- delete server/services/jellyfin.ts, webhook.ts, mqtt.ts and their tests
- strip jellyfin/mqtt imports and startup calls from index.tsx and settings.ts
- remove /jellyfin, /mqtt, /mqtt/status, /mqtt/test, /jellyfin/webhook-plugin endpoints from settings router
- clean ENV_MAP and isEnvConfigured of jellyfin/mqtt keys
- add db/index.ts migrations for series_key, duration_seconds, scan_status, scan_error, last_scanned_at (new columns absent on older dev DBs)
- move idx_media_items_series_key out of SCHEMA into migrate() so it runs after the column is added
- fix all test fixtures: drop jellyfin_id/series_jellyfin_id column refs, update MediaItem/MediaStream object literals to match current types
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
previously stop only killed the running ffmpeg process — the loop
immediately picked up the next pending job. now the abort signal
breaks the loop between jobs, so remaining items stay in the queue
column as pending jobs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
every "← Back to inbox" button (review, queue, done, per-item unapprove,
per-item reopen) now does the same thing: reset plan to pending/unsorted,
clear auto_class, delete non-running jobs. previously unapprove left
sorted=1 (sent to review instead of inbox) and each column had its own
SQL. now consistent: back to inbox always means "needs re-processing."
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
upsertJellyfinItem no longer runs analyzeItem or creates stream_decisions.
it inserts a minimal review_plans stub (pending, unsorted). all analysis
happens in processInbox. this means after scan, ALL items land in the
inbox — the "needs action" count equals the inbox count until processing
classifies them.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
scan emits pipeline_changed every 25 items, rescan endpoints emit on
completion. pipeline page listens and throttle-reloads all column data
(1s debounce) so inbox fills progressively without manual navigation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
processing runs in background, items appear in columns progressively via
SSE-triggered reloads. stop button aborts mid-run, remaining items stay in
inbox. remove skip/skip-all from inbox. fix column header height jitter by
giving subtitle slot a fixed height.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Per-item rescan now resets plan to inbox (sorted=0, auto_class=NULL),
clears external_raw, deletes pending jobs, then lets processInbox
handle language resolution + analysis when auto_processing is on
- Add getSeriesEpisodes() to jellyfin service for bulk episode discovery
- Add POST /api/review/rescan-series endpoint accepting seriesJellyfinId
+ optional seasonNumber — discovers new episodes, resets all matching
items to inbox, refreshes streams from Jellyfin, auto-processes if on
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
RescanConfig now only carries audioLanguages. Radarr/Sonarr library
loading, language resolution, and resolveSeriesTvdb callback removed
from rescan.ts, scan.ts, and webhook.ts. RescanResult no longer tracks
radarrHit/sonarrHit/missingProviderId counters. Tests updated: removed
authoritative-source and resolved-TVDB-enables-Sonarr tests (moving to
processInbox in a later task), added assertion that scan never sets
sonarr/radarr as language source.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
episodes without SeriesProviderIds.Tvdb (older Jellyfin, un-refreshed items)
fell back to the episode-level TVDB ID, which never matched Sonarr's
series-keyed library. add resolveSeriesTvdb callback that fetches the series
item from Jellyfin to get the correct series TVDB ID, with per-scan caching.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
escape hatch for items the audio pipeline can't usefully fix — e.g. a
release whose only audio track is English commentary and needs to be
purged so *arr can find a better one. two buttons on the audio detail
page:
🗑 Delete file — unlinks the file and drops our DB rows
(cascades streams, plans, decisions, jobs)
🗑 Delete & refetch — same, then asks Radarr/Sonarr to rescan
(to notice the deleted file) and trigger
an indexer search for a replacement
backend: POST /api/review/:id/delete { refetch }. the refetch step is
best-effort and its result ships under `refetch` on the response so the
UI can surface partial wins — file deleted + db clean even if Radarr
doesn't have the movie. helpers live on the existing *arr service
modules (triggerMovieRefetch, triggerEpisodeRefetch) and do the command
lookups + POST /api/v3/command calls themselves.
UI uses native confirm dialogs showing the file path. on success,
navigates back to /pipeline since the detail page points at a row that
no longer exists.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
the separate Library page was mostly duplicating what the pipeline
columns + the item detail page already show. move its two useful bits —
the stats row (total / scanned / needs action / no change / approved /
done / errors) and the scan control bar — into a compact two-row header
above the pipeline columns, drop the library items table entirely, and
redirect "/" to "/pipeline".
the scan SSE buffering logic moves verbatim into PipelineHeader.tsx so
the progress bar and the stats refresh on completion keep working. the
dead /api/scan/items endpoint and its parseScanItemsQuery +
buildScanItemsWhere helpers (plus their tests) go away with the UI;
/api/scan, /api/scan/start, /api/scan/stop, /api/scan/events stay.
nav loses "Library" — Pipeline is the only entry point now.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
adds stream_decisions.custom_language (ISO 639-2 code or null) so the
user can correct a mislabeled audio track — e.g. a Spanish dub tagged
"und" in the container — without going through Jellyfin. the override
wins over stream.language everywhere it matters: the analyzer reads it
for keep/remove decisions and track ordering, the ffmpeg command builder
writes it as both the language metadata tag and the harmonized track
title, and reanalyze preserves it across reruns and rescans.
on the audio detail page, each pending audio row swaps its language
cell for an inline <select> populated from LANG_NAMES. picking the raw
file language clears the override; anything else sets it and triggers a
server-side reanalyze so keep/remove + target_index update immediately.
a small ✎ hint marks overridden tracks. rebuilt commands tag the output
accordingly so Jellyfin reads the corrected language.
PATCH /api/review/:id/stream/:streamId/language validates the code
against LANG_NAMES (accepts ISO 639-1/2/2B aliases, rejects garbage)
and runs reanalyze inside.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
subtitle management (list/detail pages, /api/subtitles, subtitle_files
table, SubtitleFile types, predictExtractedFiles, nav link) moved to a
new sibling project at ~/Developer/netfelix-subtitles-manager/ where
it'll be rebuilt standalone later. this project now owns audio fixing
+ subtitle extraction only.
extraction still runs end to end: analyzeItem still marks every subtitle
stream as "remove from container", buildExtractionOutputs still wires
the -map 0:s:N + sidecar outputs into the ffmpeg command, and execute.ts
still flips review_plans.subs_extracted so verify.ts can check desired
state — just derived from the streams directly instead of by writing a
row per file to the now-gone subtitle_files table.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
column headers are now a fixed three-row layout (title / subtitle / button
row). every column always reserves all three rows so headers line up
regardless of contents; actions render disabled when their column is
empty instead of disappearing, which keeps the header height stable as
state changes.
the processing column gets a new "Auto-process Queue" checkbox that
mirrors the inbox's "Auto-process Inbox" toggle. backend adds an
auto_process_queue config, a maybeStartQueueProcessor() helper, a
POST /api/settings/auto-process-queue endpoint, and a hook in
enqueueAudioJob so approvals drain the queue hands-off when the toggle
is on.
reopen-all and per-item reopen now send items to the Inbox (sorted=0)
instead of back to Review. the done column's label and tooltip become
"← Back to inbox" to match, and the clear button moves to the right
slot so the header pattern (left=backward, right=forward) stays
consistent across columns.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
sortInbox is now async, yields every 10 items, and emits inbox_sort_start
+ inbox_sort_progress via optional hooks. the pipeline route handler
wires those hooks to the existing job events stream and guards against
a second concurrent sort with a 409.
the inbox column swaps its Auto Review button for a live "Sorting N/T"
counter and progress bar while the sort is in flight; the auto-process
toggle hides to give the progress the full subtitle line. the previous
behaviour was a frozen button for the entire duration of the sort.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
sortInbox used to distribute plans by the auto_class and stream_decisions
captured at scan time, which meant toggling an audio_languages entry and
then running "back to inbox" + "auto review" re-queued the item with the
stale decisions. now sortInbox re-runs the analyzer per plan against the
current audio_languages before distributing, matching the user's mental
model that auto review = re-apply the rules.
reanalyze() takes audioLanguages explicitly so callers can pass it in
once and tests can drive it without reaching into the singleton db.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- execute/clear now also resets sorted=0 so cleared items land back in the
inbox where the distributor can re-classify them; previously they got
stranded in Review with auto_class='auto', unreachable by both
"Approve all ready" and "Auto Review"
- pipelinecard: action row moves to the top so Skip/Approve/Back-to-review
sit in the same place regardless of card body height; title row follows;
file info (copy/transcode reasons) gets a dedicated row with the
ready/needs-decision badge pushed to the right
- tests: clearQueue preserves running/completed jobs, only pending plans
flip back to sorted=0 pending
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The scalar subquery I added in 7d30e6c ran one aggregate scan of
media_streams per row. On a real library (33k items / 212k streams)
a single page took 500+ seconds synchronously, blocking the event
loop and timing out every other request — Library AND Pipeline both
stopped loading.
Swap it for a single batched `GROUP_CONCAT ... WHERE item_id IN (?...)`
query over the current page's ids (max 25), then merge back into rows.
v2026.04.15.10
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Per-row audio codec summary (distinct lowercased codecs across an
item's audio streams) via scalar subquery on media_streams, rendered
as "ac3 · aac" in a new monospace Audio column.
v2026.04.15.9
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
/api/review/pipeline no longer ships the review array — it now only
carries queue state + reviewItemsTotal. Review items live behind
/api/review/groups?offset=N&limit=25 which returns complete series
(every pending non-noop episode, bucketed by season) so the UI never
sees a split group.
Lifted enrichWithStreamsAndReasons + PipelineAudioStream to module
scope so both /pipeline (queued column) and /groups (review page)
can share the same enrichment.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
GET /api/settings now returns jellyfin_api_key, radarr_api_key,
sonarr_api_key, mqtt_password as "***" when set (empty string when
unset). Real values only reach the client via an explicit
GET /api/settings/reveal?key=<key> call, wired to an eye icon on
each secret input in the Settings page.
Save endpoints treat an incoming "***" as a sentinel meaning "user
didn't touch this field, keep stored value", so saving without
revealing preserves the existing secret.
Addresses audit finding #3 (settings endpoint leaks secrets).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Queue previously processed a snapshot of pending jobs — anything approved
after Run-all clicked sat idle until the user clicked again. Now, when
the local queue drains, re-poll the DB once for newly-approved jobs
before exiting.
Also swap the looser local parseId (Number.parseInt accepted '42abc')
for the strict shared parseId in server/lib/validate.ts.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
new POST /api/execute/verify-unverified that picks every plan with
status=done + verified=0 and runs handOffToJellyfin sequentially in
the background. each handoff fires the existing plan_update sse so
the done column promotes cards as jellyfin's verdict lands. exported
handOffToJellyfin so the route can reuse the same flow as a fresh job.
done column header shows a 'Verify N' action whenever there are
unverified done plans, alongside the existing 'Clear'. one click and
the user can backfill ✓✓ across every legacy done item without
re-transcoding.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
the ✓✓ write was landing in the db but never reaching the browser.
job_update fires once at job completion (card renders ✓, verified=0),
then handOffToJellyfin takes ~15s to refresh jellyfin + re-analyze +
UPDATE review_plans SET verified=1. no further sse, so the pipeline
page never re-polled and the card stayed at ✓ until the user
navigated away and back.
new plan_update event emitted at the end of handOffToJellyfin. the
pipeline page listens and triggers the same 1s-coalesced reload as
job_update, so the done column promotes ✓ → ✓✓ within a second of
jellyfin's verdict landing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
adds POST /api/review/:id/reopen that flips done or errored plans back
to pending, clears the lingering job row, resets verified=0, and keeps
the prior ffmpeg error summary in the plan's notes so the user has
context for redeciding.
done column cards grow a hover-only '← back to review' button next to
the status badge — works identically for both the ✓/✓✓ and the ✗ rows,
since the server accepts either. also hid the existing queue card's
back-to-review behind the same hover affordance so the two columns
behave consistently and the cards stay visually calm when not hovered.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Abraham Lincoln crashed with exit 234 because the file had 14 dvd_subtitle
streams: our extraction dict only keyed on the long form (dvd_subtitle)
while jellyfin stores the short form (dvdsub), so the lookup fell back
to .srt, ffmpeg picked the srt muxer, and srt can't encode image-based
subs. textbook silent dict miss.
replaced the extension dict with an EXTRACTABLE map that pairs codec →
{ext, codecArg} and explicitly enumerates every codec we can route to a
single-file sidecar. everything else (dvd_subtitle/dvdsub, dvb_subtitle/
dvbsub, unknown codecs) is now skipped at command-build time. the plan
picks up a note like '14 subtitle(s) dropped: dvdsub (eng, est, ind,
kor, jpn, lav, lit, may, chi, chi, tha, vie, rus, ukr) — not extractable
to sidecar' so the user sees exactly what didn't make it.
also added extractErrorSummary in execute.ts: when a job errors, scan
the last 60 stderr lines for fatal keywords (Error:, Conversion failed!,
Unsupported, Invalid argument, Permission denied, No space left, …),
dedupe, prepend the summary to the job's stored output. the review_plan
notes get the same summary — surfaces the real cause next to the plan
instead of burying it under ffmpeg's 200-line banner.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
user reported ad astra got the double checkmark instantly after
transcode — correct, and correct to flag: the post-execute
verifyDesiredState ran ffprobe on the file we had just written, so it
tautologically matched the plan every time. not a second opinion.
replaced the flow with the semantics we actually wanted:
1. refreshItem now returns { refreshed: boolean } — true when jellyfin's
DateLastRefreshed actually advanced within the timeout, false when it
didn't. callers can tell 'jellyfin really re-probed' apart from
'we timed out waiting'.
2. handOffToJellyfin post-job: refresh → (only if refreshed=true) fetch
fresh streams → upsertJellyfinItem(source='webhook'). the rescan SQL
sets verified=1 exactly when the fresh analysis sees is_noop=1, so
✓✓ now means 'jellyfin independently re-probed the file we wrote
and agrees it matches the plan'. if jellyfin sees a drifted layout
the plan flips back to pending so the user notices instead of the
job silently rubber-stamping a bad output.
3. dropped the post-execute ffprobe block. the preflight-skipped branch
no longer self-awards verified=1 either; it now does the same hand-
off so jellyfin's re-probe drives the ✓✓ in that branch too.
refreshItem's other two callers (review /rescan, subtitles /rescan)
ignore the return value — their semantics haven't changed.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
each top-level card now shows a secondary button on hover ('↑ approve
above') that approves every card listed above this one in one
round-trip. uses a new POST /api/review/approve-batch { itemIds } that
ignores non-pending items so stale client state can't 409. series cards
get the same affordance scoped via a named tailwind group so it
doesn't collide with the inner episode cards' own hover state.
fix the horizontal-scroll glitch: long unbreakable audio titles (e.g.
the raw release filename) now line-wrap inside the card via
[overflow-wrap:anywhere] + min-w-0 on the span. previously
break-words was a no-op since there were no whitespace break points
in the release string.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
worked through AUDIT.md. triage:
- finding 2 (subtitle rescan wipes decisions): confirmed. /:id/rescan now
snapshots custom_titles and calls reanalyze() after the stream delete/
insert, mirroring the review rescan flow. exported reanalyze + titleKey
from review.ts so both routes share the logic.
- finding 3 (scan limit accepts NaN/negatives): confirmed. extracted
parseScanLimit into a pure helper, added unit tests covering NaN,
negatives, floats, infinity, numeric strings. invalid input 400s and
releases the scan_running lock.
- finding 4 (parseId lenient): confirmed. tightened the regex to /^\d+$/
so "42abc", "abc42", "+42", "42.0" all return null. rewrote the test
that codified the old lossy behaviour.
- finding 5 (setup_complete set before jellyfin test passes): confirmed.
the /jellyfin endpoint still persists url+key unconditionally, but now
only flips setup_complete=1 on a successful connection test.
- finding 6 (swallowed errors): partial. the mqtt restart and version-
fetch swallows are intentional best-effort with downstream surfaces
(getMqttStatus, UI fallback). only the scan.ts db-update swallow was
a real visibility gap — logs via logError now.
- finding 1 (auth): left as-is. redacting secrets on GET without auth
on POST is security theater; real fix is an auth layer, which is a
design decision not a bugfix. audit removed from the tree.
- lint fail on ffmpeg.test.ts: formatted.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
felixfoertsch