* perf(studio): timeline thumbnails come back from disk in the next session
Reopening a project re-read every video's index and re-decoded every strip:
the frame cache and the source facts lived in memory, keyed by the session.
Now each source's facts (aspect, duration) and every decoded frame are kept
in Cache Storage under the file's ETag, the version the server says it is. A
new session checks each source once with a conditional one-byte read: a 304
means the stored frames are still that file's, and a strip they cover paints
without opening a reader. A new ETag drops what was stored for the old one.
Sources are dropped whole, least recently used first, past 256 MiB. A store
that is missing or fails (private window, quota) warns once and the strip
decodes as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): a reopened project checks its videos again, and the disk index counts a frame once
The stored version a session validated was kept per video for the whole page,
so reopening a project after one of its videos was replaced at the same path
painted the old file's frames until a full reload. It is now kept per video
and project open (its contentVersion): each open re-validates once.
The index now counts a frame stored again under its name once, and merges
with what other tabs saved before writing, keeping caches this tab dropped
out, so the 256 MiB budget holds across tabs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(core): a font declared in a linked stylesheet no longer trips the bundler's font warning
* fix(core): drop the guard's explanatory comment
* fix(core): bundled fonts named only in a CSS font shorthand load
* fix(core): a font shorthand loads faces but never fails the cloud system-font check
* test(producer): regenerate static-volume-future-set for the font shorthand face
* test(producer): regenerate two more fixtures for the font shorthand face
* fix(cli): website capture stops inlining slow images and stylesheets and says what it skipped
* fix(cli): a capture ends at four minutes with its partial bundle unless told otherwise
* fix(cli): a capture's deadline starts once its browser is up, and the kept-image count is exact
* fix(cli): an image whose fetch the deadline cut off always counts as kept
* fix(core): a preview media load that fails is reported, not swallowed
* fix(core): an audio fed by source children reports its failed load; the shape lives in the protocol
* fix(core): the audio load watcher finds elements from any realm
* build(core): regenerate the inline position-edits render script
* fix(cli): check probes each audio file once, however many clips are cut from it
* fix(cli): drop an empty stray test file from the audio probe change
* fix(studio): a timeline edit on a clip the file lacks says how the clip was looked up
* fix(studio): name only the lookups a missing clip was really found by
* refactor(studio): split the missing-clip lookup words into small helpers
* refactor(studio): fit the missing-clip lookup words under the no-visible-change size
* fix(cli): a late teardown error no longer turns a finished command into exit 1
* test(cli): keep only the live invariant in the late-rejection test comment
* test(cli): shorten the late-rejection test comment
* fix(engine): late-starting MP4 and WebM videos render every frame the preview shows
The browser counts an MP4 or WebM from timestamp 0, but FFmpeg seeks from the file's
earliest stream start. A file whose first frame comes after 0 (a fragment cut from a
longer stream, an edit-list offset, a WebM with a start offset) was extracted early and
short, so renders tripped the video coverage gate. Every video seek now maps media time
to FFmpeg's timeline, and the video begins where the browser shows it.
* fix(engine): audio from late-starting MP4 and WebM plays when the preview plays it
The mixer seeked and trimmed audio by FFmpeg's clock, which starts at the file's earliest
stream. In an MP4 or WebM whose streams all start after 0, the browser plays that audio
later, so the render heard it early by the same amount. The mixer now maps media time onto
FFmpeg's clock with the same offset frame extraction uses.
* fix(engine): a late-starting file's picture shows on the clock its audio plays on
* test(producer): vfr-screen-recording baseline shows each frame at its timestamp
* fix(audio): a sound is voice only with a voice-range pitch, judged on the whole clip
Two ways the kind guess called a non-voice sound voice:
- A short hit, click or whoosh can have a speech-shaped envelope. classify() now also
measures the share of 40 ms frames with a 70-500 Hz fundamental (YIN, moved from the
CLI into core) and calls a sound voice only when that share reaches 0.15. Speech
measured 0.22-0.43; impacts, clicks and whooshes 0.
- A range query judged a clip's kind on the slice it covered, so 1.7 s of a music bed
read as voice. A film clip's kind is now judged on all of the clip that plays.
Even-paced speech (a 64 s TTS take) moves less at syllable rate and was called music;
with no beat and a voiced share of 0.25 or more it is now voice.
* chore(audio): keep each touched file's comment share at or under its base
The Comments check failed on run.ts, targets.ts and audioAnalysis.ts because the
PR added comment lines to files whose share may only go down. Drops the added
doc comments; no code change.
— Jerrai
* fix(audio): preserve clip evidence and stereo pitch during analysis
* fix(audio): bound and cache whole-clip kind analysis
* fix(audio): retain native-rate kind evidence when available
---------
Co-authored-by: Jerrai <jerrai@heygen.com>
* fix(cli): exit 2 for remaining flag validation errors
Flag and argument checks that commands make themselves still exited 1 and
reported runtime_error under --json. They now take the existing usage-error
path (failUsage / CliUsageError): exit 2 and a usage_error envelope. This
covers lambda and cloudrun render/render-batch (--width/--height, --fps,
positive-number, enum and --output-resolution flags, unknown subcommands),
capture, feedback, benchmark --runs, tts, init, snapshot, compare,
grade-compare, check, normalize-audio, media-treatment, play/present --port,
models, docs, browser and telemetry unknown subcommands, cloud render
conflicting inputs, cloud delete without --no-confirm, and inline --variables.
A few checks now run before project or stack lookup so a bad flag is reported
first. Runtime failures (missing files, network, render errors) still exit 1.
An unattended spinner now writes its start and progress text as plain stderr
lines, not only the outcome, so tools watching init can see "Transcoding".
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): reject a bad snapshot --frames, quiet spinner progress, drop added comments
- `snapshot --frames abc` (or 0, -3, 2.5) is now a usage error: exit 2, and a
usage_error envelope under --json. It used to fall back to 5 frames silently.
- An unattended spinner prints its start line once and its outcome, but no
longer one line per progress update, which flooded stderr during
remove-background and the Chrome download.
- Removes the comment lines this branch added, so the comment ratchet passes.
- The render --json test now expects exit 2 and usage_error for bad inline
--variables.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): every --json failure answers with one envelope, and --json only where supported
Under --json, failures came back as a bare string (`"error": "Command failed"`
from lint), a command-specific shape, or nothing at all on stdout (info,
compositions, render, snapshot, cloud get). Now every failure is
`{"ok": false, "error": {"code", "message", "hint"?}}`:
- jsonFailure() in utils/jsonError.ts builds it. The commands that printed
`{ok: false, error: "<text>"}` use it, keeping their extra fields (lint's
findings, validate's errors). A refused timeline edit now writes it to stdout
instead of stderr, and keeps its `reason` and `fix` fields.
- When a --json run fails without writing anything to stdout, cli.ts writes the
envelope itself: usage errors carry their own code (unknown_flag,
unknown_command, json_unsupported, confirmation_required), anything else is
runtime_error with the message the command printed on stderr.
--json is no longer accepted by every command. A command supports it by
declaring a `json` arg; the others reject it with exit 2 and the code
json_unsupported, instead of accepting the flag and printing prose.
The contract test now checks every command's --json path, and packages/cli
gets an AGENTS.md with the rules: stdout is data, stderr is messages, never
prompt without a TTY, exit 2 for usage errors, --json always returns an
envelope.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): put the real failure in the --json envelope, never the generic text or a notice
lint --json and check --json in a directory with no project answered
"message": "Command failed". The project check reports the problem, then
throws a presented "Command failed" with the real error as its cause, and the
command's own catch serialized that generic text. failureMessage() now returns
the cause, or else what the command reported through errorBox, and every
command catch that writes a --json failure uses it (lint, check, validate,
layout, snapshot, compare, grade-compare, media-treatment, tts,
remove-background, add, skills, lambda policies). compare, grade-compare and
add share failCaught() for that catch.
The boundary envelope follows the same order: the error's own message unless
it is the generic one, then its cause, then the errorBox title and hint, and
only then the stderr text. Notices (the telemetry disclosure, deprecation
lines, diag progress) are no longer captured, so validate, layout and inspect
no longer put "is deprecated" in the error message.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): render --json keeps human output off stdout so a failure answers with the envelope
Under --json a single render printed its header, lint findings and progress on
stdout, so a later failure such as `render --variables '{bad' --json` ended
with no JSON envelope. --json now silences that output the way --batch --json
already did, and --variables is parsed before the browser check so a bad value
fails without a Chrome download.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): a --strict lint abort says why, even under --json
Under --json the render is quiet, so a --strict lint abort printed nothing and
the envelope said only "Command failed". The abort is now an error box on
stderr in every mode, naming the error and warning counts and the first
finding, and the --json envelope carries the same message. Lint now runs
before the browser check, so the abort never waits on a Chrome download.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* test(cli): time the render setup cancel from the wrapper's exit
The 1 s budget started when the command did. With lint now running before the
browser check, its subprocess (about 250 ms here) sat inside that budget, and
the test failed once under load. The budget now starts when the wrapper exits,
which is the latency the test is about.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(inspector): live queries see every composition, wherever the playhead is
The Studio preview parks scenes away from the playhead, so a live inspect, timeline or
element-map query answered missing for every element in a parked composition. Each live
query now holds every scene mounted under its own holder, then hands the scene window back.
__hfHoldScenes keeps holds per holder, so the inspector letting go never drops a scene
Studio holds.
* fix(inspector): isolate concurrent live query scene holds
* fix(cli): resume a model download that stalls partway instead of failing
* fix(cli): pin the idle-only timeout; one-line doc for the resume
* fix(cli): resume from the bytes on disk and check the size before caching
* refactor(cli): split the download response handling into named steps
* fix(studio): a thumbnail reader forgets a start's failures after a minute
A pooled thumbnail reader lives for as long as clips keep asking it, and each start
byte has a retry budget of two. That budget never reset, so two transient failures at
one start early in a session made a third, twenty minutes later after the cache evicted
it, give up at once: the reader retired and the clip showed an error.
A start's failures now count for a minute. One bout of retries (8 s stalls, 0.5 s and
1 s waits) fits inside it, so a start that keeps failing still gives up; one met again
much later starts afresh. Clearing the count when a read from that start delivers bytes
would not do: a server that ignores Range sends bytes 0 onward again before it breaks at
the same byte, and would retry forever.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): the start-failure window says what it is in its name, not a comment
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): a start's retry budget resets when a read from it ends cleanly, not after a minute
The one-minute window only held while each attempt took under a minute. An attempt has
no limit before its answer, and a body that keeps sending never stalls, so a server that
ignores Range, sends slowly and breaks past a minute re-downloaded from 0 forever.
A start's failures are now forgotten when a read from it ends cleanly: its body finishes,
or mediabunny cancels it at its target. A server that breaks before the target never ends
a read cleanly, so its budget still runs out, however slowly it fails.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): a read mediabunny stops at its target ends cleanly and resets its start's budget
mediabunny's UrlSource never cancels the body: once a read reaches its target, _runWorker
aborts the signal it passed in, and the pending pull fails with an AbortError. So the reset
on cancel never fired, and any open bytes=N- read stopped at its target kept its start's
failures.
The request now listens on its own signal, which follows mediabunny's and is unlinked once
the read ends: an abort while live counts as a clean end only when mediabunny's signal is
the one aborted, never a stall or a dispose.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* ci: keep the bun cache warm and skip binary downloads in preflight
* test(ci): pin the single cache writer and preflight's skipped downloads
* test(ci): keep the cache-writer scan out of the test body
* fix(studio-server): the render route passes a MOV ProRes profile to the render
* fix(studio-server): drop the added comment line so the comment share does not rise
* fix(cli): a Windows browser that will not start says where and why, not just spawn UNKNOWN
* fix(engine): the launch diagnosis names a browser file Windows refuses to run
* fix(engine): the launch diagnosis names a browser file Windows refuses to run (engine part)
* fix(cli): add help [cmd], suggest the closest command or flag, and quiet the nags
`hyperframes help` printed "Unknown command help" with raw ANSI codes and no
suggestion. It is now an alias: `help [cmd...]` is `[cmd...] --help`.
An unknown command (at the root or inside a group like `cloud`) prints
"Unknown command: rendr", "Did you mean 'hyperframes render'?" and the group's
usage, all on stderr, and exits 2. An unknown flag suggests the flag it most
likely means ("Did you mean --output?"). Usage text printed for a usage error
now goes to stderr instead of stdout, and citty's colors are stripped from
help whenever this CLI's own colors are off.
No update, skills or app-history notice is appended to a usage error. The
skills notice shows at most once a day per project (timestamps in
~/.hyperframes/skills-notice.json) and, in an unattended run, only when a core
skill is missing.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): reject an unknown top-level command without loading every command
citty looks for an alias on every registered command before it reports an
unknown name, which loads all of them. From a source checkout without the
studio packages built, that meant `hyperframes rendr` failed in preview's
build check with exit 1 instead of suggesting `render`. The name is now
checked against the registry first.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): help with a mistyped subcommand reports it like the command itself
`hyperframes help cloud rendr` printed cloud's help and exited 0, while
`hyperframes cloud rendr` exits 2 with a suggestion. Both now exit 2 with the
suggestion. A group that also takes positional arguments (timeline takes a
project directory) still shows its help for a word it does not know.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): check prints a progress line as each pass after overlap begins
* fix(cli): check counts finished passes and drops the comment the ratchet flagged
* fix(studio): a slow thumbnail job keeps a reader that other clips still decode from
A job that timed out judged the reader by its own progress only. A poster job that got
an already-open reader while a sibling was decoding on it, and took more than its
budget to reach its one frame (a slow link, Chrome's six connections held), logged a
timeout and retired a healthy reader, so the next clip re-read the index.
The pooled reader now counts the frames every job decodes from it, and a timed-out job
retires it only when no frame came from it while the job held it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): the reader's frame count says whose frames it counts in its name, not a comment
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): undo and redo on Windows restore a file something holds open
* fix(studio-server): a restore waits for the preview's reads to close before replacing the file
* fix(studio-server): a restore checks the project folder after the preview's reads close
* perf(aws-lambda): small-chunk cloud renders stitch their output in parallel
When every chunk's payload is under S3's 5 MiB part minimum (typical for short 4K chunks), the assemble step could not copy any part server-side and pulled the whole output through one Lambda's network link. Those range-only parts now go to parallel invocations of the warm function that just rendered the chunks, about 32 MiB each and never more at once than the stack's reserved concurrency allows. The output bytes are unchanged.
* fix(aws-lambda): a throttled stitch helper builds its parts in the assemble instead
Another render can hold the stack's reserved concurrency while this one assembles, so a helper invocation may be refused. The assemble now builds that batch itself, logs it, and counts it in assemble_stages; the helper's error type is kept as the error name. Comments trimmed back to each file's share at the base.
* chore(aws-lambda): raise the package size budget to the measured size plus headroom
The parallel stitch helpers add about 1.7 KB packed and 9.9 KB unpacked; the new budget is the measured size times the checker's own 1.05 headroom.
* perf(engine,cli): a machine's GPU encoder probe runs once, not on every render
* refactor(engine): one shared probe per process, no answer kept on disk
* fix(cli): never prompt or auto-install when nobody is at the terminal
publish and upgrade called the confirmation prompt with no terminal check.
With stdin from /dev/null, publish exited 13; with an open pipe it hung.
Unattended and without --yes, both now exit 2 before doing any work, and
stderr names --yes (upgrade also names --check).
All prompts and spinners now go through ui/prompts.ts, the only module allowed
to import @clack/prompts (a lint rule and a contract test enforce it). Its
confirm, select, text and password refuse to prompt in an unattended run, and
its spinner prints only the final line, on stderr. init, feedback, catalog and
cloud delete decide whether to ask with isUnattended() instead of a TTY check
of their own.
The background auto-install no longer starts in an unattended run, and the
update notice ("Run: npm install -g ...") is not shown while an install for
that version is pending. CI now counts as set for any value but "false" or "0".
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* test(cli): neutralize agent detection in the catalog offer tests through isAgentDriven
Attendance now reads isAgentDriven(), so that is what the catalog tests stub to
stay independent of the agent running the suite.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* ci(windows): probe that a render opens no visible window
* ci(windows): the probe names what opened each window
* ci(windows): the probe spawns each binary alone, with and without detached
* ci(windows): the probe logs the processes started during the render
* ci(windows): the probe also spawns Chrome one consoleless hop down
* ci(windows): the probe logs each child started without windowsHide
* ci(windows): the probe logs every child process with its windowsHide and detached
* fix(engine): Chrome keeps the CLI's console on Windows, so a render opens no window
* fix(cli): every child the CLI starts on Windows is hidden unless it opts out
* ci(windows): the probe bisects Chrome's flags and keeps each command's output
* ci(windows): the probe names the Chrome process behind each window
* ci(windows): the probe loads a page in Chrome directly and one hidden hop down
* ci(windows): the probe prints the Chrome main command line and renders with the reporter's flags
* ci(windows): the probe narrows the CLI's Chrome arguments to the ones that open a window
* ci(windows): the probe launches the CLI's Chrome arguments one hidden hop down
* ci(windows): the probe launches Chrome through the engine's browser pool
* ci(windows): the probe varies what the engine passes to Puppeteer
* fix(engine): renders on Windows launch Chrome with a Puppeteer that keeps it attached
* refactor(cli): hideChildConsoles copies its arguments and keeps one comment line
* feat(core): name why a choppy preview window dropped frames
The playback reading now says how much of the window was frames skipped
and how much was a playing video waiting for data, and carries the p50
and p95 gap between the runtime's ticks, so a report can tell a slow
compositor from a stalled decoder.
* chore(core): drop comments the ratchet counts
* perf(producer): drawElement films keep their parallel workers instead of dropping to one
* perf(producer): shorter inversion floor comment
* test(producer): inversion suites opt into the single-worker inversion
* perf(studio): keep released timeline video frames for scroll-back
A video thumbnail frame was revoked the moment no mounted strip held it, so
scrolling back past the scheduler's cache decoded every clip again: 80 decode
passes for 40 clips on a 44-minute source, the whole first pass repeated.
Released frames now stay shared in a least-recently-released list capped by
the new sharedFrameCacheBytes budget (32 MiB of JPEG, about 11 frames for
each of a 473-clip episode's clips). Object URLs are revoked only on
eviction, and a frame any strip still shows is never evicted. Which frame a
strip shows is unchanged: the acceptance window is the same predicate.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* perf(studio): open each video source once for all its timeline clips
Every clip's thumbnail job opened its own mediabunny Input on the source URL,
so 40 clips cut from one 44-minute source opened it 80 times, re-reading and
re-parsing its 1.44 MB index each time (behind a 304, at 13-30 ms an open).
Jobs now share one pooled reader per source: the Input, its primary video
track and the track's size and duration, reference-counted and disposed
thumbnailSourceIdleMs (10 s) after the last job ends, or at once when the
source's contentVersion changes. Each job keeps its own CanvasSink: the sink
reuses one pooled canvas and a job blanks the canvases it saw on release, so
sharing one across jobs would corrupt a neighbour's frames.
A failed open leaves the pool at once, and a job that fails for any reason but
cancellation retires its reader, so the next clip opens afresh. Retries are
still counted per reader, as mediaUrlSourceOptions has done since #6477: two
jobs reading through one reader cannot tell whose read failed, and retiring a
reader whose retries are spent keeps that budget per reader rather than letting
a spent count fail later clips at once. The pool also owns the one
import("mediabunny"), which the decoder now takes from the opened source.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* perf(studio): start no timeline video decodes while the preview plays
A playing preview shares the machine's video decoder with timeline
thumbnails, and a long many-clip timeline queues hundreds of thumbnail decodes.
While the player store reports isPlaying, the scheduler now starts no new
visible or overscan video decode; a scrub's interaction request still starts,
decodes already running finish, and pausing drains the queue. Image,
composition and waveform work is unaffected.
The new thumbnailPlaybackHold module subscribes to the store's state rather
than wrapping setIsPlaying, so a reset or a project switch, which write
isPlaying: false directly, also release the held work. VideoThumbnail binds it
to the app's scheduler where video thumbnails load, which is also the path
HyperFrames Desktop runs (it embeds StudioApp, not Studio's main.tsx).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* perf(studio): bound thumbnail range reads and their read cache
mediabunny reads a URL with open-ended `Range: bytes=N-` requests and cancels
each once it has the bytes it wanted, by when a local server has streamed
megabytes past them: about 4 MB sent for each thumbnail frame decoded. The
thumbnail reader's fetch now asks for at most 512 KiB from N; a shorter answer
than mediabunny wanted makes it ask again from where it stopped (its worker
resumes a capped response when the file size is known, which the first
206's Content-Range supplies). Closed ranges and the probe reader are untouched,
and so is Studio's playback, which does not read through mediaUrlSourceOptions.
Measured on two 44-minute 1080p sources, 40 clips each (a poster and a
4-frame strip per clip), against the same branch unbounded: bytes sent fall
13-22% one decode at a time and 33% two at a time; first-pass wall time falls
4-20%. The frames decoded are byte-identical. Most of what remains is real:
a strip slot whose keyframe is outside its window decodes from that keyframe
forward through its GOP.
A pooled reader also now caches at most 16 MiB of what it read, not
mediabunny's default 64 MiB, since it outlives its clip by up to
thumbnailSourceIdleMs. 8 MiB cost requests when a clip's poster and strip
decode together; 16 MiB measured the same as 64.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): bound thumbnail ranges only once Content-Range is readable
A thumbnail reader bounded even its first range request to 512 KiB. When the
answer's Content-Range cannot be read, mediabunny sizes the file as the request
start plus the answer's Content-Length, assuming the answer runs to the end of
the file, so a bounded first answer made a 10 MB file end at 512 KiB. Browsers
hide Content-Range cross-origin unless the server lists it in
Access-Control-Expose-Headers, and Studio reads remote https sources such as
HeyGen's gen-os-static bucket, which does not. With the header hidden, every
format failed: faststart and fragmented MP4 decoded 1 of 12 thumbnails, and
moov-at-end MP4, MOV and WebM none ("no decodable video track").
A reader now bounds its ranges only after an answer of its own was a 206 whose
Content-Range it could read and that starts where the request asked; its
first request stays open-ended, as before the bound. Any answer without such
a Content-Range, a whole-file 200 included, keeps that reader unbounded for
good. Same-origin readers, where Content-Range is readable, still bound.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): count a thumbnail reader's read failures consecutively
A reader's retry budget (two retries, from #6477) counted every failure over
the reader's life. Pooled, a reader serves every clip of its source, so the
third transient failure anywhere on the source failed a clip with no retry.
The count now resets whenever a read delivers bytes, so a long-lived reader
keeps the per-clip tolerance a reader had when each clip opened its own, and
three failures in a row still give up. Bytes, not an ok status: a body that
breaks on its first read must still use up the budget. A reader that gives up
is retired, as before, so the next clip opens a fresh one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): a read stalled mid-response no longer hangs its video source
The thumbnail job's signal reached only the source's open: keyframe lookups and
the decode loop ignore it, so a job whose response stalled mid-body never ended
and never released its pooled reader. The reader was never idle, so never
closed, and every later clip that needed the same bytes waited on the same
stalled read for the rest of the session. Before pooling, each clip had its own
reader, so only the stalled clip hung.
The scheduler's timeout now aborts with a TimeoutError reason. A job that times
out drops its reader at once; a job cancelled any other way that has still not
ended a load timeout later drops it then; a cancel the job answers, a scroll
away, leaves the reader pooled. Dropping disposes the reader whoever still uses
it, which fails their pending reads too.
Disposing a mediabunny Input does not end a read already waiting on a stalled
body (its orchestrator only flags the worker, which never wakes), so
mediaUrlSourceOptions gains cancelReads(), which aborts every fetch the reader
has in flight; the worker's read then fails and mediabunny rejects the reads
waiting on it. Measured in Chromium with a server that stalls one response:
the stalled clip settles at its timeout, and the same clip asked again, now
and after 11 s idle, decodes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): a clip's own failed result keeps its source's reader
A thumbnail job that failed for any reason but a cancel retired its pooled
reader, so one clip whose frames all failed to decode, or whose canvas would
not encode, made the next clip re-read and re-parse the source's index. Those
two failures are now a ClipResultError, which leaves the reader pooled; fetch,
demux and decoder failures still retire it.
Also pins four pool behaviours mutation testing found unguarded: a reader
closed after sitting idle has left the pool, so the next clip opens a live
one; a version change closes the old reader only once the job still decoding
on it ends; a job cancelled while the track is read stops the open there; and
a scroll-away cancel keeps the reader (the last landed with the stall fix).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): keep a thumbnail reader's retry budget per start byte
ff16aed98 reset a pooled reader's retry count whenever any read delivered
bytes. That broke the #6477 give-up in two ways, both measured with real
mediabunny:
- A server that ignores Range answers 200 from byte 0, so mediabunny resumes
at the bytes it got and is sent byte 0 again. With a body that breaks after
64 KiB, those bytes reset the count every time: 61 fetches in 30 s and the
read still pending.
- mediabunny reads with two workers. A healthy stream in one region kept
forgiving a region that always answered 503: 41 fetches in 20 s, pending.
Failures are now counted per start byte, with no reset. A resume after
progress asks for a new start and gets its own two retries; asking for the
same start again uses them up whatever else the reader reads meanwhile. The
start is the byte the answer began at: the asked start for a 206 or a failed
request, 0 for a 200. Traced for the 200 case: requests at 0, 65536 and 65536
are all keyed 0, so the third failure gives up. Both repros now give up after
3 fetches ("network TypeError after 2 retries", "HTTP 503 after 2 retries"),
and failures at different starts, one per clip, never add up.
The fetch and the answer's typing move into typedFetch and typedResponse, so
fetchFn stays under the complexity gate.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): fail a stalled thumbnail read on silence, not on a timeout
dbd820706 dropped a reader when the job using it timed out, but only once the
reader had opened: a timeout during the shared open released the job and left
the stalled open pooled, so every clip of that source joined it and timed out
in turn while clips kept arriving. Dropping on a timeout was also wrong when
bytes were still coming (a slow link's 25 s open threw its parsed index away),
and dropping a reader other clips were using failed them with "Media reads
cancelled", leaving them on "no preview" until they remounted.
A thumbnail read now fails only on silence. A request waiting 8 s for its
answer, or for its next bytes, is aborted and fails as a MediaFetchError
"stalled", which mediabunny retries like any read failure: a resume after
progress asks for a new start and continues, and three silent waits at one
start (8 + 0.5 + 8 + 1 + 8 = 25.5 s, inside the 30 s load timeout) give up,
which retires the reader. This covers the open and the decode alike, keeps a
reader whose bytes are still arriving, and never disposes a reader another
clip is using, so the job-level abandon and the scheduler's TimeoutError
reason are removed. Measured in Chromium against a stalling server:
- one-shot stall in the open: the clip decodes at 8.7 s (was: it and the next
clip both timed out at 30 s);
- one-shot stall mid-decode: decodes at 8.7 s (main: hangs);
- a source that never sends a byte: each clip fails typed at 25.5 s and the
next opens a fresh reader.
Each request gets its own AbortController, which follows mediabunny's signal
and the reader's cancel without AbortSignal.any; every way a request ends
unlinks it from the reader's long-lived cancel signal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(studio): pin the thumbnail behaviours mutation testing left unguarded
- A bounded answer's Content-Range keeps the reader bounded: a closed-range
pattern that matched only open ranges flipped it unbounded after the first
bounded request, undoing cf8f4554e.
- A frame that will not encode, through toBlob or OffscreenCanvas
convertToBlob, is the clip's failure and keeps its reader. The
convertToBlob rejection is now a ClipResultError too; it escaped as the
browser's own error and retired the reader.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(studio): pin the stall watchdog's guarantees about its own requests
Four guarantees of the thumbnail reader's per-request watchdog had no test;
each mutant that broke one passed the suite:
- a silent body fails as "stalled" and aborts the fetch under it, so the
stalled connection is not left open;
- every way a request ends unlinks it from mediabunny's signal: an answered
body, a rejected fetch, a 503, no body, a broken body, a cancel;
- a probe reader's body is never timed, however slow;
- a request made after the reader's reads were cancelled starts aborted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): time a thumbnail read's silence only once its answer has started
The 8 s stall rule also timed the wait for a response's headers, and that
wait includes Chrome's queue for one of six HTTP/1.1 connections per host.
Desktop serves all media from one loopback host, where Studio's event stream
and streaming previews hold connections. Measured in Chromium against a
healthy server: with five of six connections held and 20 cancel/re-ask bursts
over the 40-clip pass, no false stall; with all six held for 40 s, two clips
failed as "stalled". The watchdog now times only a started body.
A server that never answers is now caught at the job's own timeout. The
scheduler's timeout again aborts with a TimeoutError reason. A job that timed
out without progress in its whole run (its source never finished opening, or
it decoded no frame) failed because of its source: a source that never
answers, or one that trickles bytes and never finishes. It logs one typed
MediaFetchError line ("network timeout") and retires its reader. A job that
waited for an open that finished, or decoded a frame, timed out on a slow link
and keeps the reader and its parsed index. Bytes alone are not progress, or a
trickling source would keep its reader for good.
An aborted job also ends its claim on the reader at the abort, rather than
when it settles: a job stuck in a read that ignores its signal no longer keeps
the reader open. A retired reader still waits for the clips decoding on it.
Measured in Chromium: a source that never answers fails each clip with one
typed line at 30 s and the next opens a fresh reader; with all six
connections held for 40 s, the two starved clips fail typed, none as a stall.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(studio): move the thumbnail reader's explanations into names
The comment ratchet (scripts/comment-ratchet.mjs) fails a source file whose
share of comment lines rises over its merge-base, and a new file above its
package's share. Five files of this branch did. The reasoning they carried is
in the commit messages and the PR description; what stays in code is named:
SILENT_BODY_STALL_MS, POOLED_READER_CACHE_BYTES, failuresByStart,
inFlightReads, decodedByJob, timeoutWithoutProgress and madeProgress. No
behaviour changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(cli): define every command through one factory and test them all as an agent
Every command now comes from defineCommand in utils/defineCommand.ts. A lint
rule bans importing defineCommand from citty anywhere else, and a new contract
test walks the command registry in cli.ts, checks each command was built by the
factory, and runs it the way a coding agent does: no terminal, stdin closed,
CI=1, CLAUDECODE=1 and a home folder of its own.
Usage errors now exit 2: unknown flags, extra positionals, and citty's own
parse errors (missing argument, bad value, unknown or missing subcommand).
An unknown flag names the help command instead of printing nothing else.
isUnattended() in utils/attendedTerminal.ts is the one answer to "can a person
respond here". It is true without a TTY on stdin or stdout, under CI (any value
except "false" or "0"), when the existing agent detection finds an agent, or
with HYPERFRAMES_AGENT=1. Color is off when it is true.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* test(cli): spawn preview in the agent-mode test only when the studio packages are built
From source, preview refuses to load until the studio workspace packages are
built, and the CLI test jobs do not build them. preview is still checked for
being defined with the shared factory.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(cli): count only agent-run shells as unattended, and exit 2 for missing arguments
isUnattended() used the telemetry fingerprint detectAgentRuntime(), which also
matches a person typing in Cursor's or Windsurf's terminal (TERM_PROGRAM) and
any Replit shell (REPL_ID, REPLIT_USER). It now uses isAgentDriven(), which
keeps only markers an agent sets on the shells it runs: CURSOR_AGENT counts,
TERM_PROGRAM=cursor, TERM_PROGRAM=windsurf and the Replit variables do not.
The hand-written argument checks in lambda, cloudrun, lambda policies,
capture, remove-background, init (--resolution), tts (--lang) and cloud render
(--idempotency-key, --dry-run) exited 1. They are usage errors and now exit 2.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- .claude/settings.json: the typecheck step piped through grep inside
single quotes, which cmd.exe splits into a broken pipeline, so the hook
denied every commit on Windows. On every OS, `|| true` also made the step
pass regardless of type errors. Test-file errors are now filtered in JS
and any remaining `error TS` line fails the step.
- scripts/check-studio-copy-dashes.mjs: report paths with forward slashes
so EXEMPT entries match on Windows.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* codegen: resync cloud client from EF 354b4cf4
* chore(codegen): keep page_motion_blur in the generated render request
The CLI already sends page_motion_blur (#6087), and experiment-framework adds
it to the public spec in a pending change. Keep it here so the sync does not
break the CLI typecheck in the meantime.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: James <james.russo@heygen.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
