CEP `csInterface.evalScript` callback is broken in Premiere Pro 26.0.x —
nothing called from the panel ever returns, so importFiles deadlocks. Adobe's
path forward is UXP. This is the minimum viable port that restores the
Import Proxy / Import Hi-Res workflow.
Scope (v2.0.0):
- Connect to a Dragonflight server (URL + Bearer token; persisted)
- Asset library (search, refresh, grid with thumbnails)
- Import Proxy via streamed download → Project.importFiles
- Import Hi-Res via presigned S3 URL → Project.importFiles
Layout:
manifest.json UXP v5, host=premierepro, minVersion=26.0.0
index.html Panel shell
styles.css Mirrors web UI dark tokens
src/ui.js DOM helpers, toast, progress, formatting
src/api.js HTTP client (Bearer; manual redirect-follow drops auth
when hopping to a different host per UXP security policy)
src/library.js Asset grid render + selection
src/import-flow.js Streaming download (fs.createWriteStream) +
premierepro.Project.importFiles into rootBin
src/main.js Bootstrap, event wiring
build/pack.mjs Packs into .ccx; installs via UnifiedPluginInstallerAgent
Coexists with services/premiere-plugin/ (CEP) — keeps the CEP panel for any
features that still work there while running v2.0.0 for import. Future v2.x
will add live preview, conform, timeline export, settings.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Writes timestamped pre/post lines to C:/Temp/df-import-log.txt around the importFiles call so we can see whether importFiles hangs (only pre line present) or returns and evalScript callback gets lost (both lines present). Diagnostic only.
app.project.importFiles() can deadlock if a hidden Premiere modal appears (off-screen, behind window, etc) — the evalScript callback never fires and the panel spinner hangs forever.
Two changes:
1) Pass suppressUI=true to all five importFiles call sites (main.js inline IIFE + 4 in premiere.jsx). Premiere proceeds even if it would have prompted (audio sample rate, project link, scale-to-frame, etc).
2) Wrap importFileToPremiereProject in a 60s timeout race so even if importFiles does block, the panel surfaces a real error instead of leaving the spinner stuck.
Bumps to v1.2.2.
downloadFile() uses native https.get which bypasses the window.fetch interceptor that injects Authorization. Same-server URLs (proxy /video) hit requireAuth and 401. Inject the Bearer header manually when url starts with state.serverUrl.
Also add a 15s setTimeout so an unreachable presigned URL (or CEP-Node TLS hiccup on broadcastmgmt.cloud) fails fast with an error instead of hanging the spinner forever.
This is the real cause of the login loop. mam-api sets its session cookie
with Secure=true (production config). express-session refuses to emit a
Secure Set-Cookie unless req.secure is true. With `app.set('trust proxy')`
on, req.secure derives from X-Forwarded-Proto.
web-ui's nginx was unconditionally sending `X-Forwarded-Proto: $scheme`.
Inside the web-ui container nginx listens on port 80, so $scheme is always
"http" — regardless of whether the outer NPM proxy terminated TLS. mam-api
saw http, decided the connection was insecure, and silently dropped the
Set-Cookie from the login response. Login succeeded server-side (session
row landed in PG, last_login_at updated) but the browser never received a
cookie, so the very next /auth/me check came back 401 and AuthGate bounced
to the login screen. Infinite loop.
The previous Connection: "upgrade" → $connection_upgrade fix wasn't wrong
(the hardcode is a real latent bug worth fixing) — it just wasn't the
proximate cause.
Fix: a second `map` directive forwards the outer X-Forwarded-Proto through
when present, falling back to $scheme only when no proxy header exists (so
direct localhost curls still work). Both /api/ and /capture/ now send the
correct value upstream, mam-api sees https, req.secure is true, Set-Cookie
flows through, login works.
Verified by curling the existing direct-to-mam-api path: with X-Forwarded-
Proto: https on the request, Set-Cookie comes back; without it, no
Set-Cookie. That's the exact difference between web-ui-proxied and
direct-to-mam-api in our previous diagnostic curls.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Login was infinite-looping in production. Server side was healthy (sessions
landing in PG, /me returning 200 to a manually-signed cookie) but the
browser never received `Set-Cookie`. Bisected the proxy chain layer by
layer with direct curls on the box:
- mam-api direct (port 47432) → Set-Cookie present
- web-ui nginx (port 47434) → Set-Cookie STRIPPED
- NPM (https://dragonflight.live) → Set-Cookie stripped (because web-ui ate it)
Root cause was this in /api/ and /capture/:
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection "upgrade";
The literal "upgrade" was being sent on every request, not just real
WebSocket negotiations. Nginx then routes the upstream response through
its tunnel/upgrade code path, which doesn't preserve all response headers
the same way — Set-Cookie got silently dropped. mam-api doesn't speak
WebSockets today so it never sent a 101, and the bad pattern went
unnoticed until session-cookie auth shipped.
Fix is the standard conditional pattern: a `map` directive at the top of
default.conf computes $connection_upgrade as "upgrade" only when the
client actually requested Upgrade, otherwise "close". Both location blocks
now send `Connection $connection_upgrade` instead of the hardcoded literal.
WebSocket support on either location continues to work unchanged.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
User reported infinite login loop on dragonflight.live. Root cause: openresty
fronts both http:// and https:// without redirecting, and a user landing on
http:// gets the Set-Cookie response silently dropped — cookies are Secure-only
when TRUST_PROXY=true, and the CORS allowlist refuses the http:// origin.
Result: login appears to succeed, next request has no session cookie, AuthGate
bounces back to login.
Two defensive layers (the openresty box is not in our reach):
- web-ui index.html: tiny inline redirect; if location is http://dragonflight.live,
rewrite to https:// before anything else runs. Bounded to that exact hostname
so local / LAN access on http://172.18.91.x stays as-is.
- mam-api: emit Strict-Transport-Security on HTTPS responses when AUTH_ENABLED=true.
After one successful HTTPS visit, browsers auto-upgrade future http:// requests
on their own — closes the loophole even if someone bypasses the index.html JS.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- requireAuth bearer path now selects api_tokens.bound_hostname and users.role,
populates req.tokenBoundHostname and req.user.role. /cluster/heartbeat can
now authenticate via a bound api_token (issued via POST /auth/tokens with
bound_hostname).
- routes/tokens.js POST accepts bound_hostname; GET returns it so users can
see which tokens are bound.
- Remove /cluster/heartbeat from SERVICE_PATHS so requireAuth runs on it (the
bearer auth handles the gate; the heartbeat handler still enforces the
body.hostname === bound match).
- /auth/me now returns role (final-review I2). Closes the gap where every
signed-in user appeared as 'viewer' in the UI regardless of actual role.
- loadUser SELECTs role for session auth.
- Backend tests still 37/15/0/22 — no test changes needed; existing token
CRUD tests stay passing since bound_hostname is optional.
- On connect success: hide form, show compact connected-bar with hostname
- On disconnect: clear assets, reset buttons, restore form
- Wire disconnect-btn click to disconnectFromServer()
On startup the full form shows. On successful connect the form hides and a
compact connected-bar appears with the server hostname and a Disconnect button.
Migration 023 was fixed in 9dc572b to use '00000000-0000-4000-8000-000000000000'
because 'v' isn't a valid hex digit, but the DEV_USER_ID constant in
middleware/auth.js still referenced the original '...000000000dev'. Every
route that passes DEV_USER_ID as a query parameter (users list, login lookup,
setup-required count) was throwing 22P02 invalid input syntax for type uuid.
The errors were swallowed by Promise.allSettled in the SPA's data load so the
app appeared to work in dev mode, but enabling AUTH_ENABLED=true would have
broken login entirely.
Final-review findings:
- Mount usersRouter at /api/v1/users in addition to /api/v1/auth/users so the
existing SPA Users page works; add PATCH /:id for inline edits (display_name,
role, password).
- Add X-Requested-With: dragonflight-ui to raw XHR/fetch paths that bypass
apiFetch (file uploads, SDK uploads, EDL export) — without it, requireUiHeader
403s before reaching the route.
- Exempt SERVICE_PATHS (/cluster/heartbeat) from requireUiHeader so node-agent
heartbeats keep working when NODE_TOKEN is unset.
- Remove stale auth.js.bak.
Fixes three issues in the authentication system:
C1: Add boot-time warning when AUTH_ENABLED=true but TRUST_PROXY!=true.
Without TRUST_PROXY=true behind nginx, req.ip becomes the proxy IP for all
clients, collapsing per-IP rate limiting into a shared pool. Operators must
explicitly set TRUST_PROXY=true to make per-IP rate limiting effective.
C2: Mount requireUiHeader middleware in test helpers (auth.test.js,
users.test.js, tokens.test.js). The CSRF header validation was not being
exercised in the test suite. Tests now send X-Requested-With: dragonflight-ui
headers that are actually validated by the middleware.
I1: Implement bounded rate-limit Map with MAX_ENTRIES=10000 and LRU eviction.
Unbounded Maps are vulnerable to spray attacks: attackers can force memory
exhaustion by requesting with distinct IPs. Now we evict the oldest entry
(by insertion order) when the map reaches capacity.
Code-review feedback:
- Dummy hash for user-enumeration-defense timing was 63 chars (bcrypt strings
are 60 chars). Worked by accident because bcrypt 5.x is lenient about
trailing chars; a future tightening would silently regress the timing
defense. Replaced with a real pre-computed bcrypt hash.
- last_login_at UPDATE now logs errors instead of silently swallowing them,
matching the pattern in requireAuth for api_tokens.last_used_at.
- Removed dead import of comparePassword from auth.test.js.
Code-review feedback: startsWith('/cluster') was a prefix match that exposed
destructive operator endpoints (POST /containers/:id/restart, DELETE /:id,
GET /devices/blackmagic/*) unauthenticated. Only POST /heartbeat is genuine
node-agent traffic; everything else in cluster.js is operator/UI surface
that should go through requireAuth. Long-term: issue node-agent a bound
api_token and drop the carve-out entirely.
Code-review feedback:
- Hard-fail boot when AUTH_ENABLED=true and SESSION_SECRET is unset, so
express-session can't silently use an in-memory random secret that
invalidates sessions on restart and breaks multi-node clusters.
- CORS rejection now returns cb(null, false) instead of cb(new Error)
so misconfigured origins surface as clean CORS errors in the browser
instead of HTTP 500s. Log a warn line for operator visibility.
- pruneSessionInterval units comment.
Code-review feedback: writing last_seen_at = now before loadUser() lets
the stamp persist if the lookup throws (resave:false still writes when
modified), extending the idle window without confirming the user exists.
Also clarify DEV_USER_ID is a specific placeholder, not a generic sentinel.
Code-review feedback: ON CONFLICT (id) only catches id collisions; a pre-existing
'dev' username would trigger a unique_violation on the username index and roll
back the migration, hard-failing the mam-api boot. Switch to bare ON CONFLICT
DO NOTHING so any unique conflict is no-op-safe.
CEP's embedded Chromium (used by Premiere Pro panels) does not support
oklch() color syntax. All color tokens were rendering as invalid/transparent,
causing the panel to appear unstyled. Converted all oklch() values to their
precise hex/rgba equivalents via OKLab→sRGB math. No design changes.
- Capture screen now polls /cluster/devices/blackmagic/signal every 3s
- Per-port chips show signal state (RECEIVING/CONNECTING/LOST/ERROR/IDLE) with pulsing dot
- BMD SVG card diagram rendered per node card
- Sidebar nav badge on Capture item shows live/total port count (pulsing green dot)
The Dashboard page was rendering as plaintext because all .dash-* CSS
classes (dash-section, dash-onair-*, dash-jobs-*, dash-cluster-*,
dash-statusbar, etc.) were missing. Added them with the full dark-theme
design-system styling matching the rest of the app.
The Cluster page's .stat-row and .stat-card classes were also missing,
causing node statistics (counts, CPU, GPUs, memory) to render unstyled.
Added grid-based stat row and card styles.
- remove requireAuth from all route files
- delete auth.js, tokens.js, users.js routes
- delete auth middleware
- remove session middleware and all auth deps from index.js
- delete login.html and auth-guard.js from web-ui
Scope (locked in via planning Q&A):
- Identity: local accounts only (PG users table) + existing bearer
tokens for headless callers.
- Transport: httpOnly cookie session for browser, Bearer for API.
- RBAC: admin / editor / viewer roles, plus an orthogonal
is_client flag for external (agency, talent, customer) accounts.
- Bootstrap: ADMIN_BOOTSTRAP_USER + ADMIN_BOOTSTRAP_PASSWORD env
seed the first admin on a clean install. Set ADMIN_BOOTSTRAP_RESET
to force-reset the named user (break-glass).
- Rate limit: in-memory, 10 fails per 15min per (IP, username).
- Password policy: \u22658 chars, mixed case, digit, symbol; small
blocklist of common passwords; cannot equal username.
- Self-service: change own display name + password. Everything
else (role, is_client, other-user mgmt) is admin only.
- Audit log: append-only table, indexed by actor + event_type +
created_at, populated by every auth/admin event.
Files added:
- services/mam-api/src/db/migrations/022-auth-rework.sql
users.is_client + last_login_at + failed_attempts; audit_log
table with FK to users (ON DELETE SET NULL).
- services/mam-api/src/middleware/audit.js
Fire-and-forget audit() helper. Caller never awaits, failure
logs but never throws — auditing cannot break the request
that triggered it.
- services/mam-api/src/middleware/passwordPolicy.js
Shared checkPassword(pw, { username }) used by setup, user
create/update, and self-service password change.
- services/mam-api/src/tasks/bootstrapAdmin.js
Runs after migrations. No-ops unless ADMIN_BOOTSTRAP_USER +
ADMIN_BOOTSTRAP_PASSWORD are set AND (users table empty OR
ADMIN_BOOTSTRAP_RESET=true).
- services/mam-api/src/routes/audit.js
Admin-only GET /audit (paginated, filter by event_type /
actor / target / date) and GET /audit/event-types.
- services/web-ui/public/modal-account-settings.jsx
Profile + Password tabs. Triggered by sidebar user button.
Files rewritten:
- services/mam-api/src/routes/auth.js
- POST /login: regenerate(), no manual save(); audit success/
fail/lockout; updates last_login_at + failed_attempts.
- POST /logout: destroys session, audits logout.
- GET /me: returns is_client + last_login_at. Synthetic admin
when AUTH_ENABLED=false.
- GET /setup-status: drives login.html UI state.
- POST /setup: blocked once any user exists; password policy.
- POST /password: self-service. Requires current pw, runs
policy, audits, invalidates other sessions implicitly via
users.js if changed by admin.
- PATCH /me: self-service display_name update.
- services/mam-api/src/routes/users.js
- is_client field in create/update/list/get.
- Guardrails: cannot delete or demote last admin, cannot
delete self, admins cannot be flagged is_client.
- Password change invalidates all sessions for that user
(DELETE FROM sessions WHERE sess->>'userId' = id).
- Audit on every mutation.
- Password policy enforced.
- services/mam-api/src/middleware/auth.js
- requireAuth now exposes req.user.is_client.
- New requireRole(["admin","editor"], { rejectClients: true })
helper. Applied to cluster, sdk, capture routes (infra).
- Synthetic user when AUTH_ENABLED=false has is_client=false.
- services/mam-api/src/index.js
- Loads bootstrap admin after migrations.
- Wires /api/v1/audit.
- Cleans up an earlier comment block.
- services/web-ui/public/login.html
- Password hint added next to setup-mode password field.
- services/web-ui/public/shell.jsx
- Sidebar user footer is a button that opens AccountSettings.
- CLIENT badge next to role when is_client=true.
- Nav filters: clients lose ingest tree + jobs + editor;
viewers lose ingest + editor; only admins see the Admin
section. Power button hidden when synthetic user.
- services/web-ui/public/screens-admin.jsx
- Users table: new Client column with inline toggle.
- InviteUserModal: Client checkbox + password hint, gated
off when role=admin.
- Last login column replaces Created in primary view.
- CSV export includes client + last_login.
- services/web-ui/public/data.jsx
- ZAMPP_DATA.ME carries is_client + display_name.
- services/web-ui/public/index.html
- Loads dist/modal-account-settings.js.
- services/web-ui/public/styles-rest.css
- .user-row grid widened to 6 columns.
- docker-compose.yml
- Plumbs SESSION_COOKIE_SECURE + ADMIN_BOOTSTRAP_* env vars.
Deploy:
cd /opt/wild-dragon
git pull origin main
# In .env:
# AUTH_ENABLED=true
# SESSION_SECRET=<openssl rand -hex 48>
# ADMIN_BOOTSTRAP_USER=admin
# ADMIN_BOOTSTRAP_PASSWORD=<strong>
docker compose build mam-api web-ui
docker compose up -d --force-recreate --no-deps mam-api web-ui
Auth work is parked until after ship. While AUTH_ENABLED=false:
- login.html now auto-redirects to / on load (no one should ever see
the login screen while auth is off; it was confusing).
- sidebar power button is hidden entirely when /auth/me returns a
synthetic user, so there's no broken-feeling no-op control.
- Removed connect-pg-simple createTableIfMissing flag in case
v9.0.1's handling of that option was responsible for the recent
boot 502 (the schema is created by migration 021 anyway).
The /auth/login + session.regenerate() + cookie fix from c34a721
stays in place — when we re-enable auth it'll work end-to-end. The
sessions table from migration 021 stays. Operator action to restore
auth later: set AUTH_ENABLED=true + SESSION_SECRET=<random> in the
mam-api environment and restart.
Login was returning 200 + correct user JSON + writing a row to the
sessions table, but emitting zero Set-Cookie headers. Root cause:
session.regenerate() → set fields → session.save() → res.json()
Calling session.save() manually writes the store but bypasses
express-session's res.end() hook, which is the only path that adds
the Set-Cookie header to the response. The cookie was never sent to
the browser even though the session existed server-side — hence the
redirect loop.
Fix: remove the manual save(). Set the session fields and call
res.json() directly inside regenerate()'s callback; express-session
handles store write + Set-Cookie automatically on res.end().
The redirect loop after successful login was almost certainly the
`sessions` table never being created. `schema.sql` defines it but
only runs on first-init via the postgres entrypoint; instances
bootstrapped via mam-api's own migration loop never got the table.
express-session's `req.session.save()` then failed silently and the
cookie pointed at a sid that wasn't in the store — every subsequent
request looked like a brand-new visitor.
- New migration 021-ensure-sessions-table.sql (idempotent).
- connect-pg-simple now configured with `createTableIfMissing: true`
as belt-and-braces.
- `POST /auth/login` now explicitly waits for session.save() and
surfaces both regenerate() and save() errors instead of treating
them as 'success'. Logs sid + req.secure + req.protocol so we can
confirm trust-proxy is doing the right thing behind NPM.
Three concrete issues kept the login flow broken on dragonflight.live:
1. mam-api trusted no proxy headers, so behind nginx/Cloudflare the
session cookie's `secure` flag and the rate-limiter's IP keying
both saw the wrong values. Now sets `app.set('trust proxy', 1)`.
2. Session config was tied to NODE_ENV and lacked sameSite/name. Now:
- SESSION_COOKIE_SECURE env (default: true when AUTH_ENABLED) so a
site behind HTTPS gets Secure cookies regardless of NODE_ENV.
- `sameSite: 'lax'` for predictable post-login redirects.
- Renamed to `df.sid` so it's obvious in DevTools.
- `rolling: true` extends the 7-day TTL on active use.
- SESSION_SECRET is now required when AUTH_ENABLED=true; the
server refuses to start with a dev default in prod.
3. login.html silently showed the sign-in panel even when no users
exist or auth is off:
- New GET /auth/setup-status reports {needs_setup, user_count,
auth_enabled}.
- login.html calls it on load and auto-flips into setup mode when
needs_setup is true, or shows an explicit "auth is off" flash
when auth_enabled is false (the previous symptom: logout button
did nothing because /auth/me returned a synthetic admin no matter
what).
- Added a `.flash.info` style for the new neutral notice.
4. Sidebar logout used to call /auth/logout then `window.location
.reload()`. With auth off that reload landed back on the synthetic-
admin app and looked like nothing happened. It now redirects to
/login.html in all states so the operator sees feedback (and the
server-side messaging about auth being off) instead of a no-op.
Deploy notes for zampp1:
- Set AUTH_ENABLED=true and a random SESSION_SECRET in the
mam-api environment (e.g. /opt/wild-dragon/.env).
- Restart mam-api.
- First load of /login.html will auto-route to the setup form so
you can create the first admin.
RustFS returns empty bodies for ranged GETs whose start offset is past
~5.9 MB on single-file proxy MP4s. HEAD reports correct size, full GET
(`bytes=0-`) works, but `bytes=8179166-` comes back 206 + correct
Content-Range header with zero bytes. Confirmed via direct S3 probe
against broadcastmgmt.cloud/dragonmam (see scratch tests).
Workaround in mam-api `GET /api/v1/assets/:id/video` until the proxy
worker emits HLS (planned v1.2.1):
- HEAD the object first to learn total size (also gives ETag /
Last-Modified for conditional requests).
- No-Range / unparseable-Range / pre-EOF requests \u2192 plain pipe.
- Parsed `bytes=N-M` requests below RUSTFS_RANGE_SAFE_START
(default 5_500_000) \u2192 direct ranged GET, RustFS handles fine.
- Anything reaching into the broken zone \u2192 stream from offset 0,
drop bytes below start, stop at end. Memory stays flat; extra
bandwidth = (end+1 - requested-size) per seek.
- Genuinely out-of-range \u2192 416 with Cache-Control: no-store so the
browser doesn't poison its cache.
Also stashes (not yet wired up) the HLS pieces we'll need for the
follow-up: `segmentToHls` ffmpeg helper + `uploadDirectoryToS3`
worker s3 helper. Harmless additions; not referenced by any code path
yet.
Confirmed against the affected asset (a72aaa03-...): bytes=0-100k +
50% +100k native pass-through; 70% +100k and near-EOF previously hung
the browser, now stream correctly via the stitched path.
Refs #143.
Frontend / UX / a11y
- Sidebar collapse/expand toggle with localStorage persistence (#142)
- Settings sections wrap inputs in <form> with Enter-to-submit + native
validation; password autocomplete=new-password (#141, #138)
- Asset thumbnails get descriptive alt text (#140)
- Production deploy now precompiles JSX via esbuild and loads the
production React UMD instead of dev builds + in-browser Babel (#139,
#122)
- Search wrapper gets role=search; global search input gets aria-label,
role=combobox, aria-controls/aria-expanded/aria-activedescendant
wiring (#137, #135)
- Dashboard and Library no longer share the same nav icon (#136)
- Sidebar collapses off-canvas with a topbar menu button below 768 px;
mobile default is collapsed (#134)
- --text-3 bumped to #8B92A0 for WCAG AA contrast on --bg-0 (#133)
- Schedule and Library routes were rendering empty inside the .main
flex container — switched to flex:1 + min-height:0 (#131, #132,
editor + asset detail get the same fix)
- Jobs nav badge now polls /jobs?status=active every 10 s and reflects
the live count (#130, #113)
- aria-label sweep on every icon-only button (#126)
- Premiere panel release list moved to window.PREMIERE_RELEASES in
data.jsx; Editor + Settings read from the same source (#125)
- Typo setPgMclips → setPgmClips (#124)
- Stray console.error / console.warn calls gated behind
window.DF_LOG.{warn,error} (#123)
- Hardcoded /api/v1 paths route through window.ZAMPP_API_PREFIX (#115)
- Schedule rows no longer crash on null recorder_id (#117)
- EditorKeyboard guards against document.activeElement === null (#116)
- Unmount-safe timers for PasswordResetModal, Containers, Editor (#111)
- Player seek clamps below totalMs, server-side range clamping +
uncached 416 on EOF, client-side EOF-stall watchdog (#143)
- Duration badge overlap fix on narrow asset cards (#52)
Backend / security / reliability
- GET /recorders fixed N+1: single LATERAL JOIN for live_asset_id;
Docker inspects bounded to actually-recording rows (#121)
- Upload disk-storage (multer.diskStorage) streams parts to S3 instead
of buffering 500 MB in RAM (#120)
- /assets list clamps limit to MAX_LIMIT=500 to prevent OOM (#119)
- SDK upload archive listing + post-extract sanitize block zip-slip /
tar-slip and symlink escapes (#118)
- Migrations track applied state in schema_migrations, run in a
transaction, and exit non-zero on failure (#107)
- node-agent BMD_COUNT override uses BMD_DEVICE_PREFIX; filesystem
detection wins (#109, #127)
- GPU_COUNT override now merges with nvidia-smi enrichment (#108)
- /cluster/heartbeat requires a node-bound token or admin user;
tokens carry bound_hostname (#106)
- /recorders/:id/start error responses no longer echo the Docker
create payload — env vars stay out of client responses (#105)
- /recorders/probe restricts schemes (srt/rtmp/rtsp/udp/rtp), blocks
private + loopback hosts for non-admins, denies common service
ports (#104)
- Scheduler tick guarded by a Postgres advisory lock; pending/running
rows claimed via UPDATE...RETURNING + FOR UPDATE SKIP LOCKED to
survive multi-node deploys (#103)
- UUID validateUuid('id') param middleware on every /:id route (#102)
- Error handler scrubs Postgres error messages and 5xx detail (#101)
- Graceful SIGTERM/SIGINT shutdown — stops scheduler, drains the HTTP
server, ends the pool, 25 s force-exit watchdog (#100)
- AMPP sync moved from fire-and-forget to a persisted retry queue
(ampp_sync_status / attempts / next_attempt_at + scheduler retry
loop with exponential backoff) (#77)
Migrations
- 019: api_tokens.bound_hostname (#106)
- 020: assets.ampp_sync_status + retry bookkeeping (#77)
Other
- Defer #92 Growing-files per-upload toggle, #80 Audio tab, #57
Dashboard redesign, #56 Editor SPA polish phase 3, #114 S3
migration tool to v1.3
The BMD card SVG renderer (window.BMDCards) was created in an earlier
session but never wired into index.html, so the new video-presence
indicator from a44d8bd was silently bailing at the !window.BMDCards
guard. Loading it alongside the other helpers in /js/.
Adds per-port video signal state to the admin Cluster panel:
- New GET /cluster/devices/blackmagic/signal endpoint joins recorders by
node_id+device_index and queries each active capture container's
/capture/status (local: http://recorder-<id>:3001, remote: api_url/
sidecar/<container_id>/status). Returns receiving/connecting/lost/
error/idle/no-recorder per port plus framesReceived and currentFps.
- bmd-card.js render() now accepts portSignals (Map or object) and
overlays a colored dot on each BNC connector with pulse animation
for receiving/connecting states.
- screens-admin.jsx Cluster panel polls the new endpoint every 5s,
feeds the signal map into both the port chips (now show
RECEIVING/CONNECTING/LOST + fps) and the BMD SVG card diagram
rendered below them via a new BmdCardPanel component.
- styles-fixes.css adds bmd-card-* styles for the SVG diagram and
bmd-port-signal --pulse animation.
mam-api /video endpoint:
- S3 InvalidRange (httpStatusCode 416) was being caught and returned as 500
via next(err), which the video element treats as a fatal load error and
freezes the player. Now we catch the specific 416 case, do a no-range
HEAD-equivalent to learn the real file size, and return proper 416 with
Content-Range: bytes */<total> so the browser can recover.
screens-asset.jsx — player health + buffer visualization:
- New states: playerState ('idle'|'loading'|'playing'|'paused'|'seeking'|
'waiting'|'stalled'|'error'), playerError, buffered (array of {start,end}
ms from HTMLMediaElement.buffered), stallStart, stallElapsedMs
- Wired video element events: onProgress, onWaiting, onStalled, onPlaying,
onCanPlay, onCanPlayThrough, onSeeking, onSeeked, onError
- onError captures MediaError code+message into a console.error and the
on-screen badge so freeze causes are now visible
- Status badge overlay (top-right of player): shows SEEKING / BUFFERING /
STALLED / ERROR + elapsed seconds since the stall began
- PlaybackBar renders buffered ranges as translucent grey segments so you
can see what the browser has loaded vs. what's still pending — makes
seek-related freezes immediately obvious
Replaced sync execFileSync('docker') approach (no docker CLI in container)
with async Docker socket HTTP API calls:
- POST /containers/create with nvidia runtime + DeviceRequests
- POST /containers/:id/start
- Poll inspect until not running
- GET /containers/:id/logs, strip 8-byte frame headers, parse csv
probeGpusViaSmi() runs once at startup before the first heartbeat.
Result cached in _gpuCache; detectHardware() reads cache on every heartbeat.
Falls back to /dev/nvidia* scan if probe fails or runtime unavailable.
nsenter approach failed (requires SYS_ADMIN in container).
nvidia-smi bind-mount failed (Alpine vs Ubuntu glibc incompatibility).
Working solution: spawn 'docker run --rm --gpus all ubuntu:22.04 nvidia-smi'
via the Docker socket. The NVIDIA Container Runtime injects nvidia-smi and
driver libs into any container with --gpus all, regardless of the base image.
ubuntu:22.04 is already cached on GPU nodes.
Result: GPU reported with name, memory_mb, driver_version — shows as BOUND
in the cluster UI.
nvidia-smi bind-mount failed due to Alpine vs Ubuntu glibc incompatibility.
Fix: nsenter --mount=/proc/1/ns/mnt -- nvidia-smi runs in the host's mount
namespace where glibc and all NVIDIA driver libs are present.
Requires pid: host in docker-compose.worker.yml (already has network: host).
nsenter is provided by util-linux in Alpine — already in the image.
Falls back to direct nvidia-smi call (for glibc-based containers), then
to /dev/nvidia* file scan if all attempts fail.
index.js:
- detectGpusViaSmi(): runs nvidia-smi --query-gpu=index,name,memory.total,
driver_version and parses the output into structured GPU objects with
name, memory_mb, driver, device — the same fields the cluster UI uses
to determine BOUND status
- Falls back to /dev/nvidia* file scan if nvidia-smi isn't available
docker-compose.worker.yml:
- Bind-mount /usr/bin/nvidia-smi and libnvidia-ml.so.1 from host into
node-agent container (read-only). These are the minimum binaries needed
for nvidia-smi to execute inside the container.
- Mounts are optional — Docker ignores them silently if paths don't exist
(e.g. on nodes without NVIDIA hardware)
_normalizeNode:
- Maps cpu_usage, mem_used_mb/mem_total_mb from actual API fields
- Reads capabilities.gpus: name, memory_mb, device, bound status
(bound = nvidia-smi confirmed driver, detected by name+memory_mb)
- Reads capabilities.blackmagic + blackmagic_model: model, port count,
device paths
Node detail panel:
- GPUs: name, VRAM, device path, BOUND/UNBOUND badge (green if driver active)
- Capture cards: model name, port count badge, per-port device name
with online/offline color coding
Stat row: adds Capture ports total count card
Topology SVG: shows GPU count and BMD port count under each node label
Fix: removeNode uses node.dbId (UUID) not node.id (hostname)
executor.js:
- transcodeVideo() now accepts videoMinRate, videoMaxRate, videoBufSize
- When set, passes -minrate/-maxrate/-bufsize to FFmpeg for ABR/VBR mode
- libx264 operates with per-scene quality variation within the envelope
proxy.js:
- Target average: 750k (gpu_bitrate_mbps=0.75)
- Min: 375k (50% of target), Max: 998k (~133%), Buffer: 2× max
- Gives effective range of ~500k-1M depending on scene complexity
- Log now shows VBR min-max-avg
- GPU fallback also passes VBR params
- Default videoBitrate changed from 10M to 750k in executor.js
S3 at broadcastmgmt.cloud (RustFS/openresty) returns 403 on range
requests that include an Origin header on presigned URLs. The HMAC
signature only covers 'host' in X-Amz-SignedHeaders, so the browser's
cross-origin Origin header breaks signature validation.
Reverted: /stream and /video no longer redirect to signed S3 URLs.
Fixed: /video now pipes through Node with:
Cache-Control: private, max-age=3600
ETag and Last-Modified forwarded from S3
This means the browser caches video segments for 1h. On seek the
browser checks its cache first — only uncached byte ranges hit the
server. Combined with the 1.5Mbps proxy (was 4Mbps), seeks should
be responsive for clips under ~10 minutes.
- GET /assets/:id/stream now returns a signed S3 URL directly (4h TTL)
instead of pointing to the /video pipe endpoint. Browser streams
directly from S3 — no Node.js bottleneck, S3 handles range requests
natively for smooth seeking.
- GET /assets/:id/video now redirects (302) to a signed S3 URL.
Belt-and-suspenders: any code still calling /video gets redirected.
- proxy.js: default bitrate changed from 10Mbps to 1.5Mbps, audio
default from 192kbps to 128kbps. DB settings already updated to
1.5Mbps. Cuts proxy file size ~6x for the same quality content.
Existing proxies need re-generation at new bitrate.
The badge initially showed '0' before any poll completed. Toggling
display via JS expects an initial display:none so the badge does not
flash in the tab nav on first connect.
Root causes found:
1. Scheduler crashing every 15s: assets table has no error_message column.
Fix: remove error_message from UPDATE in scheduler.js (#66 regression).
2. Clip freezing: client-side filmstrip seek loop runs on main thread,
seeks same proxy the player is streaming → both stall → freeze.
Fix: replace browser seek loop entirely with server-side FFmpeg worker.
3. No dedicated filmstrip worker: filmstrip was never pre-built server-side.
Changes:
- services/mam-api/src/db/migrations/018-add-filmstrip-s3-key.sql
Add filmstrip_s3_key TEXT column to assets table
- services/worker/src/workers/filmstrip.js (new)
BullMQ worker: downloads proxy, runs FFmpeg fps filter to extract
28 evenly-spaced JPEG frames, base64-encodes them, uploads JSON
array to S3 at filmstrips/<assetId>.json, stores key in DB
- services/worker/src/workers/thumbnail.js
Queue filmstrip job automatically after thumbnail completes
- services/worker/src/index.js
Register filmstrip worker (concurrency=2), export filmstripQueue
singleton, close it on SIGTERM
- services/mam-api/src/routes/assets.js
- filmstripQueue added
- POST /reprocess?type=filmstrip now supported
- GET /:id/filmstrip returns signed S3 URL for JSON frames
- services/mam-api/src/routes/jobs.js
filmstrip queue visible in Jobs UI
- services/web-ui/public/screens-asset.jsx
Replace browser seek loop with fetch of /assets/:id/filmstrip
→ fetch S3 JSON → render frames. Zero browser-side video seeking.
Right-click and Files tab re-generate via API endpoint.
- FilesTab: fetch /assets/:id/thumbnail (returns signed S3 URL JSON),
display the resolved URL in <img> instead of pointing directly at the
endpoint which returns JSON not image bytes
- Transcoding: settings updated on ZAMPP1 to gpu_transcode_enabled=false,
codec=libx264 — NVENC not available in worker container (no GPU passthrough)
The proxy worker already has a CPU fallback but this prevents the
unnecessary failed GPU attempt on every job
Two bugs:
1. Frame 0 sets currentTime=0 but probe starts at t=0 after onloadedmetadata,
so 'seeked' never fires (no position change). Promise hangs until the 15s
global timeout kills the whole build. Fix: when currentTime is already at
target (within 0.05s), call done() immediately without waiting for seeked.
2. Seeks into unbuffered regions of large MP4s can stall indefinitely.
Fix: 3s per-frame timeout captures the current decoded frame and moves on,
so a slow/stalled seek doesn't block the remaining 27 frames.
Filmstrip:
- Right-click on the filmstrip opens a context menu with
'Re-generate filmstrip' and 'Re-generate proxy'
- filmstripKey state forces the build effect to re-run on demand
without waiting for a streamUrl/totalMs change
- Context menu dismisses on click, contextmenu, and scroll
Files tab (replaces empty Versions tab):
- Proxy: status badge, S3 key path, inline video preview, re-generate button
- Hi-res master: status badge and S3 key path
- Thumbnail: status badge, S3 key path, inline thumbnail image, re-generate button
- Filmstrip: status badge, frame count, scrollable strip of first 14 frames,
re-generate button (disabled while building)
The /video endpoint requires session auth (requireAuth middleware).
crossOrigin='anonymous' strips cookies from the request → 401 → video
never loads → 15s timeout → filmstrip stays empty for all clips.
Same-origin video does not need crossOrigin for canvas drawImage — the
taint restriction only applies to cross-origin resources.
- Editor: overlay Coming Soon screen over NLE timeline (code preserved,
bumper sits at z-index 100 with backdrop blur). Links to download
ZXP and Windows installer directly from the bumper.
- Settings → Capture SDKs: new Premiere Panel section lists v1.0.0
and v1.0.1 with ZXP + Windows Installer download buttons.
Both releases embedded as static files in web-ui under /downloads/.
- nginx: /downloads/ location serves files as Content-Disposition
attachment with 24h cache.
Files added:
services/web-ui/public/downloads/dragonflight-premiere-panel-1.0.0.zxp
services/web-ui/public/downloads/dragonflight-premiere-panel-1.0.0-windows-setup.exe
services/web-ui/public/downloads/dragonflight-premiere-panel-1.0.1.zxp
services/web-ui/public/downloads/dragonflight-premiere-panel-1.0.1-windows-setup.exe
Bug fixes:
- #91: dockerApi() 10s socket timeout (Docker daemon hang)
- #77: await syncToAmpp() with .catch() — no longer fire-and-forget
- #75: migration 016 — add 'proxy','import' to job_type enum; add 'completed' to job_status
- #73: BullMQ orphan job cleanup on hard asset delete
- #70: batch-trim jobs table gets expires_at; trim-status auto-expires stale rows
- #66: scheduler tick marks stale live assets (>2h) as error
- #63: migration 017 — partial unique index prevents concurrent live asset overwrite
- #61: recorders.js uses getS3Bucket() not stale process.env.S3_BUCKET
- #60: already fixed (copy nulls proxy/thumbnail keys, requeues proxy)
- #40: already fixed (All projects clears openProject)
- #64: already fixed (sourceType/needsProxy handled)
- #90: GET /jobs now includes DB jobs table (trim jobs visible in UI)
- #74: nginx Content-Type header preserved; multer 500MB file size limit
- #68: GET /upload returns in-progress ingesting assets
- #58: /stream and /video endpoints fall back to original file for all video types
- #55: recorder poll .catch() logs auth errors cleanly; redirect stops interval
- #52: thumb-status and thumb-duration moved inside position:relative wrapper
- #50: ProjectCard gets onContextMenu handler with rename/delete menu
- #49: project context menu dismisses on contextmenu + scroll events
Features:
- #93: POST /assets/:id/reprocess?type=proxy|thumbnail — force re-queue any asset
Asset ⋯ menu now shows 'Re-generate proxy' and 'Re-generate thumbnail' buttons
UI:
- Logo: brightness(0) invert(1) filter applied consistently in sidebar, launcher,
and login — white logo pops on dark UI; inline style removed from login.html
track?.@_currentExplodedTrackIndex is invalid JS syntax — @ is not a
valid identifier character. Replaced with track?.['@_currentExplodedTrackIndex']
so the worker process no longer crashes on startup.
Four critical fixes:
- Remove overflow:hidden on tlRef so Timeline.init's scroll survives re-renders
- Don't call _renderClips() inside mousedown (was destroying event target mid-drag)
- Use refs for undo history to eliminate stale closure in onClipsChanged callback
- Change .tl-clip-area overflow:hidden to overflow:visible so pointer events reach clip edges
Dispatch df:bins-changed custom event from onBinDrop and
AssetContextMenu.moveToBin so the bin rail counts update
immediately after moving an asset into a bin.
RenameProjectModal is exported to window from screens-projects.jsx,
so Library screen must reference it via window object and use
React.createElement instead of JSX syntax.
- Fix 500 error when creating bins: missing updated_at column on bins table
(migration 013 adds the column, schema.sql updated)
- Add drag-and-drop support for moving asset cards/list rows onto bin rail items
with visual droppable highlight
- Add right-click context menu on project rail items (Rename/Delete)
- Expose RenameProjectModal via window so Library screen can reuse it
- Bins context menu already existed — was hidden by the 500 error
Add 'trim' to job_type enum, create temp_segments table with
expiry/job/asset indexes, and add conform_source_sequence_id
to assets for lineage tracking.
Closes#33
End-to-end debugging against a live Premiere Pro 2025 + auth-enabled mam-api
surfaced four real bugs that made v1.0.0 install cleanly but never load,
plus the missing auth flow. All four are fixed and the panel is verified
connected (status dot green, Reconnect button shown, project list populated).
- manifest.xml: a comment in the <Resources> block contained "--" (inside
"--enable-nodejs"/"--mixed-context"), which is illegal per the XML spec.
CEP 12's strict parser logged
ERROR XPATH Double hyphen within comment
and skipped the panel entirely. Comment rewritten without double hyphens.
- manifest.xml: lacked the Version="X.Y" attribute on <ExtensionManifest>
and used a non-standard AbstractionLayers/empty <ExtensionList/>
structure. CEP rejected it with
Unsupported Manifest version ''
Manifest rewritten to the standard CSXS 7.0 schema (ExtensionList +
DispatchInfoList + RequiredRuntimeList), matching the working AMPP
panel template.
- main.js: re-declared `const csInterface = new CSInterface()` at top
level even though CSInterface.js already declared the same binding.
CEP 12 shares script-realm lexical scope across <script> tags, so the
second const threw
Identifier 'csInterface' has already been declared
The throw fired before setupEventListeners(), so the Connect button's
click handler was never attached. This is the root cause of the
original "clicking Connect does nothing" symptom; everything else was
secondary. Removed the duplicate declaration; main.js now uses the
binding from CSInterface.js.
- No auth support against AUTH_ENABLED=true servers. mam-api supports
Bearer tokens (POST /api/v1/tokens), so added:
• API token input field (password-masked) next to Server URL
• localStorage persistence on every keystroke
• window.fetch monkey-patch that injects
Authorization: Bearer <token>
on every request whose URL starts with the configured server.
Signed S3 download URLs are NOT touched.
Drive-by fixes that came out of the same debugging pass:
- Server URL input listener was 'change' (fires on blur); switched to
'input' so typing-then-clicking-Connect immediately commits.
- restoreSettings() now strips trailing slashes from the stored URL so
older saved values like 'http://host/' stop producing //api/v1 404s.
- CSS selector `input[type="text"].server-url` didn't match the new
password input → the token field was unstyled and effectively invisible.
Generalized to `input.server-url`; restructured the connection bar into
`.connection-controls--stacked` (flex column) of two `.server-input-row`
rows so two input fields fit cleanly.
- Build scripts now parse ExtensionBundleVersion from both element form
(<ExtensionBundleVersion>X</...>) and attribute form
(ExtensionBundleVersion="X"), since the manifest rewrite switched
schemas.
Version bumped 1.0.0 → 1.0.1. New artifacts committed at
services/premiere-plugin/build/releases/v1.0.1/ (.exe 2 MB, .zxp 35 KB).
v1.0.0 left in place so editors who downloaded it can verify they're on
the broken version.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
DELETE /jobs/:id was throwing "404 not found" when the operator tried to
cancel a running job. BullMQ refuses job.remove() while a job is in the
active state; the route caught that error and fell through to the
404 branch, which was misleading because the job actually exists — the
queue was just refusing to drop it from under the worker.
Fix:
- Detect 'active' state explicitly and call moveToFailed(err, '0', false)
first. Token '0' bypasses the per-worker lock check (the operator-side
cancel doesn't hold the worker lock). That transitions active -> failed
and frees the queue's concurrency slot.
- If moveToFailed itself fails (lock owned by a live worker), fall back
to job.discard() so at least the result is thrown away.
- If remove() then fails (stalled, broken state), drop the job's Redis
key directly via queue.client. Last-resort obliteration.
- Stop swallowing getJob() errors — if Redis is sad, surface it via
next(err) instead of returning a misleading 404.
- Return { cancelled: true } when the job was active, so the client
can show "Cancelled" rather than "Removed" in any future toast.
While here: thumbnail jobs now run with concurrency 4 by default
(proxy 2, conform 1, import 1 unchanged). Every queue defaulted to
concurrency 1 before, so a single stalled job blocked the entire queue.
All three are overridable via PROXY_CONCURRENCY / THUMBNAIL_CONCURRENCY
/ CONFORM_CONCURRENCY env vars for nodes with more headroom.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Jobs page only exposed a delete button for queued + done jobs, so a
stalled-active job (worker died holding a BullMQ concurrency slot) had
no way out from the UI. Operators were watching the queue back up
behind a single stuck thumbnail job with no kill switch.
- Running jobs now show a "Cancel" button (red text). Confirm copy
spells out that the worker may run a few seconds longer in the
background but the queue slot frees up immediately.
- Failed jobs now show the X icon for delete in addition to the
existing Retry button.
- Both routes hit the same DELETE /jobs/:id endpoint; BullMQ's
job.remove() works on any state including stalled-active.
- handleDelete takes an optional mode ('cancel' | 'delete') only to
customise the confirm prompt and error toast wording.
Right-aligned the action cell so the Retry/Cancel/Delete buttons sit
flush right like the rest of the table's actions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Before this commit /public had two parallel UIs: the React SPA (index.html
+ screens-*.jsx) and a stack of pre-SPA standalone pages (home.html,
recorders.html, jobs.html, ...). The SPA replaces every standalone page,
nothing in the .jsx tree links to them, and the only outside references
were login.html redirecting to home.html and the nginx fallback pointing
at home.html.
Delete 16 standalone pages (~9.2k lines of dead markup, ~430KB on disk):
_primitives-smoke.html api-tokens.html capture.html cluster.html
containers.html edit.html editor.html home.html
jobs.html player.html projects.html recorders.html
settings.html tokens.html upload.html users.html
Keep:
index.html — the React SPA shell
login.html — the sign-in / setup screen
Wire the redirects to the SPA:
- login.html post-signin: home.html -> /
- nginx try_files fallback: /home.html -> /index.html
After this, sign-in lands the operator on the real React app instead of
the stale 2025-era home page. The Editor screen continues to embed the
separate editor service via the /editor/ nginx proxy (unaffected).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
apiFetch now redirects to /login.html when the server returns 401, so
flipping AUTH_ENABLED=true on mam-api gives the user the login screen
instead of a half-loaded app that silently failed to fetch /auth/me.
While AUTH_ENABLED=false the server's /auth/me still returns a synthetic
200 user, so this branch is dormant — safe to deploy ahead of the env
flip on the server. After the flip the operator visits /login.html
(directly or via auto-redirect), runs the "Create admin account" flow
once, and lands back on the SPA with a real session.
Guards against a redirect loop if login.html itself somehow lands here.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Right-click any event block to open a context menu (Edit, Cancel,
Copy schedule ID, Delete) — actions per status mirror the List view so
the two surfaces stay in lockstep. Menu is viewport-clamped and
dismisses on outside click / scroll, same pattern as the asset menu in
the Library.
Drag-to-resize works for pending schedules only (the schedules PUT
rejects edits to running rows, and terminal statuses are read-only):
- Drag the left edge to move the start time
- Drag the right edge to move the end time
- Drag the body to shift the whole block in time
All gestures snap to 15-minute increments to match the new-schedule
click snap. Minimum duration is clamped to 5 minutes; the block clamps
to the visible day on both edges. While dragging the title shows the
preview range ("Start time → end time") and the block lifts with a
project-tinted shadow.
A short pointer click (< 4px travel) still opens the edit modal — the
click and drag share the same pointerdown so the operator never has
to know which gesture they made first.
Implementation: replaces the <button> block with a <div> hosting three
zones (left handle / body / right handle). Pointer events with
setPointerCapture so drags survive losing the cursor over the block,
and pointerup demotes back to click if travel was below threshold.
Optimistic local update on resize, PUT /schedules/:id with just the
two changed time fields, refetch to reconcile.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Drops dragonflight-premiere-panel-1.0.0-windows-setup.exe (2 MB) and
dragonflight-premiere-panel-1.0.0.zxp (35 KB) at
services/premiere-plugin/build/releases/v1.0.0/ so the binaries have
stable URLs on the forge without needing a separate release artifact
flow.
Heads-up: this commits 2 MB of binary into git history. Future bumps
should use Forgejo Releases (release assets are external to git history
and easy to delete) rather than another commit under releases/.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Time was clipping the full "done May 22 · 2:23 PM · 6h ago" string on
terminal-state rows; Progress's bar felt cramped next to the percent.
Node carries only "primary" / "—" so it can shrink, and Priority's
"normal" / "high" badge doesn't need 80px either. Net widening absorbed
by the flexible Asset column.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The Jobs screen only displayed an asset name when the enqueueing code
stuffed assetName into the BullMQ job data. YouTube imports did that;
upload-triggered proxy/thumbnail jobs didn't — so everything except
YouTube showed em dashes in the Asset column.
Fix it centrally: after we collect jobs from BullMQ, look up names
in one bulk SELECT against the assets table for any job that has an
assetId but no asset_name. Applies to /jobs, /jobs/:id, and the SSE
events stream. Lookup failures fall through silently rather than
500-ing the whole list.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
End-to-end verification on a fresh Windows machine surfaced three issues:
1. pwsh isn't installed by default — Windows ships powershell.exe (5.1).
Switched all script invocations + docs from `pwsh` to `powershell`.
2. .NET's strict XML parser rejects manifest.xml because the <Resources>
comment legally contains `--` (inside `--enable-nodejs`/`--mixed-context`
CEF flag names). Switched build-installer.ps1 to regex extraction,
matching what build-zxp.mjs already does.
3. winget installs Inno Setup 6 to %LOCALAPPDATA%\Programs by default, not
Program Files (x86). Added the user-scope path to the ISCC.exe fallback
list.
Verified: `powershell -File build-all.ps1` produces both artifacts —
dragonflight-premiere-panel-1.0.0.zxp (35 KB, signature valid)
dragonflight-premiere-panel-1.0.0-windows-setup.exe (2 MB).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The EPG JSX components in screens-ingest.jsx ship with the YouTube branch
but the matching stylesheet got lost during the parallel-branch shuffle.
This adds the missing .epg-* block to styles-rest.css and replaces the
dead .cal-* (month-calendar) rules left over from the previous design.
What the styles cover:
- .epg-page / .epg-toolbar — top-level flex layout + date nav row
- .epg-status — sticky "on air" strip with pulse halo on the live dot
- .epg / .epg-corner / .epg-gutter / .epg-canvas-head / .epg-canvas —
the 2x2 sticky grid (top ruler + left gutter both sticky)
- .epg-ruler / .epg-ruler-tick — hour ticks
- .epg-row + .epg-block + .epg-block.live/.failed/.past — event blocks
with project-color 4px inner bar (no side-stripes; impeccable ban)
- .epg-now / .epg-now-pip — vertical hot-red now-line with broadcast glow
- .epg-week + .epg-week-day — stacked 7-day sections for week view
- .epg-empty — recorder-less / loading empty state
Also adds PRODUCT.md and DESIGN.md so future design passes have the
context files the impeccable skill requires. Both drafted from the
existing codebase (tokens, screen patterns) rather than synthesised
from a prompt.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Premiere-running check passed a Boolean to Exec's var ResultCode (Integer)
parameter — Pascal type error. The block also did nothing useful: it only
checked but never warned or prompted. Drop it. {InstallDelete] of the legacy
folder still works through the Tasks checkbox + Check function.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The Tokens screen referenced showCalc / setShowCalc in the Cost calculator
button and modal but never declared the state hook, so the component
threw ReferenceError on mount and rendered blank.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The initial pass referenced zxp-sign-cmd@0.2.2 which never shipped (latest
is 2.0.0) and used the v1.x callback API. v2 is promise-based — rewrote
build-zxp.mjs accordingly.
Also commits the freshly-generated self-signed cert + passphrase from the
first local build run. From now on every build reuses these so Adobe's
ZXP signature-continuity rule is satisfied across versions.
Verified end-to-end: `npm install && node build-zxp.mjs` produces
dist/dragonflight-premiere-panel-1.0.0.zxp (34.7 KB), signature verifies,
cert valid until 2051.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Replaces the manual robocopy / install-windows.ps1 flow with two real
distributable artifacts:
- dragonflight-premiere-panel-<version>.zxp (Mac + Win)
- dragonflight-premiere-panel-<version>-windows-setup.exe (Win)
The Windows installer copies the bundle to %APPDATA%\Adobe\CEP\extensions,
sets PlayerDebugMode=1 for CSXS 8..13, registers an uninstaller, and
offers to remove any legacy com.wilddragon.mam.panel folder so editors
don't end up with duplicate panels.
The .zxp is signed with a self-signed cert generated on first build and
committed to build/cert/ so signature continuity is preserved across
builds (Adobe rejects ZXP upgrades with a different cert fingerprint).
Also migrates the CEP bundle ID from com.wilddragon.mam.panel to
net.wilddragon.dragonflight.panel to match the wild-dragon -> dragonflight
repo rename. Manifest, .debug, CSInterface.js, install docs, and the
growing-files quickstart all updated.
build/ is normally swept by the root .gitignore; added an explicit
negation so the packaging pipeline stays tracked.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds Ingest → YouTube. UI takes a URL + project, API enqueues a BullMQ
"import" job, worker shells out to yt-dlp, lands the MP4 in S3 at the
same originals/{assetId}/... path uploads use, then hands off to the
existing proxy queue. Imported assets share one lifecycle with uploads
from that point on.
Worker container picks up yt-dlp + python3 (apk on alpine, apt on the
GPU variant). The new 'import' queue is registered in jobs.js so it
appears in the Jobs SSE stream and retry/delete work for free.
Spec: docs/superpowers/specs/2026-05-23-youtube-importer-design.md
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds a paste-URL ingest path under Ingest → YouTube. Worker hosts
yt-dlp, downloads to S3, then hands off to the existing proxy +
thumbnail pipeline so imported assets share one lifecycle with uploads.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Nginx was serving css with `expires 1y; Cache-Control: public, immutable`,
which combined with version-less <link href="styles-rest.css"> meant every
browser permanently pinned whatever stylesheet it cached first. Users were
seeing pre-polish-round-2 CSS even after the new image was deployed —
the calendar grid rendered as a vertical stack of weekday names because
the .cal-* rules didn't exist in the cached file.
Move css into the same bucket as js: must-revalidate via ETag. Fonts,
icons, and raster assets stay in the immutable 1y bucket since they don't
change between deploys.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
