Behaviour is unchanged: on Free the plans page opens straight away, a
mid-run shortfall goes to add-on credits for the run's total, and cost is
never brought up unprompted. Lines like "without asking or naming a price"
and "never tell the person prices" read as hiding cost, so the skills now
say the page shows the price and payment happens only there, by the
person, and that cost is not volunteered.
Claude-Session: https://claude.ai/code/session_01HebQXQjEiSeVmhvTbJvSAw
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): the cancel route replies once the adapter's stop has settled
* fix(studio-server): shorten the cancel route comments
* fix(studio-server): drop the added cancel comments so the comment share stays down
* fix(studio-server): the cancel route clears its timer when the stop throws
* 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>
* feat(registry): add path-traced-3d example with raster preview and path-traced export
An opt-in Three.js starter: previews with raster rendering and exports with
three-gpu-pathtracer, from one scene file. three 0.180.0, three-gpu-pathtracer
0.0.24 and three-mesh-bvh 0.9.5 ship as one local, pinned ES module with their
licences and a rebuild recipe, so renders make no network requests and 2D
projects never load the tracer.
Setup is one buildReady promise that awaits library readiness promises only
and ends with a throwaway frame and a one-pixel readback. Frames are drawn in
the hf-seek listener through waitUntil, each time from zero with a fixed
sample count. Math.random is seeded while the scene and tracer are built,
because object ids order light sampling and the tracer's noise textures come
from it.
Adds a skill reference and a docs guide for the setup and seek contract, the
frozen clock, why scheduler.yield() waits hang, and sample/bounce cost.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(registry): list path-traced-3d in CLI docs and tighten its notes
Add the example's section to the CLI examples doc, which docs.test.ts
requires for every registry example. Throw if the tracer reports compiling
but exposes no compile promise, instead of spinning renderSample(). Say that
three's object uuids set the tracer's sort order, that later-created objects
need the seed too, and that Date and performance.now read the current seek
time during a render rather than staying at 0.
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(core): blur scan and eventful reads put back the DOM a timeline call changed
With data-motion-blur-events on, the mover scan runs the timeline to the end with
callbacks on and rewinds with callbacks on. GSAP fires a tl.call on a backward
crossing too, so the rewind left the earliest call's DOM effect in place: the
callbacks fixture showed label "B" from frame 0, before its 0.5 s call.
Calls cannot be un-fired, so the scan (and eventful per-frame reads) now journal
DOM mutations with a MutationObserver, land on the frame with callbacks on as
before (keeping JS-side call parity), then undo the journal back to the
pre-scan DOM, which is what a plain seek to the frame produced.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* test(engine,producer): callbacks label-early pins become ordinary outside cases
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* perf(core): the DOM undo skips attributes and text already back at their old value
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* fix(cli): billing stops checkout work once the payment wait ends, and a token refresh can't hang it
The wait for a purchase ends when the account shows it or at --timeout.
Checkout work still in flight then kept going: a price read that answered
late still created and printed a checkout, an Enter-time checkout that
answered late still opened Stripe, and the command waited on that work
past the purchase or the deadline. A token refresh with no deadline could
hold any of these reads forever.
readPlan, readOffer and createCheckout take the wait's stop signal, joined
with their own deadline, and stop waiting on a sign-in refresh once it
fires. Billing checks it after the price read and after each checkout
before printing or opening anything. The token refresh POST gives up
after 20 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
* refactor(cli): billing's checkout failure and its two billing reads each live in one place
The comment-share ratchet and the fallow audit failed on #7216: a new comment
in billing.ts, readPlan and readOffer repeating the same eight-line request,
and offerCheckout over the complexity threshold. The reads share getBilling,
the failure report moves to reportCheckoutFailure, and the comment goes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
* fix(cli): billing's stop signal fires at --timeout, not only when the poll returns
The checkout path stopped once the polling task ended, but the poll's
readPlan carried no signal. A poll read stalled across the deadline kept the
signal live, so a held offer could still create and print a checkout after
--timeout. A timer now aborts the controller at the deadline on its own, the
initial and polling plan reads take the same signal, and the deadline counts
from the start of the command.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(cli): a paywall refusal carries its page link in its own url field
An agent hit the Pro paywall, found the link only inside the fix sentence,
and handed the person a shell command instead of the page. The link now
also comes as url.
* fix(cli): auth status says hyperframes is the plan, not the HeyGen app's
An agent read user.subscription.plan (the HeyGen app's pro) as the
HyperFrames plan and ran past Free. With the HyperFrames plan read,
billing_notice now names which is which.
* fix(cli): a paywall refusal with url null names no purchase link
HeyGen sends url: null when it can't sell to this account right now. The
refusal now says so and carries no link, as reverse start does since #7031;
a missing url (older server) still gets the plans or credits page.
* fix(cli): reverse start makes its --dir project, and init takes the video already in it
Bug bash: production has no reverse/start route yet, so every run came back unrecorded,
and reverse start then made no folder and no hyperframes.json. The agent downloaded the
video into the folder it named, and init refused it as not empty, so agents moved files
around. reverse start now makes --dir and stamps it with the skill when no session is
recorded, and init accepts a folder whose only other files are its --video or --audio.
* refactor(cli): say in names what the comments said, so the comment share holds
* fix(cli): init keeps an in-folder source safe, and saved_to only names a stamped config
- Only an audio or video file in the folder counts as the source; a file named
index.html, meta.json or AGENTS.md is refused, so the scaffold never writes over it.
- An in-folder .mp4 that needs transcoding goes to <name>.h264.mp4 beside it,
instead of being both ffmpeg's input and output.
- reverse start names saved_to only when hyperframes.json now holds the skill;
a broken config is left as it was and saved_to is null.
* fix(cli): init transcodes an in-folder .MP4 beside it on case-insensitive disks too
original.MP4 and original.mp4 are one file on macOS and Windows, so the
same-file check now ignores case. The test's ffprobe call gets its --.
* fix(cli): init tells the source in the folder by the file, not its spelling
On a case-blind disk --video original.mp4 names Original.MP4, so the folder
check and the copy now compare device and inode; a source named in another
case is accepted and never copied onto itself.
* fix(cli): init transcodes only to a name with nothing at it, so a linked alias is never the output
The folder guard admits every alias of the source by file identity, but the
transcode output was chosen by path spelling. A hard link or symlink at
name.mp4 or name.h264.mp4, or a source named through a linked folder, made
FFmpeg (-y) rewrite the original. The output is now the first of name.mp4,
name.h264.mp4, name.h264-2.mp4, ... with nothing at its path (lstat, so a link
counts), which also covers a case-only spelling on a case-blind disk.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A crop on yuv420p rounds an odd size or offset down to even. readClip
split the 32x32 output as 33x33 bytes, so a 10-frame clip came back as
9.4 frames and every frame after the first was misaligned; an odd offset
also shifted the chroma. Raw reads now convert to rgb24 (or gray) before
the crop, which is exact at any size and place. extract and diff stay on
yuv and add exact=1.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Adds 13 text-free fixtures from the 60-composition capture motion blur
sweep and one test per defect, each pinned by its current symptom so it
fails once the fix lands. Pins use the existing outside and inside
invariants where they see the defect, and a second check against the
engine's 32-sample sub-frame blur where they do not (E, J, K, M).
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(core): a scrolled list keeps its offset when blur moves the root
Chrome resets a scrolled container's offset when the root leaves or enters the
document, so the first blurred frame of a fresh page or render worker drew the
list at offset 0. Offsets are read at move time and put back before the re-seek,
so a container that became scrollable after the scan is covered too.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* test(engine): scroll-list is an ordinary outside case, not a pinned known bug
The scroll offset now survives the root's move, so the three engine differential
cases and the multi-worker render case expect blur on to equal blur off, and the
measured constants for the old symptom go.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* chore(core): drop the scroll-offset comment, the comment-share ratchet may only fall
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* docs: download links name where the visit began
Docs download links carry the visit's source (ref, utm_*, ad-click kind), like the
homepage's links. Mintlify's Card download button now carries them and the PostHog id
it was dropping.
* docs: keep the attribution comment to public names
* docs: take over only the Card's click, so Mintlify's own link handlers still run
Stopping every download click hid the navbar CTA's docs.navitem.cta_click. Only Mintlify's
Card (data-client-link) ignores a changed href; every other link is followed by the browser
from the tagged href, as before. A source restored from storage carries only its own names.
* fix(cli): check skips a contrast sample measured over a layer that is mid-motion
* fix(cli): move the backdrop overlap read out of sampleOf, and add no comment lines
* fix(cli): shorten the measuredInMotion comment
* feat(studio): alt marquee removes the covered items from the selection
Shift already adds marquee hits to the selection on the canvas and the
timeline; there was no way to take a few items back out of a large
selection. Both marquee paths now carry a MarqueeSelectionMode
(replace | add | subtract) resolved from the pointer modifiers in one
helper, with Alt winning over Shift. Subtracting the primary re-anchors
on a surviving member; subtracting everything clears the selection. The
band and candidate outlines turn the danger colour while subtracting.
Closes#5586
* fix(studio): same alt marquee rules on the canvas and the timeline
Both surfaces now let the modifiers held at release decide: the timeline
re-applies the marquee from the pointerup event before it finishes, so
letting go of Alt mid-drag turns a subtract back into a replace, as the
canvas already did. Alt at pointerdown no longer bakes a timeline base
mode; Cmd/Ctrl still sets add. While subtracting, the canvas outlines
only candidates that are in the current selection, so the band never
suggests removing something that is not selected. The timeline commit
helpers move to timelineMarqueeCommit.ts to keep the hook under the
file-size cap.
Some hosts drop a connection to a closed loopback port instead of refusing it
(WSL 2 mirrored networking, a firewall in stealth mode). The Vite Studio probe
on 127.0.0.1:5190 then waits for fetch's 10 s connect timeout, and since
every render, check and snapshot looks for a Studio through the history owner,
each of them paid 10 s. Aborting the fetch does not help: the connect attempt
keeps its socket, and the process, alive for the full 10 s.
The default probe now connects with node:net under a 1 s bound and destroys
the socket, then fetches only from a host that accepted.
Co-authored-by: Whoosh <nhatkha1407@gmail.com>
A piped, non-quiet local render now writes one versioned @hf-result JSON line
after success. Batch manifest rows carry the same reference_background value.
A rejected window.__hf.buildReady hold now sets window.__hfStartupError naming the key and the
reason, so render, snapshot and check fail instead of capturing a scene that was never built.
Fixes#5603
* fix(cli): a browser install goes on when Windows holds a file for a moment
* chore(cli): shorten the lock-retry comment
* fix(cli): keep the lock check private
* chore(cli): wrap the lock-check comment
* fix(producer): pad/trim audio from the delivery PCM, not its own AAC
The assemble stage's duration fix decoded audio.m4a and encoded it again.
That second AAC generation read about 2 dB hotter on a measured mix
(-1.8 dBTP out of the audio stage, +0.2 after the trim), so the whole-mix
cut then lowered an already compliant film by 3.4 dB, overshooting to
-2.5 dBTP. A sibling project with a slightly louder music bed escaped the
cut, so the two landed about 4 LU apart.
- The local render keeps the audio stage's PCM in the work dir until
assembly; pad/trim and any peak correction there encode from it.
- A whole-mix cut in the audio stage is applied to that PCM too, so it
still encodes to exactly the delivered AAC. If it cannot be rewritten
it is removed and assembly falls back to the AAC.
- The distributed plan does not keep the PCM, so its plan dir is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(producer): name the assemble stage's PCM input by what it is
The comment-ratchet does not let assembleStage.ts gain a comment line, so
the new input says what it holds in its name instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
* fix(producer): assembly never reuses a PCM whose delivered cut was not applied
When the audio stage cuts the whole mix and then fails to rewrite the kept
PCM, it tries to delete the PCM. Deletion is best-effort and can be refused,
for example on Windows, so the uncut PCM could stay on disk. The audio stage
handed it to assembly because the file still existed, and assembly would
then re-encode a mix louder than the AAC that was actually delivered.
The cut now reports when the PCM no longer matches the delivered AAC. The
audio stage keeps the PCM only on an explicit success, not because the file
exists, and assembly otherwise falls back to the verified AAC.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
* test(producer): the fake-FFmpeg PCM rewrite writes its output
The true-peak correction now rewrites the kept PCM with FFmpeg and reports
the PCM as stale when that output never appears. The fake FFmpeg in this
test answered the rewrite without writing anything, so a correct run looked
like a failed rewrite. The fake now writes the output, as real FFmpeg does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HebQXN7wsMK6GqT3MNtqXw
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(core): data-motion-blur-phase sets when the shutter opens
Page motion blur always centres the shutter on the frame. Add a phase in degrees of a
frame, -360 to 360: the window starts at t + phase / 360 / fps and lasts shutter / 360 /
fps. Unset stays centred, so a page without the attribute renders the same bytes.
Both tiers take the window from one place, shutterTimes in frames.ts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(core): the motion blur report carries the shutter phase
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* chore(core): drop the phaseDeg doc comment, the comment-share ratchet may only fall
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(skills): every published skill reports its own run
`hyperframes events --skill=<name>` exists so a skill can say it ran, but
only reverse-any-video and figma asked the agent to call it, so the other
skills' use was invisible. Each of the nine published skills now opens with
a one-line usage beacon naming itself. The line says what it sends (the
skill's name only) and that the telemetry opt-out applies. The router's
`init` line also passes `--skill=hyperframes`, so renders of new projects
carry the skill.
check-core-skills fails a core skill that lacks its own beacon; each core
skill's size budget grows by 192 B for the line.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(cli): checkout links name the CLI and the skill that sent them
Only `billing subscribe` marked its pricing link `from=cli`, and the
terminal's Press-Enter Pro checkout sent no return page, so a purchase made
from the terminal looked like one made on the site. Now:
- `pageUrl` adds `utm_source=cli` and `skill`: the one named, else the
project's own `authoringSkill` from hyperframes.json.
- `billing` takes `--skill`. The terminal Pro checkout returns to the plans
page marked `from=cli&skill=...`, and retries without the return page if
the server does not allow it.
- Pricing links the server hands back (reverse's upgrade link, a refused
media call's link) gain `from=cli` and the skill when they lack them.
- The Pro gates name their skill: reverse -> reverse-any-video, templates ->
style-sampling, paid media calls -> media-use, in the CLI's printed fix and
in those skills' billing commands.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): report a refused checkout return page, and say where --skill goes
When the server refuses the terminal checkout's return page, the CLI still
sells Pro without it, but that purchase loses its CLI origin. A
cli_billing_checkout_return_refused event now makes that visible.
The --skill help said the skill is recorded with the purchase. It is sent
with the pricing link; the purchase itself does not store it yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(skills): every published skill reports its own run
`hyperframes events --skill=<name>` exists so a skill can say it ran, but
only reverse-any-video and figma asked the agent to call it, so the other
skills' use was invisible. Each of the nine published skills now opens with
a one-line usage beacon naming itself. The line says what it sends (the
skill's name only) and that the telemetry opt-out applies. The router's
`init` line also passes `--skill=hyperframes`, so renders of new projects
carry the skill.
check-core-skills fails a core skill that lacks its own beacon; each core
skill's size budget grows by 192 B for the line.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(skills): say the usage beacon goes out with the CLI's usual telemetry
The beacon line said it sends only the skill's name. The events command uses
the CLI's normal telemetry, which also carries machine details, the install
id and, when signed in, the account. The line now says so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
README links to hyperframes.studio, hyperframes.heygen.com and heygen.com
carry utm_source=github&utm_medium=readme, the values the README's download
links already use. Three README links went through docs redirects that drop
the query string (/showcase, /guides/claude-design, /guides/open-design);
they now point at the pages themselves.
The docs navbar's Community and changelog links carry src=docs-navbar, and
the export page's reinstall link src=docs-export, beside the existing docs-*
tags.
Links the CLI prints to our sites (init's docs links, the Lambda docs, the
publish claim URL, the HeyGen API-key page) carry utm_source=cli. Discord
invites stay untagged.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): tell a failed billing checkout apart from a payment not finished in time
* fix(cli): keep waiting for payment after a failed billing checkout
* refactor(cli): split billing checkout opening out of offerCheckout
* fix(fonts): stop failing cloud renders on var() shorthands, font-X-400 names and linked font sheets
* style(producer): shorten the project font sheet comment
* style(core): shorten the field font families test header
* fix(fonts): split font size parsing and fit the CLI size budget
* fix(producer): read project font sheets for family names only, never fetch from them
* refactor(cli): remove OpenRouter as a vision provider for capture
* refactor(cli): tidy comment and formatting after OpenRouter removal
* refactor(cli): shorten the thinking-budget comment
* chore(skills): refresh the skills manifest for the film reference
* docs(catalog): every card on screen plays a small looping video instead of a live player
* docs(catalog): the preview generator also writes each card's small loop and poster
* docs(catalog): a search box over the cards; it fills the ?q= filter the gallery already reads
* fix(catalog): end ffprobe options before the card file, as the argv contract requires
* fix(catalog): a pause control for the card loops; none play in a hidden tab or under 25% visible
* docs: say HyperFrames is free, and what the Studio app's Pro plan adds
* docs(readme): the app picture's alt text names windows too
* docs(faq): the free plan's exact monthly numbers, from the live pricing page
At 900 px the root composition card sat on the edge of the list's thumbnail margin, so a small layout change between runner images decided whether it asked for a thumbnail. A 1400 px window puts every card well inside it.
* fix(cli): check stops after 2 minutes of video copies and says how to finish
* fix(cli): check copies only the stretch of a video the composition plays
* docs(cli): check copies only the played stretch of a long video
* docs(cli): shorten two comments on the copy window
* fix(cli): finding crops and frames get the same copy windows as the check
* fix(cli): drop a comment the ratchet counts
* refactor(studio-server): keep runFfmpeg under the complexity gate
* fix(cli): check thins its craft samples when frames are slow, so the pass ends near 90 seconds
* style(cli): format the craft loop
* fix(cli): cap the craft stride so the widened grid stays under the dead-stretch threshold
* fix(engine): a background image that never finishes decoding no longer stalls the frame
* fix(engine): capture tolerates a page that returns no list of late background images
* fix(engine): a background decode timeout is reported at every caller, url redacted
* fix(engine): a relative background url keeps only its path in the decode timeout report
The runner image ships Chrome 156.0.8078.47 or .79 run to run, and .79 renders one more thumbnail on open, so the studio-open ceilings failed on changes that never touched Studio. The ceilings file now records the major it is measured on.