Delete permanently removes a course from disk (confirmed with the full
path first) and cleans up any Hidden/Next Up/Recently Viewed
references to it. Ignore marks a specific pair as confirmed-not-
duplicates so it stops resurfacing in scans, without suppressing
either course's other matches; ignored pairs are listed and reversible
under "Ignored matches," and now travel with backup/restore.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Replace the dashboard-only footer link with a "Files" item in the
persistent bottom tab bar (Home/Notes/Help/Settings) on every page, so
it's reachable the same way everywhere instead of only from the
dashboard.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Fix Sort Unsorted picking a wrong destination on words that are common
clutter across many categories (a creator's name, generic marketing
filler) by ranking matches on a folder's own deliberate name over
incidental sibling-course words, and downweighting borrowed words that
recur across many categories. Add a folder-picker/create-new/rename UI
to override any proposal.
Turn /unsorted into a general File Management page: move Refresh
Library, Bulk Rename, and Manage Library (hide/rename) here from
Settings, add a Move action to Manage Library using the same picker,
and add Duplicate Courses and Clean Up Stale References tools. Add a
footer link to the page from the dashboard.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Scans an Unsorted folder at the library root and proposes destinations
for new courses by matching keywords against the existing category
tree, suggesting new subfolders when nothing matches closely, and
flagging items for manual review when nothing matches at all. Review
and apply happen on a dedicated /unsorted page linked from Settings;
nothing moves until the user confirms.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The 34 kept themes were split 24 light-mode / 18 dark-mode - the light
ones all shared the same washed-out near-white background (L=0.95),
too bright per feedback. Added a force_mood override to build_theme()
and regenerated all 24 as dark instead, same accent character, colored-
dark background rather than near-white.
That conversion collapsed a lot of previously-distinct light backgrounds
into similar dark neutrals, so re-ran the same color-distance dedup pass
across the full set: 8 more turned out to be near-twins once everything
converged to dark (including a 3-way orange-on-dark cluster trimmed to
one). Net: 53 candidates -> 34 kept, all dark/colored, zero white
backgrounds - down from the 42 (24 light/18 dark) shipped last commit.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Sourced from Figma's "53 Unique Website Color Schemes" resource page,
which only shows rendered mockup screenshots (no raw hex data) - extracted
dominant colors via canvas pixel sampling, then a small pipeline
(tools/generate_website_scheme_themes.py) maps each scheme's real colors
into this app's palette shape, with contrast enforced against both the
background and the white text every .btn uses.
Cut 11 of the original 53 after review: 2 for a neon-bright accent (the
saturation cap only bounded lightness at first, so ~28 of 53 accents
landed at pure 100% saturation - fixed at the pipeline level, calibrated
against what the existing hand-picked themes actually use), and 9 for
being near-duplicates of another kept scheme once mapped into actual UI
colors (measured by real color distance, not eyeballed) - including one
pixel-identical pair and a 4-way cluster trimmed to its two most distinct
members.
Settings dropdown is grouped into <optgroup>s (Base, Editor Themes, and
5 categories matching the source page's own curation) rather than one
flat 54-option list.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Auto-generated course thumbnails: find_course_thumbnail() falls back to
grabbing a frame from a course's first video via ffmpeg when there's no
manual cover image, cached to .offlineu_thumbnail.jpg so it only ever
runs once. Folded into the "Precompute" prewarm button, now "Precompute
Lengths & Cover Art".
Library-wide "Remaining" stat: sums every course's already-cached
duration (no ffprobe, just a JSON read) into the dashboard's Library
Stats card, alongside the existing "Watched" figure.
Backup restore: new /api/backup/restore endpoint and a "Choose Backup
File..." control in Settings, with zip-slip and missing-course guards.
Also fixed a gap in the export itself - next_up.json and
outline_config.json weren't being backed up before, so restore wouldn't
have been a true round trip.
Move Surprise Me from a full-width button above the course list to a
dice-icon button in the dashboard header, swapping with the course-name
badge depending on whether a course is loaded.
Update README to cover all of the above.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Settings gets a "Precompute Video Lengths" button that walks the whole
library in a background thread, populating every course's persistent
ffprobe duration cache up front instead of paying that cost the first
time each course card is viewed. Progress polls live while it runs and
picks back up correctly if you navigate away mid-scan.
Separately, the loaded-course lesson tree only ever showed a lesson's
duration once you'd actually played it (from progress.json) - a freshly
opened course had no time info anywhere, including the course header's
"~X remaining" line, which existed but was always empty as a result.
apply_progress_to_tree now falls back to the same ffprobe cache for any
lesson without a known duration yet, so per-lesson lengths and the
remaining-time estimate work from the very first visit. Refactored the
duration-cache read/write into a shared helper so the library browser,
the prewarm button, and this tree view all go through one path.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Add ffmpeg to the Docker image so ffprobe can read each video/audio
file's duration server-side. Results are cached to a small persistent
per-course JSON file keyed by (relative path, size, mtime), so a file
only gets probed once - a container restart or the library scan's
in-memory cache expiring never re-runs ffprobe on unchanged files.
Course cards in the library browser (grid, list, and search) now show
total runtime alongside the media-file count, so you know the time
commitment before opening a course. Durations under a minute are
suppressed rather than showing a noisy "0m".
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Compute a cheap, uncached completion percentage per course (completed
lessons / media files) and surface it as a badge + thin progress bar on
library cards in both list and grid view, and in search results - untouched
courses stay unbadged so the list doesn't get noisy with "0%".
Replace the native alert() dialogs used for bulk-select validation and
course-load errors with a small slide-in toast component that matches the
rest of the UI instead of a jarring browser popup.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Replace card-header emoji with a matching stroke-style SVG icon set that
inherits theme color via currentColor. Drop the accent-stripe border from
every card and keep it only on the Your Courses card, the one primary
action. Merge Continue Watching and Recently Viewed into a single
true-recency list so an old completed item can no longer outrank a newer
in-progress one across two separately-capped lists.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Three visual/structural changes:
- Grid cards without a thumbnail (nearly all real courses in this
library) showed the same generic graduation-cap icon, making the
poster grid hard to visually scan. Replace it with a deterministic
color+initial per course name (same idea as Slack/Google Photos
default avatars), so cards without real cover art still look
distinct from each other.
- Notes/Help/Settings only existed as small footer text links,
meaning a scroll to the bottom of every page to navigate. Add a
persistent bottom tab bar (Home/Notes/Help/Settings, current page
highlighted via a new active_tab context var) to the dashboard,
Notes Hub, Settings, and Help pages. Deliberately left the lesson
page alone - it already has its own fixed-bottom notes panel, and a
second fixed bar there would just eat into an already content-dense
page.
- The no-course dashboard stacked five informational/queue cards
(Library Stats, Next Up, Pick Back Up, Recently Added, Continue
Watching, Recently Viewed) before "Your Courses," so the actual
library browser was the last thing on the page. Move it up to right
after the Surprise Me banner - it was already unconditional, so this
only changes where it renders, not when.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Three quality-of-life additions to the Library browser and course
pages:
- A course page shows a "Resume: <lesson>" button when there's a
last-visited lesson, using Course.last_accessed_path (already
populated by apply_progress_to_tree, just not surfaced before) -
no more scrolling the tree to find where you left off.
- "Surprise Me" picks a random not-yet-fully-watched course and loads
it, for when there's too much library to decide what to watch.
Reuses the already-cached get_all_course_dirs() and the same
per-course completed/total shape _scan_library_activity() already
computes for Stale Courses - no new scanning.
- A select-mode toggle in the Library browser adds checkboxes to every
row/card (list and grid) with a bulk-action bar to hide or queue
several courses/folders at once, via two new endpoints
(/api/hidden-paths/bulk, /api/next-up/bulk) that reuse the existing
single-item set_path_hidden()/set_path_queued() in a loop. Selection
is scoped to the current folder view and clears on navigation.
Also fixes runTranscriptSearch() going silently blank on zero results
instead of showing a "no matches" message, and updates the Help page
to cover all of this.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
There was no caching anywhere in the app, so every request re-walked
the SMB-mounted library from scratch - the dashboard alone triggered
2 duplicate full-library directory walks, 100+ individual
.offlineu_progress.json opens, and ~10-15 full per-course recursive
rglob() scans, none of it shared between requests. Measured ~90s per
dashboard load against the real mounted NAS, with a second immediate
reload taking just as long - proof nothing was being reused.
Add a minimal in-process TTL cache (5 min) and apply it at the actual
hot spots: the shared course-directory listing (get_all_course_dirs,
replacing 6 independent iter_all_courses() walks), the per-course
media-count/thumbnail summary (_course_summary, also now reused by
list_library_directory instead of a third duplicate implementation),
the per-course tree scan (get_course_tree - safe to cache since
progress is always re-applied fresh on top, never baked into the
cached structure), and the transcript-search subtitle index (the
worst offender - previously re-read and lowercased every subtitle
file in the library on every search).
Invalidates immediately on hide/show, rename, and library-path
changes; a "Refresh Library" button on Settings covers files added
directly on the NAS outside the app.
Measured after the fix, same real NAS mount: dashboard ~90s -> 1.3s
warm, Notes Hub ~1.15s warm, Library browser ~3.6s warm.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Three comparison-driven upgrades against typical course/media platforms:
- Lesson page gets a collapsible "Lessons in this section" list so you
can jump between siblings without leaving the page, instead of only
global Prev/Next buttons. find_lesson_in_tree() now also returns the
DirectoryNode that owns the lesson (its .lessons list is the sibling
set) since DirectoryNode has no parent pointer; a new
get_section_lessons() resolves each sibling's progress the same way
view_lesson() already does for the current lesson.
- Course tree section headers show "X/Y watched" instead of a flat item
count, via DynamicCourseParser._calculate_completion_stats (already
computed recursive completion for any node, just never called
per-section) exposed as a Jinja global.
- Library browser gets a list/grid toggle for browsing large category
folders by poster-style thumbnail instead of a 32px-icon list.
Client-side only, persisted via localStorage, defaults to the
existing list view; both folder browsing and search results already
funnel through the same render function so grid mode covers both.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
view_lesson() called update_lesson_progress() with no arguments on
every page load, which defaults completed=False and progress_seconds=0
and silently overwrote whatever was already saved - just opening an
already-watched lesson reset its progress/completed state unless the
client happened to report real values within the next 15 seconds.
Replaced with touch_lesson_accessed(), which only bumps last_accessed.
Also fixes the "Resume from Xs" feature this fed into: it read a value
that was never populated on this route, and even when given a real
one, checked activeMedia.duration synchronously before metadata had
loaded, so it silently never applied. Now waits for loadedmetadata (or
resolves immediately if already available) and defers to an explicit
note-timestamp jump (?t=) when both are present.
Notes Hub gets a live search box filtering by note text, lesson title,
or course name (client-side, no reload), with a "no notes match" state
- useful now that a single category folder can hold dozens of courses'
worth of timestamped notes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
_SECTION_NAME_RE used \b after keywords like "chapter", but \b doesn't
fire between "Chapter" and an immediately-following "_" since
underscore counts as a word character too - so a course whose chapter
folders were named "Chapter_1-Introduction" (no space) failed to match
and got misclassified as a plain folder instead of a course.
Swap \b for a (?![a-z]) negative lookahead, which correctly rejects
only a following letter (e.g. "Chapterhouse") while accepting a digit,
underscore, hyphen, space, or end of string.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
_looks_like_course() treated any folder with 2+ media-holding subfolders
as a single multi-section course, with no check that those subfolders
actually looked like sections. A category folder holding many unrelated
courses (e.g. Claude/, General/) could accidentally satisfy that if
several of its courses happened to keep episodes directly at their own
top level rather than nested under a chapter folder, collapsing the
whole category into one bogus course.
Now every media-holding subfolder must read as a chapter/section label
(Section 1, Module 2, or a numbered chapter like 01-Introduction) for
the parent to count as one course - verified against the real library
mount that ChatGPT/Claude/Copilot/General all correctly show as
folders again, while individually-structured courses still detect
correctly.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Replace the single free-text note per lesson with a list of timestamped
note entries, migrated transparently from the old format. Adds a
floating notes panel on the lesson page with an always-reachable
quick-capture bar (press N to pause and jump in), inline edit/delete,
click-to-seek timestamps, and a live "Saved Xs ago" indicator. Notes
Hub, the study guide export, and the Outline push all updated to
render multiple timestamped notes per lesson.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Six more dashboard features, all built on the per-course progress-file
scanning infrastructure from earlier sessions:
- Notes Hub (/notes): every lesson note across the library in one place,
newest first, reusing the existing /recent/open cross-course jump route.
- Transcript search: opt-in checkbox on the Library search box, searching
inside .srt/.vtt files. Kept cheap by doing a raw substring match as the
filter step and only extracting a snippet for files that actually match.
- Stale course nudges ("Pick Back Up"): courses with progress that haven't
been touched in 14+ days.
- Study guide export: compiles a course's notes into one downloadable
markdown file, section/lesson structure preserved.
- Next Up queue: an ordered, persisted "what to tackle next" list with
up/down reordering and a queue button on every course row.
- Activity heatmap: a 90-day contribution-style calendar in the Library
Stats card.
_scan_library_activity() now does one walk over every course's progress
file per dashboard load; library stats, stale courses, and the heatmap all
derive from that single scan instead of three independent ones.
Also refactored the duplicated lesson-progress-key lookup (used in three
places now) into _resolve_lesson_progress_key(), and flagged a pre-existing
bug found along the way (not fixed here, kept out of scope): subtitle files
never actually attach to a Lesson object, so the video player's caption
track has never had anything to render.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Lesson page's global keyboard shortcuts (space/arrows for play-pause/
seek/volume) fired regardless of focus, so typing a space in the notes
textarea toggled the video instead of typing. Now skipped entirely
whenever a text field/select has focus.
- Outline topics no longer spawn their own top-level collection each time.
A topic is now a document inside one fixed, user-configured collection
(Settings -> Outline Integration -> Default collection), and a lesson's
note becomes a child document nested under its topic - matching how
notes actually get organized in a real Outline instance instead of
cluttering the collections list with one per topic.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Pick a topic on a lesson's note and it pushes to that topic's collection
in a self-hosted Outline instance when you leave the page; notes without a
topic stay local-only. Settings gets a new Outline Integration card
(URL/token, Save, Test Connection).
Credential handling: the API token lives in its own outline_config.json
in DATA_DIR, deliberately kept out of the general settings flow (GET
/api/settings is fetched on every page load by theme.js - no place for a
secret to ride along). GET /api/outline/config only ever returns whether
it's configured, never the token itself.
The topic chooser reflects Outline's live collection list rather than a
locally cached copy, and resolves a newly-typed topic name to a real
collection immediately (find-by-name-or-create) rather than waiting until
the note is pushed. That's not just an optimization: the push itself fires
via navigator.sendBeacon() on pagehide, which can't read a response, so a
page that fires pagehide more than once for the same load (a back/forward-
cache restore, for instance) would otherwise re-send the same "create a
new topic" intent every time and spawn duplicate collections. Verified
live against a local mock Outline server that firing pagehide repeatedly
for the same lesson creates the collection/document once and updates
thereafter.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Five more usability features on top of the search/thumbnails batch:
- Recently Added: dashboard card of courses by folder mtime, separate from
Recently Viewed (watched vs. just showed up on disk).
- Bulk find/replace rename across every course/folder name at once, with a
mandatory preview step before anything touches disk. Refactored the
single-item rename route to share the same validate/apply logic.
- Estimated time remaining on the loaded course's stats card, computed only
from lessons that have actually reported a duration.
- Library-wide stats overview: total courses, lessons tracked, time
watched, daily streak - read from each course's small progress file
rather than re-scanning course contents.
- One-click backup/export zip of settings, hidden-paths, recently-viewed,
and every course's progress/notes.
Also fixes a real performance bug found along the way: search was routing
through list_library_directory, which computes a full recursive file count
and thumbnail lookup for every course at every level regardless of match -
turning a search into an O(every file in the library) scan. Gave search its
own lightweight directory-only walk (iter_all_courses), now shared by
Recently Added and the stats overview too, so the expensive per-course work
only runs for courses that actually match.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
search_library_courses was walking via list_library_directory, which
computes a full recursive file count (rglob) and thumbnail lookup for
every course at every level - turning a search into an O(every file in
every course) scan regardless of how few results actually match. Give
search its own lightweight directory-only walk and defer the expensive
per-course lookups until after a name match is confirmed, so cost now
scales with matches found rather than total library size.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Eight usability additions on top of the mobile redesign: course cover-image
thumbnails, a debounced library search, a Continue Watching section split
from Recently Viewed, per-lesson playback speed control, an installable
PWA manifest/icons, name sort in the library browser, per-lesson notes, and
a bulk mark-course-as-watched action. Also fixes two real bugs found along
the way: lesson_view.html was missing the viewport meta tag (so lesson
pages weren't actually mobile-responsive), and update_lesson_progress
always overwrote the whole progress entry, which would have silently
deleted a saved note on the next routine autosave.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
rmsitzandClaude Sonnet 5