completes
with 50 results, and the container is left with no orphan
chrome/Xvfb/ffmpeg processes
Happy to drop the `BYPASS_REUSE_BROWSER` half entirely if you would
rather not carry a default-off path — the helper persistence stands on
its own.
Co-authored-by: helgehelge123 <helge.neumann@zollsoft.de>
Hardcover replaced its ~500 char JWTs with short opaque personal access
tokens ("hc_pat_..."), and the connection test rejected anything under
100 chars before a request ever left Shelfmark, so every newly created
key failed with "API key seems too short".
The length floor now applies only to keys without the hc_pat_ prefix; a
prefixed key goes straight to Hardcover, which is the authority on
whether it is valid. Also strip a pasted "bearer " prefix regardless of
casing -- Hardcover's docs tell users to paste the token into an
"authorization" header, so the prefix rides along on the copy, and the
old case-sensitive removeprefix() sent it through as part of the token.
The API key field now names the expected shape.
Note that Hardcover's PAT path currently answers every hc_pat_ token
with a 500, a fabricated one included, while non-PAT tokens still get a
clean 401. So a new key cannot connect yet regardless of this change --
that failure is server-side and not something this code can reach.
Refs #1240
Newznab searches hardcoded category 7000 for ebooks and 3030 for
audiobooks,
so indexers using custom IDs returned no results or the wrong ones. Add
NEWZNAB_EBOOK_CATEGORIES and NEWZNAB_AUDIOBOOK_CATEGORIES (tag lists,
defaulting to 7000 and 3030) and resolve the search categories from
config.
Values are parsed leniently — list or comma/whitespace separated,
non-numeric
entries skipped, duplicates dropped — and fall back to the standard IDs
when
empty, so a cleared field can't silently widen the search to every
category.
NEWZNAB_AUTO_EXPAND remains the way to do that on purpose.
Results carrying a custom ID outside the standard 7000-7999 / 3030
ranges were
typed as "other", which routed custom-category audiobooks as ebooks.
Trust the
searched content type when a result carries a category we explicitly
asked for.
Also drop the unused NEWZNAB_BOOKS / NEWZNAB_AUDIOBOOKS constants from
api.py —
a third copy of the same hardcoding.
Closes#1208
An IRC audiobook search returned nothing while OpenBooks, reading the
same @search answer from the same channel, listed results. Three
separate defects were discarding them.
The audiobook format list was maintained by hand in four places and had
drifted. The settings UI offered only m4b/mp3/m4a/zip/rar, and that list
is the only one a user's config can be built from, so flac, opus, ogg,
aac, wav and wma were unreachable everywhere — even though the IRC
parser recognized them, the IRC sorter ranked them (dead code that could
never fire), archive extraction knew them and Prowlarr searched for
them. A FLAC audiobook was invisible in search and, if it arrived
anyway, rejected after download as "format not supported".
AUDIOBOOK_FORMATS and ARCHIVE_FORMATS now live once in core.utils and
every layer derives from them, which also restored the missing .opus in
the post-download scan's trackable extensions.
Widening the default alone would not have reached anyone already
affected: initialize_default_configs() writes field defaults only when a
tab has no config file yet, so an existing install keeps its persisted
m4b/mp3 list forever. migrate_audiobook_formats rewrites a list that
still matches the old default exactly and leaves every other value
alone — re-enabling formats someone had deliberately turned off would be
worse than leaving them narrow.
The IRC parser filtered by file extension alone. Multi-file audiobooks
ship as a .rar or .zip of MP3s, which matched neither SUPPORTED_FORMATS
nor SUPPORTED_AUDIOBOOK_FORMATS, so they fell out of the ebook bucket
and the audiobook bucket both. Results are now classified before the
format filter is applied: an audio extension means audiobook, an ebook
extension means ebook, and for a container — where the extension says
nothing about the contents — the release name decides. An ebook archive
stays out of audiobook results.
RESULT_LINE_REGEX matched \w+ after any dot, so a line carrying no file
extension parsed as format "5mb" out of "::INFO:: 620.5MB", taking the
title and the size down with it and guaranteeing every downstream filter
dropped it. Any decimal size did this. The extension is now matched
against the known formats, so such a line falls through to the simple
pattern and comes back as "unknown", which the rest of the parser
already handles. ALL_RECOGNIZED_FORMATS became an ordered tuple in the
process: it was a set, so which extension won for a line naming two of
them depended on set iteration order and could vary between restarts.
Refs #1129
LOG_LEVEL never reached the app logger: env.py hardcoded the level to
DEBUG or INFO, so INFO lines kept appearing under LOG_LEVEL=error. Read
it from the env var and advanced settings, normalize unknown values to
INFO, and expose it as a setting. entrypoint.sh now normalizes
gunicorn's level too, so a typo falls back to info instead of stopping
the container from booting.
Z-Library gates the first hit on /md5/<hash> with a 503 whose only
payload is a Set-Cookie; echoing that cookie back returns the 302 to the
real page. html_get_page dropped it and re-ran the same rejected request
on every retry, ending in "No download URL resolved". Retry once with
the
cookies the 503 issued.
Fixes#1185Fixes#1187
## Summary
- add a **Change Category** torrent completion action and conditionally
show its post-import category/label setting
- update kept torrents only after a successful library import
- support qBittorrent categories, Transmission labels, Deluge's Label
plugin, and rTorrent's `custom1` label
- preserve existing Keep/Remove behavior and document the new
environment setting
## Behavior
Category changes happen from `post_process_cleanup`, after output
transfer and post-processing complete. This keeps the existing hardlink
flow unchanged. An empty post-import category is a no-op, and client API
failures are logged without turning a successful library import into a
failure.
## Validation
- `pytest -n 0 tests/prowlarr/test_qbittorrent_client.py
tests/prowlarr/test_transmission_client.py
tests/prowlarr/test_deluge_client.py
tests/prowlarr/test_rtorrent_client.py tests/prowlarr/test_handler.py
tests/config/test_generate_env_docs.py` — 161 passed
- `ruff check` on all changed Python files — passed
- `basedpyright` on the changed client implementation files — passed
- full `basedpyright shelfmark/download/clients` currently reports two
pre-existing errors in the new debrid connection-test code at
`settings.py:546` and `settings.py:563`, outside this PR's diff
- multi-architecture Docker images built successfully for `linux/amd64`
and `linux/arm64`
Fixes#1138Fixes#1141
## Problem
Two language editions of one book resolve to the same canonical title,
so they render to the same path and the second gets a `_1` collision
suffix. Audiobookshelf treats a folder as exactly one library item, so
the pair becomes a single book with both files as tracks and a summed
runtime.
Shelfmark already parses and displays the language. It just never
reached the template engine.
## `{Language}` template variable
A template like `{Author}/{Title}{ (Language)}/{Author} - {Title}` now
yields:
```
/library/J K Rowling/Harry Potter (sv)/J K Rowling - Harry Potter.m4b
/library/J K Rowling/Harry Potter/J K Rowling - Harry Potter.m4b
```
The untagged edition's path is byte-identical to today, so no existing
layout shifts.
Three details worth flagging:
**The value is casefolded.** On a case-insensitive filesystem `(SV)` and
`(sv)` would collapse back into one folder, reintroducing the exact
collision being fixed.
**Values meaning "we don't know" render nothing** rather than producing
`Project Hail Mary (unknown)` folders. Anna's Archive reports that
string literally (`direct_download.py`, `language = detected or
"unknown"`).
**The frontend wasn't sending the release language at all**, so the
token would have stayed empty for exactly the audiobook sources in the
report. Prowlarr and AudiobookBay do not put language in `extra` the way
`direct_download` does, hence the payload plumbing. It reads
`release.language`, never `book.language` — the latter is the provider's
canonical edition and would mislabel a translation, with a regression
test for that specifically.
Not gated to audiobooks: Calibre-Web-Automated stages ingested files by
basename and discards folder structure, so the rename (filename)
template is the only lever those users have. Verified that form works:
`J K Rowling - Harry Potter (sv).epub`.
## Language consolidation (#1141)
Three release sources each carried their own alias map, all resolving to
the same ISO 639-1 codes, alongside a bundled database that only one of
them used. Adding a language meant editing three places.
Aliases now live in `data/book-languages.json` beside the code and name
they belong to, and `shelfmark/core/languages.py` resolves any of them —
two-letter code, ISO 639-2 three-letter in either the bibliographic or
terminological form, or English name. Prowlarr and AudiobookBay drop
their tables. Direct Download keeps its own path-parsing heuristics,
including the ambiguous short codes that collide with English words
(`de`, `en`, `no`, `in`), and takes only the alias data.
This also closes a coverage gap. MyAnonamouse offers 62 languages;
Prowlarr mapped 37, and an unmapped code is *dropped* rather than passed
through, so the other 25 carried no language at all — leaving
`{Language}` empty and the collision unfixed for Latin, Farsi, Tamil,
Urdu and the rest. Seven languages MAM offers had no database entry at
all: Bosnian, Burmese, Estonian, Icelandic, Manx, Scottish Gaelic,
Sanskrit.
Also fixes the Traditional Chinese code, which used a U+2011
non-breaking hyphen. Nothing compares against the ASCII spelling today
so it was latent, but it would silently defeat the first thing that did.
## Validation
Verified end to end against a live Prowlarr and MyAnonamouse, not just
unit tests. A real search returning both an English and a Swedish
edition, through the actual `queue_release` → `DownloadTask` → naming
path:
```
STEP 1 real MAM search -> 37 releases, languages: ['en', 'sv']
STEP 3 queue_release -> task.language='sv'
STEP 4 build_metadata_dict -> metadata['Language']='sv'
STEP 5 build_library_path -> /library/J K Rowling/Harry Potter (sv)/...
two language editions resolve to DIFFERENT folders: True
```
The refactor is pinned by a snapshot of both per-source maps taken
*before* they were deleted. All 131 aliases are asserted to still
resolve to the same code, one parametrised test each, so a regression
names the specific alias.
Also verified: the filename-only template, the retry round-trip
(`serialize_task_for_retry` → `_restore_task_from_retry_payload`, plus a
legacy payload with no `language` key), and placeholder handling.
Added a `KNOWN_TOKENS` ordering invariant test — `find_placeholder()`
does a substring `.find()` in list order and nothing protected that
contract, so a future token in the wrong position could silently shadow
an existing one. And a lockstep guard on the frontend, since
`KNOWN_TOKENS` is hand-duplicated in TypeScript.
**One caveat worth stating.** Three MAM codes are confirmed by
observation (`ENG`→`en`, `SWE`→`sv`, `MAL`→`ml`, the last from a real
`[MAL / EPUB]` Tagore release). The remaining ~59 are derived from ISO
639-2 rather than observed, because MAM's catalogue is overwhelmingly
English — enabling 27 extra languages still yielded only one non-English
hit across 258 results. Mitigated rather than closed: both 639-2
variants are present for every language where they differ, and a wrong
alias is an unused entry while a missing one loses the language. Happy
to correct any code a maintainer knows differs.
## Test results
2056 Python tests pass (up from 1906). Frontend typecheck, lint, format
and 126 unit tests pass.
Pre-existing failures on my machine, unchanged by this branch and
unrelated: `tests/bypass/` needs `seleniumbase`, and
`tests/config/test_entrypoint_permissions.py` uses bash-4 syntax that
macOS bash 3.2 rejects.
---------
Co-authored-by: delize <4028612+delize@users.noreply.github.com>
Co-authored-by: CaliBrain <calibrain@l4n.xyz>
#1140 fixed the guid-only dedup that hid results from filter-specific
indexer entries, but shipped the new behaviour on by default:
PROWLARR_COLLAPSE_DUPLICATES defaulted off, so every existing Prowlarr
user got extra rows for any release that two indexer entries both
returned, and the setting only let them opt back into what they already
had.
Default it on. The dedup key stays indexer-qualified, so the entries are
still distinct internally; collapse then merges them back to one row,
resolved by the Prowlarr priority rather than by query order as before.
The visible result set matches what users had prior to #1140, and anyone
who wants the per-entry rows (freeleech and the like) turns the setting
off.
Beyond the noisier list, the default mattered because split rows differ
only by indexer name while sharing a title, size and peer count. Two of
them have distinct source_ids, so the queue's duplicate guard does not
fire, and the second grab's find_existing() matches the first by
infohash and runs post-processing over the same download again,
delivering the book twice.
The search-side fallback in config.get(..., True) is flipped to agree
with the field default. In production the field default governs, since
the config cache is seeded from the registry and the fallback only
applies to an unregistered key. The source tests monkeypatch config.get
with a plain dict lookup, though, so the fallback is what they exercise:
leaving it False would have kept every default-behaviour test asserting
the opposite of what ships. A new test pins the two together.
The deduplication tests now opt out explicitly, since they assert the
split itself. test_collapse_off_by_default_keeps_both_rows becomes a
pair, one for the untouched setting collapsing to a single row and one
for opting out.
docs/environment-variables.md is regenerated rather than hand-edited. It
was already stale on main, so it also picks up RTORRENT_AUDIOBOOK_LABEL,
DIRECT_DOWNLOAD_LANGUAGE_FROM_PATH, and reworded IRC_SEARCH_BOT and
RTORRENT_LABEL text from earlier merges.
The rest is fallout from the ruff 0.16.0 bump in #1139, which enabled a
much larger default rule set and started formatting Python code blocks
in Markdown: _find_existing_alias_user() uses min() instead of
sorted()[0] (FURB192, currently failing Python Quality on main), and the
two READMEs get their code blocks reformatted.
Results were deduplicated on `guid` alone. When one tracker is
configured in Prowlarr as several indexer entries differing only by a
server-side search filter, all of them return the same guid for the same
torrent, so every entry but the first was silently discarded. Freeleech
and other filter-specific releases became invisible, replaced by the
unfiltered entry's copy, and which copy survived depended on query
ordering rather than user intent.
Include the indexer id in the dedup key so the entries stay distinct.
Release.source_id is qualified the same way. It keys the release cache
and becomes the download task id, so rows sharing a guid would otherwise
collide and a grab would route through whichever entry cached last,
defeating the point of showing them separately. The handler's task
matcher still accepts a bare guid or infoUrl so tasks queued before this
change still resolve.
Ordering now follows the priority already configured in Prowlarr (1-50,
lower preferred) rather than a new setting, since users curate that
ranking there and filtered entries are typically ranked ahead of their
unfiltered counterparts. The enabled-indexer list is fetched once and
reused for the enrichment check, so this costs no extra round trip.
Releases carry extra.indexer_priority, and the sort dropdown gains an
"Indexer priority" entry, ascending, alongside the existing alphabetical
"Indexer" sort; SortOption grew a default_direction for that, defaulting
to desc so "Peers" is unchanged.
PROWLARR_COLLAPSE_DUPLICATES (default off) optionally collapses a
release back to one row, resolved by the same Prowlarr priority. Left
off, every entry that carried a release keeps its own row, which is what
makes filtered results visible again.
Identity handling is defensive about partial payloads: a result that
cannot be identified is never dropped or merged, and collapse only
merges on a strong identifier (guid/downloadUrl/magnetUrl/infoUrl)
because merging on title alone would discard genuinely different
releases that share a name.
Fixes#1137
Co-authored-by: delize <4028612+delize@users.noreply.github.com>
[qBittorrent
5.2.0](https://www.qbittorrent.org/news#sun-may-03rd-2026---qbittorrent-v5.2.0-release)
(May 2026) added support for API key-based authentication in addition to
the existing username/password-based authentication.
This commit adds support for qBittorrent API key authentication to
Shelfmark, configurable via environment variable or settings UI. If an
API key is set at the same time as the username/password, API key will
be preferred for authentication.
Requires `qbittorrent-api` 2026.5.3, the version that added the
`api_key` argument, or newer. `403 Forbidden` responses are not retried
with API key authentication because a retry has no chance of succeeding.
Tested end-to-end with my live qBittorrent 5.2.3 instance (WebAPI
v2.15.1).
<img width="785" height="616" alt="image"
src="https://github.com/user-attachments/assets/8064e10e-9a7f-49f0-804e-4c441d23a1fc"
/>
- Add configurable completed-path wait for external clients : - Add a
configurable Advanced setting, DOWNLOAD_CLIENT_COMPLETED_PATH_TIMEOUT,
for how long Shelfmark waits after a torrent or usenet client reports
completion before treating the completed path as missing. - Keep the
default at the existing 60-second grace period, with a maximum of 3600
seconds
- Add e2e testing
- Add e2e testing
Should fix#861
Adds support for `content_type=combined` in URL search parameters,
letting users force combined-mode searches via a bookmarkable link
rather than relying on the last-used preference from localStorage.
The override is applied only in Universal mode and only when combined
mode is actually available (universal enabled, `show_combined_selector`
on, neither content type blocked by policy). Otherwise, it's silently
ignored, consistent with how `content_type=ebook`/`audiobook` already
behave outside Universal.
Existing `content_type=ebook`/`audiobook` URLs now also force combined
mode off, so the URL is authoritative regardless of prior preference.
Also adds a per-user `FORCE_COMBINED_SEARCH` setting that locks combined
mode on whenever it's available.
URL `content_type=ebook`/`audiobook` overrides are also superseded by
force-combined for the same reason: the search bar wouldn't let users
switch back, so honoring the URL param would leave them in a state they
couldn't escape from.
- Updated mirror selection
- Removed built-in mirror options, users must provide their own
configurations
- Set Universal search to default, added ability to disable direct
source
- Updated documentation
- Updated makefile
- Adds a non-root startup path at user 1000:1000 - skips privilege
escalation and ownership checks. Works e.g. for kubernetes deployments
(user 1000:1000 and runAsNonRoot enabled).
- Remove startup check/chown commands for user-owned folders. Checks can
be done with a "Test destination" button in settings which performs a
test write. Users are responsible for fixing their own permissions.
- Update docs
Closes#515
## Summary
This adds a new `audible` metadata provider backed by the Audimeta API.
The provider supports:
- Audible/Audimeta metadata lookup without authentication
- region selection (`us`, `ca`, `uk`, `au`, `fr`, `de`, `jp`, `it`,
`in`, `es`, `br`)
- ASIN book lookup
- ISBN lookup with fallback search
- series suggestions and series-order browsing
- richer audiobook metadata such as narrators, runtime, rating,
subtitle, cover, publisher, and series info
- configurable Audimeta base URL, timeout, cache usage, default sort,
and unreleased filtering
## Notes
A few Audimeta-specific integration details were needed:
- send a meaningful `User-Agent`, otherwise Audimeta rejects requests
with `403`
- send the `cache` parameter in the format Audimeta expects
- use `keywords` for general search instead of `query`, which gave
poor/irrelevant results for title-style
searches
## Validation
Tested locally with:
- `python -m py_compile shelfmark/metadata_providers/audible.py`
- `python -m pytest tests/metadata/test_audible.py -v`
- `python -m pytest
tests/metadata/test_metadata_provider_capabilities.py -v`
Also verified manually in a Podman test container:
- searching for `Discount Dan` returns Audible title `B0DXLXRNGG`
- book details and series metadata load correctly
## Scope
This PR intentionally keeps the change localized to the provider layer
and docs:
- new Audible provider
- provider registration
- provider docs
- generated environment variable docs
- Tweak manual search toggle position
- Refinements to the Hardcover list dropdown behavior
- Hide the content type dropdown when a content type is blocked for a
user
- Fixes to Hardcover author parsing to strip out initialed names
- Remove `env_supported=false` for security config options.
- Add admin config for self-settings options visibility. Remove delivery
preferences or notifications from the view.
- Add option to use Booklore's Bookdrop API destination instead of a
specific library
- Add download path options for all torrent clients
- Add new booklore API file formats
- Renamed cookie for better login persistence with reverse proxy
- Updated fs.py to try hardlink before atomic move from tmp dir
- Fix transmission URL parsing
- Fix scenario where file processing of huge files starves the
healthcheck
- Large enhancements to custom scripting, including passing JSON
download info, more consistent activation across output types,
decoupling from staging behavior, and added full documentation.
- Refreshed available AA URLs
- Fixed potential redirect from AA itself causing mirror cache errors
- Added fully customizable mirror list in UI
- Segmented rotation behavior to Auto mode only
Fixes#588
- Base URL option in settings for reverse proxy setups
- Fix NZB downloads not deleting on completion
- Fix handling for audiobook files over 100+ parts
- Fix prowlarr search timeout
- Fix prowlarr categorisation for expanded searches
- Added new book output option **upload to Booklore**, available in
download settings
- Got annoyed at my messy processing code while implementing Booklore so
refactored the whole thing
- Full black box file processing testing with randomised configuration
- Deluge: Connect via WebUI auth for simplified setup
- Added env vars documentation, auto generated via script, and unlocked
most settings to be used as env vars
CaliBrain
Gavin McFallandClaude Fable 5.1
splitsec2
Marcelo Rodrigo
Vinicius Gabriel
Jorge Lima
zab1996andRyan
jakesterpdxandClaude Fable 5
Tilian B
Adam Vigneaux
Alex SchittkoandCaliBrain
spindrift
Alex
cadric
arjunsrinivasan1997
Alex