* fix(cli): wait for the cloud's file check as long as the file's size allows before giving up
* style(cli): format the complete-wait test
* fix(cli): keep the complete-wait settings private to the upload module
* refactor(cli): split the upload retry decision and the test drain to keep complexity down
* test(cli): drain the slow upload test server at a fixed rate whatever the chunk size
* fix(cli): two cuts at the same age are the same cut whatever bytes each got through
* fix(cli): sample upload progress on its own interval so a short try still reports its bytes
* style(cli): format the cloud upload files
* fix(cli): cloud render zip carries files the composition references under dot-folders
* fix(cli): a missing link href no longer fails the cloud zip; media refs still do
* chore(cli): shorten the referenced-file comment
* fix(cli): only a missing video, audio, img src or poster stops the cloud zip
* fix(cli): a stylesheet url() never stops the cloud zip; only the no-fallback media refs do
* feat(check): every check step prints a line on stderr and repeats it every 30 s while it runs long
* feat(check): name the saved report file on stderr and tell agents not to use npm run
* fix(check): print a step's line once when the step starts, not on every progress ping
* style(check): wrap two long lines for oxfmt
* fix(check): say the report file is written unless a not-recorded warning follows
* chore(skills): refresh the skills manifest hash for the check doc line
* fix(check): keep the start line free of the not-recorded warning text
* fix(cli): a cloud upload cut at a fixed age stops after two identical cuts and shows its progress
* test(cli): size the upload tests so the kernel cannot buffer the whole file
* test(cli): keep the upload test server quiet when the client drops the connection
* test(cli): shorten the upload test server comment
* fix(check): count a Korean number and its counter as one word in line reading time
* fix(check): keep segmenting Thai, Lao, Khmer and Burmese when counting reading words
* feat(cli): audio --read --layer music|sfx|voice reads a one-layer file whole
A caller that knows a file holds a single layer can declare it: the file is
read whole as that layer with no stem separation and nothing uploaded.
Without the flag the read is unchanged. An unknown layer is rejected.
Consumer: heygen-com/experiment-framework#66322. Proposal OK'd by Rory
(C0BEH487Q7R, 2026-11-07).
Co-Authored-By: Jerrai
* chore(cli): drop added comments from audio read layer
Co-Authored-By: Jerrai
* fix(cli): price GPT Image 2.6 and every GPT Image reference picture
The media-use image estimate priced every non-default model as GPT Image 2.5 and
charged references at $0.0082 after the first. fal bills every reference picture,
the first one too, at about $0.0125, and the price book now does the same. GPT
Image 2.6 and 2.6-flare take 2.5's prices; nano-banana-2.1 gets its own; any
other model is not priced ahead instead of guessed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* chore(cli): drop the new pricing comments to keep the comment share flat
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* feat(cli): the sign-in page the browser lands on follows the design
* style(cli): format the sign-in page files
* fix(cli): the sign-in page picks Cancel or failure inside the page builder, and drops unused code
* fix(producer): a failure before the first frame reports frame capture as its stage
* fix(producer): a failure after the effect scan reports frame capture, with a stable code
* fix(engine): a render no longer finishes without an image that loaded late
* style(engine): format the frame gate
* fix(engine): an image outside the frame no longer stops a render
* fix(engine): a render stops and warns when a frame's image never arrives, with no cloud retries
* fix(producer): a cloud chunk keeps the missing image's error name so retries stop
* feat(studio): show the agent's storyboard in a Storyboard tab beside the preview
The board an agent writes per the storyboard recipe (boards/storyboard.html,
or a root storyboard.html) now shows in Studio the way HyperFrames Desktop
shows it: a Preview | Storyboard strip over the preview panel, only while the
project has a board. Studio reads the page's class="frame" boxes into a board
of stills, with acts, titles, times and words, and one frame at a time with
arrow keys. A page with no frames shows as written in a script-less iframe.
A board written while Studio is open brings its tab to the front, as Desktop
does at the agent's stop. A rewrite of the page, a still beside it or
STORYBOARD.md reloads it in place, and a dot marks a rewrite seen from the
preview. "Show in preview" goes back to the film at the frame's start.
Notes, drawings and the go belong to Desktop's agent, so the board's bar
carries the desktop-app button. Its funnel events now say which button was
pressed (placement: header or storyboard).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* refactor(studio): split the storyboard view and its frame reader into smaller parts
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* docs(skills): the agent's storyboard stop points at the Storyboard tab, not the file
With the board shown beside the preview in Studio as in Desktop, the recipe's
stop and style-sampling's stop 1 tell the person to look in the Storyboard
tab, and to start a preview and give its link when none is running, instead
of handing over boards/storyboard.html to open from disk.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(studio): the storyboard listens on Studio's own file-change channel, and its notes are trimmed
The board opened its own /api/events stream, which the Vite dev server
behind the runtime smoke does not serve. It now subscribes through
subscribeFileChanges, the one channel picker the external-change
coordinator already uses (test adapter, Vite HMR, then SSE), so dev and
built Studio deliver the same changes to both.
The new files' comments are cut to the package's share, and the recipe's
Storyboard-tab line moves to the paragraph that names the board, leaving
the stop step as it was.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(studio): the board hears file changes through the coordinator, not a second subscription
The external-change coordinator stays the one subscriber to the file-change
transport (HMR in dev, SSE when built), as externalFileChangeOwnership
asserts. It announces each decoded change through fileChangeFanout, and the
storyboard listens there. Mounting EditorShell alone, as its tests do, no
longer opens an EventSource, and the coordinator file stays under the line cap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* feat(studio): the storyboard's Open in desktop app asks the app to open on the board, and its card says why
- The storyboard bar's button posts `doc: "storyboard"` with its page id. The preview server passes only that value
on, and `openInDesktop` writes it into `.hyperframes/desktop-cta.json` beside the page id, so the app can open the
project on its Storyboard tab with Build in sight (hyperframes-internal reads it).
- The guide card opened from the storyboard says "Approve this storyboard in the desktop app" and "Comment on a frame,
draw on it, then press Build", and its last step says the app opens on this storyboard. The header's card is
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* feat(studio): approve the storyboard in Studio; comments on frames stay in the desktop app
A Studio user without the desktop app had no way to approve the board from the tab, and the bar said approval was
in the desktop app, so the tab read as a dead end. Studio talks to the person's own agent through their chat, so the
bar's Approve copies "The storyboard is approved. Build the video from it." for them to paste to it, and says so in
a toast (or spells the words out when the clipboard refuses). Commenting and drawing on frames stay in the desktop
app. The storyboard recipe tells the agent those pasted words are the person's go.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(studio): the storyboard's desktop card offers commenting, since approving now happens in Studio
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(studio): hand-off intent before dispatch, steady film under the board, stills outside the board folder
- openInDesktop writes desktop-cta.json before sending the folder, so a running
app reads the storyboard ask on arrival; a plain or header open drops an
earlier storyboard ask; a failed open puts the earlier record back.
- PreviewStage keeps the film at one place in the tree with or without a board,
so a board appearing never remounts it; the automatic switch goes through the
same pause as a click, and a play while the board is up is paused.
- The board reloads when a still it draws changes, including a root board's
stills/ and a boards/ page's ../assets.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* chore(cli): fit the Storyboard tab in the CLI's size budget; reword a record helper's doc
The CLI tarball carries Studio's build, which now holds the Storyboard tab; merged
with main it measures 7,982,002 bytes packed, 7 bytes past the 1% tolerance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(studio): hold the player while the storyboard covers the film
A pause after the fact let J's reverse loop schedule its next frame, so the
hidden film kept seeking. The stage now takes the player's own hold while the
board is up (and only when no one else holds it), so J, Space and playback
requests are refused before they start; Preview releases it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(core): motion blur reports what a composition holds, as counts
The runtime scans a composition once per blur scan, on both tiers, and reports
counts of what it holds (canvases, video, frames, custom elements, clips, masks,
scroll containers, 3D, blend modes, and movers that overflow, sit under other
content, hold fixed children or carry pseudo-elements outside their box), plus
whether pose reads need callbacks. Counts only: no ids, classes, text or URLs.
The engine puts the non-zero counts on its [motion-blur] line and hands the
parsed report to the perf summary; the producer aggregates workers (first report
wins, failed frames sum) and the CLI sends motion_blur_* properties on
render_complete, from the CLI and from Studio renders.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* fix(cli): disclose blur counts in telemetry docs, raise inspector budget
The runtime fingerprint code is bundled into @hyperframes/inspector, which
puts its unpacked size 2.2 KB past the 1% tolerance. The telemetry docs now
list the composition counts the CLI sends with a blur render.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* feat(cli): check_report counts the browser-playable copies a check waited for
* style(cli): shorten the copy facts comment under the line limit
* fix(cli): a check stopped by the copy wait stays browserSkipped; count only abort-caused unfinished
* style(cli): format the copy facts tests
* fix(templates): pin the five style-sampling packs with the release-check fixes
New bundles for all five creator templates:
- science-explainer: the end card shows over the last section (assemble.py isolates
every mounted clip), instead of being hidden in every film.
- collage, 3d-brand, 3d-true-crime: add_sound.py runs on ffmpeg 4.2 and says what
failed instead of raising an AttributeError.
- all five: a voice clone is the one the person names as theirs; `media-use voices`
lists the whole workspace's clones and is never picked from by count or language.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(templates): re-pin the science pack so a Desktop voice tag is a pick, not consent
The science pack took Desktop's `[voice: your HeyGen voice ...]` tag as
settled and asked nothing. Desktop's own-voice picker lists every clone the
workspace can use, so the tag only records which voice was picked. The new
M9003 bundle uses a picked, named or id-given clone only when the person
says it is theirs or that its owner agreed, and asks once otherwise. The
other four packs are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): media-use voices says whose clones it lists, and --voice takes language codes
`media-use voices` lists every clone the account's workspace can use, a teammate's
included, and HeyGen does not say who made each one. The command's description and
output (`whose`) now say so, and the style-sampling remix rule no longer reads
"One clone: use it": a clone is used only when the person names it as their own.
`--voice en`, `en-US`, `zh`, `zh-CN` and `cmn` were sent to HeyGen as language names
it does not list, so it answered with a multilingual voice ("Gail in car") and the CLI
used it. Codes now map to the English language name HeyGen lists, and a multilingual
voice offered for a language the request did not name is refused with the fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* chore(skills): regenerate the skills manifest
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* refactor(cli): keep languageName and the voices note module-private
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(skills): a clone's id is not consent; use one only when the person says it is theirs
The remix guide used a voice id the person gave, or a unique prefix of one,
without asking whose clone it was. The workspace list includes teammates'
clones, so either path could voice a film with someone else's clone. Both
paths, and the `whose` line of `media-use voices`, now use a clone only when
the person says it is theirs or that its owner agreed. An id that is not a
clone in the list is a stock voice and is used as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* fix(skills): an id missing from the voices list is not a stock voice
`media-use voices` returns one page of private voices (20 by default), so a
teammate's clone can be absent from it. The remix guide treated an absent id
as stock and used it without asking. Now a given id or prefix is used only
when the person says it is theirs, that its owner agreed, or that it is a
HeyGen stock voice; otherwise the agent asks in its one round of questions.
The `whose` line says the list is one page and absence proves nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): cloud render logs a line a minute while a job holds its status
Outside a terminal, `cloud render` printed one line per status change, so a 3D shot
that renders for 20 minutes showed "rendering 2.1s" and nothing else. Agents read
that as a hang and killed jobs that later completed. The log now repeats the status
with its elapsed time once a minute, and the style-sampling remix guide says the
cloud gives no frame count and how to check a job the client stopped waiting for.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* chore(skills): regenerate the skills manifest
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
* refactor(cli): keep the still-running interval module-private
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXPiLcYKQfvPWB3Fgxy
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): installing Chrome on a busy Windows PC no longer takes ten minutes
* fix(cli): the priority reason lives in the PR, one helper sets either priority
* feat(core): auto motion blur draws with the filter tier
Capture stays available as data-motion-blur-tier="capture". auto now draws with the
filter tier (reasonCode tier-auto), so the capture-only defects leave the default path.
Removes the translation-only hand-off and its Discovery field, adds a filter arm to the
differential, and updates the check, guide, skill and Blur panel label.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* docs(core): measured cost and known limits of the filter tier, the new auto default
spacex-launch (77 movers) takes 2.9x the CPU of no blur on the filter tier and 1.7x on
capture; say so, and list the filter tier's measured limits (sweep D, E, F, I, O, B).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* docs(core): word the filter tier's cost and limits as measured, not as mechanism
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* docs(core): do not say every filter-tier limit is pinned by a sweep test
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* test(engine): split the blur differential so each file fits the flake gate's 120 s run
The filter arm moves to its own file and skips fixtures that already draw with the filter
tier; inside cases get their own file; a fixture's blur-off page is opened once per file.
Shared helpers live in __fixtures__/motionBlur/differentialPage.ts.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* feat(cli): a line_unread finding carries the frames it measured around where the line was first read
* style(cli): shorten the measured field's comment
* feat(cli): measured frames at both ends of the line, with fewer comment lines
* fix(cli): a font missing from this computer no longer stops a cloud render, and is named
* style(cli): format the font preflight files and trim two new comments
* style(cli): the embedSystemFonts doc fits in two lines
* fix(cli): normalize-audio says where the target stays when a limit holds its gain
* fix(cli): normalize-audio held line counts the clip's data-volume in both modes
* style(cli): format the normalize-audio held-line change
* chore: release v0.9.17
* chore(producer): raise the unpacked size budget for windowed video frames
The new modules and this version bump put the unpacked package 49 bytes
past the 1% tolerance. No tests or source maps are in the tarball.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): normalize-audio --reference stops short of the peak ceiling instead of refusing
* docs(cli): the README says normalize-audio stops short of the ceilings instead of refusing
* fix(cli): end a stalled cloud upload after 60 s and send it again up to three times
* fix(cli): bound the in-memory project zip upload by progress like a file, watch the socket
* refactor(cli): split the upload retry and stall watch into small functions
* fix(core): snapshot and preview show a clip on the frame render shows it
* fix(cli): snapshot pulls a video's frame on the same on-screen rule
* refactor(cli): drop two doc lines the names already carry
* fix(cli): an unknown command answers in a few lines and says what to run instead
* fix(cli): shorten the captions pointer and drop the redirect doc comments
* fix(cli): point the captions hint at the hyperframes skill's guide, and test the path exists
* chore(cli): regenerate the skills manifest for the hyperframes-cli skill edit
* feat(cli): check --budget stops the whole check in time and reports what it found
* fix(cli): keep check's comment share where it was
* fix(cli): a --budget that runs out before any check step is waiting is not an unhandled rejection
* feat(cli): info --media lists every clip the render hears, on the film's clock
* fix(cli): info --media leaves out a clip that starts past its media's end
* fix(cli): keep the heard-clips doc line within 120 characters
* fix(core): hlg footage tone-maps to sdr at the bt.2408 reference white
* fix(core): an HLG clip whose stream leaves its transfer unset keeps HLG's white
An SPS with an unspecified transfer takes the stream's tag, as its frames do; a failed keyframe read is not cached; the grey test needs ffmpeg 8.
Two holes in `billing subscribe` / `billing add-credits`:
- After paying on the pricing page, "Press Enter to pay" was still waiting. Pressing it before the
next 2-second poll opened a second Stripe checkout. Enter now reads the plan again: nothing opens
when the page has already covered it, and a partial top-up on the page is priced from the new
balance.
- One failed first plan read made the account look like Free with 0 credits: it offered Pro to a
Pro account and priced add-on credits for the whole run, not the shortfall. The first read is
retried; if the plan still can't be read, no terminal checkout opens and the page finishes it.
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): every settings file follows HYPERFRAMES_CONFIG_DIR
* fix(cli): the media-use skill copy reads its config from HYPERFRAMES_CONFIG_DIR too
* test(cli): expect the config folder as a normalized path on every platform
* chore(skills): refresh the media-use manifest hash
* fix(cli): a moved settings folder starts from ~/.hyperframes, keeping an opt-out
* fix(cli): drop the updater's unused path import
* style(cli): keep the settings-seed comment under 120 characters
* fix(cli): a preview's session follows HYPERFRAMES_CACHE_DIR when it is set
* docs(cli): background previews follow HYPERFRAMES_CACHE_DIR
* test(cli): unstub the preview test's env in a block body
* test(cli): clear HYPERFRAMES_CACHE_DIR in the CLI's test home
* fix(cli): registry cache, search model, catalog and catch-up follow HYPERFRAMES_CACHE_DIR
* style(cli): format the media search cache test
* test(cli): compare the model path against an absolute cache folder on Windows too
* fix(cli): both catalog readers treat an empty HYPERFRAMES_CATALOG_ARTIFACT_DIR as unset
* fix(cli): capture takes a refused screenshot once more before naming it
* fix(cli): keep the screenshot retry's doc to one line
* test(cli): refuse the same capture shot twice now that one refusal is retried
* fix(engine): headline a pending setup as a setup timeout, not zero duration
When the runtime never reaches render-ready, __hf.duration reads 0 because
the bridge masks it, so pollHfReady's timeout reported "Composition has zero
duration" even for compositions with a declared data-duration whose
window.__hf.buildReady promise was still pending. Headline that case as
"Composition setup did not finish within <timeout>ms", skip the duration
inference hint that cannot apply yet, and note that the render clock stays
at 0 until the first frame. Keep its transient retry classification and give
it its own CLI diagnostic code.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(engine): name animation frames in the setup-timeout hint
Setup that awaits requestAnimationFrame hits the same render-clock rule as
setup that waits on Date or performance.now: frames only fire on a seek, and
the first seek waits for setup. Say so, and point at the waits that work.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(engine): stop suggesting scheduler.yield() for setup waits
Chrome runs scheduler.yield() continuations ahead of ordinary tasks, so a
setup loop that polls with it starves the setTimeout-driven work it waits on
(Three's compileAsync polls with setTimeout(10)) and never finishes. Plain
headless Chrome 152 hangs on a ten-line page that does exactly that. Point
authors at setTimeout and say why yield-polling does not work.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(engine): point setup waits at the library's readiness promise
HyperFrames guidance is no system timers, so the hint should not tell authors
to wait with setTimeout. Point them at the library's own readiness promise
(Three.js renderer.compileAsync) and keep the warning about scheduler.yield()
polling loops.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(engine): keep the setup-timeout hint to what it can diagnose
A scheduler.yield() polling loop starves the page, so the readiness poll never
returns and the render reports a capture stall instead of this message. Drop
that sentence; the hint covers animation-frame and clock waits only.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>