* fix(studio): a preview reload survives its pane unmounting mid-load
When the preview pane unmounted while a shadow reload was loading, its
Player removed the shadow frame but the reload kept timing it, so the
edit was dropped with "Its page had got to: no-page" and the previous
preview stayed. The reload now pauses while its frame is off the page
and starts its budgets again when the pane mounts a new frame.
The Player's comment on callback refs said the mutable-ref branch was
the only one in use; the shadow reload passes a callback ref, so the
comment now says why those are left alone.
* refactor(studio): name the left-page flag instead of explaining it in comments
* style(studio): format the shadow reattach guard
Ten rules each parsed the same <style> block with postcss, so a large
style block cost ten parses and ten trees for the garbage collector.
The 40 000-rule linear-time test ran close to its 2 s limit on CI and
failed there by chance. One shared parse per block brings that lint
from about 400 ms to about 60 ms.
* fix(core): show a paused video frame with the same color as the playing video on Mac
On macOS Chromium the video layer shows 8-bit SDR video with Apple's video gamma while a canvas shows the decoded pixels as sRGB, so a paused frame looked lighter-to-darker than the playing video. One module now owns the conversion and the paused stand-in canvas is displayed through it.
* fix(core): drop a comment line to keep the file's comment share down
* fix(core): correct a paused frame only for a source the scan reports as 8-bit, on both hold paths
Source facts now come from the shared media lookup (origin-aware, exact then longest suffix then normalized), unknown depth is left alone, and the native-only seek hold gets the same display helper.
* fix(core): shorten two comments to the line limit
* feat(studio): record the layer name on each history entry's target
* feat(studio): record the clip's name on timeline move and resize entries
* chore(studio): drop doc comments the names already carry
The media panel's Volume slider wrote a keyframe at the playhead, so a clip with a fade (or none) changed only around where the playhead sat. Volume now sets the clip's gain, the whole fade moves with it, and the keyframe toggle still adds automation on purpose. The separate Gain slider is folded into it, and the Audio Gain dialog no longer refuses clips that have a volume lane.
Adds minimax-h3-image (2K) and minimax-h3-image-4k to media-use generate,
refuses aspect ratios and reference counts each /v3/images model does not
take, and re-bases every image price on the live rows: an API key pays 1.5x
the vendor price, a HyperFrames plan the hf_image rows at x1.6545.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): publish keeps Library files out for projects on iCloud Drive
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): treat Library/CloudStorage cloud drives like iCloud Drive when publishing
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(producer): render servers only load assets from inside the project
The HTTP render handlers and the distributed planner now compile with
outside-project imports off: markup that reaches a file beyond the project
directory (through ../, an absolute path, a file: URL or a symlink) is no
longer copied into the compiled directory, served to the page, probed or
extracted. Local `hyperframes render`, `--start/--end` range renders,
`hyperframes benchmark` and Studio renders pass allowOutsideProjectAssets
and keep importing outside files exactly as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(producer): render the symlinked-assets fixture as a trusted local render
The render-symlinked-assets fixture links src/shared to a sibling folder
outside the project to keep local renders working like preview. The
regression harness now reads renderConfig.allowOutsideProjectAssets from a
fixture's meta.json and passes it to the in-process and
distributed-simulated renders, and this fixture sets it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(producer): render servers read compositions and media only from inside the project
Without the trusted-local opt-in, a sub-composition or entry file that
leads outside the project once symlinks resolve is refused, the same as a
../ sub-composition already is. Local media that resolves outside the
project, compiled and download folders, including absolute paths, file:
URLs and percent-encoded ../ paths, is reported as a missing source and
never probed or extracted. Remote media fetched during the render keeps
working. A missing file is no longer probed for the compile-time advice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(producer): raise the unpacked size budget for the project-folder checks
The producer tarball grows by about 12.5 KB unpacked: the new checks live
in the compiler and render stages, which the index, public-server and
distributed bundles each include. No test files ship in the package.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* chore: release v0.9.37
* docs(release): add #7576 to v0.9.37 notes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): publish leaves out private files outside the project and says so
A project you open may come from someone else, and its markup can point at
files outside the project folder. Publish and cloud render now leave out an
outside file that sits in a hidden folder (any path segment starting with a
dot past the folders the project itself sits in) or under /etc, /proc, /sys
or /dev, checked both as written and after resolving links.
Whenever an outside reference is left out, for this reason or because it is
an env file or Git data, the CLI now prints which project file holds the
reference and the reference as written. Other outside references are copied
into _ext/ as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): feedback's repro publish also names files it leaves out
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): publish also leaves out project links to private outside files, and AppData/Library
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): only the home folder's AppData and Library count as private
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): a referenced project link to a private outside file stays out too
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): resolve both paths the same way and skip AppData/Library when the project sits there
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): templates pull unpacks only plain files and folders
`templates pull` already refuses entries that climb out of the target
folder. It now also reads the archive's long listing and refuses any entry
that isn't a plain file or folder (a symlink, hard link or device), the
same rule Desktop applies when it unpacks example films and templates.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(cli): name the templates pull entry-type check instead of commenting it
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
0.9.35 made snapshotProject and snapshotFiles fall back to scratch-only
exclusions when .hyperframesignore can't be read. That is right for change
detection, but a host that builds an archive from the same snapshot then
includes files the ignore file meant to exclude, including ones named on its
valid lines.
The snapshot is strict again by default and throws invalid_source_ignore.
Change detection opts in with { lenientIgnore: true }, and projectFingerprint,
which names no files, always reads leniently.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The audio-missing check resolved every <audio src> on disk, including a
src naming a share on another host (\\host\share, //host/share). On
Windows that lookup opens a connection to that host, and Studio runs lint
as soon as a project opens. The check now skips a src whose root is a
network share other than the project's own, as the other lint rules
already do. Media on the project's own share, local absolute paths and
project files are checked as before; the render still reports a file it
cannot find.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(core): motion blur preview draws fewer filter copies at lower preview quality
The filter tier's preview draws 16 copies at preview quality 1/2 and 8 at 1/4 and
below; renders keep 32. Preview facts now list those levels in lighterAt once the
filter tier has movers, so a host that steps quality down on choppy playback can do
so for a film whose only load is blur.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(core): name the blur preview copies instead of commenting them
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* chore: release v0.9.35
* docs(release): note the render stop and container 403 from #7563 in v0.9.35
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix: track local Studio sharing CTA intent
* fix: tag local Studio and CLI share upgrade links with where they came from
The local Studio share modal's Upgrade to Pro link opens
/pricing?from=cli&entry=local_studio_share, and hyperframes share's source-files
notice links /pricing?from=cli&entry=cli_share, so the pricing page can tell
which link a visit came from and return the visitor to their terminal. The
modal accepts its own pricing page with any query, still only on
hyperframes.dev's two origins. share_upgrade_clicked carries
cta=share_modal_upgrade.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor: name the local Studio upgrade CTA and drop the explaining comments
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix: keep the local Studio upgrade link's return page unchanged
Only entry=local_studio_share is added to the local Studio share window's
Upgrade to Pro link. from=cli would have told these buyers to go back to their
terminal after checkout, but they return to a browser tab. hyperframes share's
link keeps from=cli, as the CLI's other upgrade links do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(share): keep change detection working with a gitignore-style .hyperframesignore
The project scan used by change detection now reads .hyperframesignore
through the strict source reader. Cloud render reads the same file as
gitignore syntax and its docs show /exports/, which the strict reader
refuses, so the scan threw and hosts lost change detection, export
freshness and share status for that project.
Change detection now falls back to scratch-only exclusions when the file
cannot be read. Source capture keeps the strict reader and still refuses it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(share): check snapshotFiles tracks the file, not just names it
snapshotFiles returns every requested path, keying an excluded one as
undefined, so key presence alone did not prove the file was tracked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(core): keep git metadata and env files out of bundled HTML
The bundler now checks every file it reads against the project's private
paths (git metadata and env files, spelled directly or through a link)
and leaves a refused reference as authored instead of inlining it.
Bundle options gain a canRead predicate; Studio's preview passes its own
private-file rule so the preview document matches what the file routes
serve.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): check capture read grants against the decoded path
A capture page's read grant now looks at the path the way the static
preview route decodes it, including encoded separators, so a URL that
resolves to a dotfile or into a dot-folder gets no CORS grant. Ordinary
project files keep loading in thumbnails.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): accept local share requests only from loopback connections
The local share routes already require loopback Host and same-origin
headers. They now also require the connection itself to come from this
machine, so a Studio bound to a network address still keeps sharing
local to the person running it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): ask the bundler for Studio's private-file rule as plain data
The preview route passed a fresh `canRead` callback on every build. A host that builds on a
worker thread cannot send a function across, so it built on its own thread, and it could no
longer share one build between two same-content requests.
The bundler now takes `privateFiles: "studio"`, a plain string, and applies Studio's
private-file rule itself. That rule (git metadata, env files, and the app's records under
`.hyperframes/`) now lives in `@hyperframes/core/safe-path`, and studio-server reuses it.
The `canRead` option is removed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): do not serve a preview main file linked to a private file
When index.html (or <id>.html) is a link to an env file or into git metadata, the
bundler refuses it and the route falls back to reading the main file from disk. That
fallback now skips a candidate that is a private project file, by the same rule the
file routes use. A main file linked from outside the project is still served.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(producer): never inline git or env files as render sub-compositions
The render compiler resolved data-composition-src and inlined the file it named, so
`data-composition-src=".env"` put the env file's text into the render page. It now
refuses git metadata and env files with the same shared rule the bundler uses. The
pre-render check names the refused reference and stops the render.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore: raise the inspector unpacked size budget to 3,722,000 bytes
The inspector already sat about 36.8 KB over its 3,684,092 budget, just inside the 1%
tolerance. Moving Studio's private-file rule into core added a few hundred bytes to the
bundles it ships, which pushed it to 3,721,419 bytes and over the tolerance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(parsers): find helper programs by walking PATH in-process
The trusted lookup ran `which` by name through the full PATH before
applying the trusted-folder rule, so a relative or project-local PATH
entry could supply the lookup program itself. Walk the trusted PATH
folders in-process on every platform, as Windows already did, for
findTrustedExecutable, findFfBinary and the doctor's unzip check.
The walk keeps what `which` did for valid installs: first match in PATH
order, regular files only, and a leading `~` in a PATH entry read as the
home folder.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): start the media-use engine with the Windows working-directory lookup off
mediaUseEngineEnv now sets NoDefaultCurrentDirectoryInExePath for the
engine on Windows itself instead of relying on the CLI's own environment,
and keeps a value the person set. When no PATH folder passes the trusted
rule, PATH names a folder that is never created rather than an empty
value, which a shell or execvp reads as the working directory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(skills): helper scripts skip the Windows working-directory program lookup
On Windows a program started by name is looked up in the working
directory before PATH. The plugin launcher and the media-use, hyperframes
and figma helper scripts run inside project folders and start ffmpeg,
ffprobe, Python and other tools by name, either through the launcher or
directly from a standalone skill install.
Each skill now ships a small module that sets
NoDefaultCurrentDirectoryInExePath for the process (keeping a value the
person set), and every script that starts programs imports it first, so
the helpers and every program they start skip the working directory.
Tools installed on PATH resolve as before; macOS and Linux are
unaffected. A test checks that each such script loads the module.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): ship the Windows lookup guard with the CLI's media-use scripts
The CLI build replaces the published media-use scripts/lib folder with
packages/cli/src/media-use/lib. That folder had no windows-exe-search.mjs,
so every bundled media-use script that imports it, including the engine
behind `hyperframes media-use`, failed to load on every platform.
The guard is now a generated copy in the CLI lib, next to the skill copies.
A parity test checks that each scripts/lib module a published media-use
script imports exists in the CLI lib.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(skills): load the win32 stub by file URL so the launcher test runs on Windows
On Windows, `node --import D:\...` reads the drive letter as a URL scheme
and fails before the launcher starts. The test now passes a file URL and
reports the child's stderr when it exits non-zero.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(skills): resolve the Python helper's ffmpeg and drop relative PATH entries
extract-audio-data.py ran ffmpeg by bare name, which Windows looks up in
the working directory first. It now runs the absolute path that
trusted_program.find_program finds in absolute PATH folders outside the
working directory. A miss raises the same FileNotFoundError as before.
On macOS and Linux the shared guard now drops relative and empty PATH
entries, which point into the project folder, and leaves absolute ones in
order. If none is absolute, PATH is unset rather than left empty.
The guard census now also requires every skill Python script that imports
subprocess to resolve programs through trusted_program.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(skills): satisfy the Python UTF-8 and ffprobe argv sweeps
trusted_program.py carries the same UTF-8 stdio pin as every skill Python
script. The guard test's planted-program probe uses a literal-only argv,
which the ffprobe argv contract test reads as a capability check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
formatDuration rounded seconds after splitting off minutes, so 59.96s printed
"60.0s" and 119.97s printed "1m 60.0s". formatBytes compared KB against the MB
threshold before rounding, so 1,048,575 bytes printed "1024.0 KB". Round first,
then split or compare.
Takes over #5817 by @rudycelekli, whose commits could not be merged unsigned.
Co-authored-by: Rudy Celekli <rudy@gradiahq.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(core): give the preview parent a typed editing port to prepare for an opaque frame
Studio reads the preview frame's window and document directly, which stops
working once the preview runs in an opaque origin. This adds the runtime half
of the replacement, with no Studio consumer changed yet:
- runtime/editingPort.ts: a versioned request/reply protocol, a validator
that turns any bad request into a typed error, and a promise-based parent
client with request ids, timeouts and dispose. Exported as
@hyperframes/core/runtime/editing-port.
- runtime/editingAgent.ts: on the parent's connect, the runtime hands
window.parent a MessagePort once and answers hello, runtime-snapshot,
wait-for-seek and set-loop-start on it. A connect from anything but the
parent is ignored and does not use up the handoff.
Playback transport stays on the existing control protocol.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(core): type the handed-off port under the runtime's strict index checks
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(core): emit the editing-port module so the published subpath resolves
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(producer): raise the packed size budget for the runtime's editing agent
The producer ships the preview runtime, which now carries the editing agent
and its request validator. The packed tarball measured 878,763 bytes in CI
against an 870,000 budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): let opaque preview frames load project fonts and module scripts
A preview framed with an opaque origin, such as the Compositions hover
preview, makes its font and module-script loads in CORS mode with
`Origin: null`. Without a CORS grant they failed: the frame fell back to a
system font and module scripts never ran.
The preview asset route now answers those two loads with
`Access-Control-Allow-Origin: null`, `Vary: Origin` and
`X-Content-Type-Options: nosniff`. The grant is chosen by `Sec-Fetch-Dest`,
which the browser sets and page script cannot: `font` for a font file and
`script` for a .js/.mjs file, served with the matching content type. A
page's own `fetch()` is always `empty`, so it still gets no grant and
composition script still cannot read project files.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(studio-server): keep opaque-frame grant comments within the package share
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): let opaque preview frames load project fonts
Narrow the opaque-frame grant to fonts. Every sandboxed frame sends the
same `Origin: null` and fetch metadata, so a CORS grant for module
scripts could not be limited to frames the Studio itself serves. Module
scripts stay unloaded in the hover preview for now; fonts keep their
grant, and `Sec-Fetch-Dest: script` loads of .js/.mjs get none.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): load fonts and modules only in the Studio's own preview frames
A CORS grant to `Origin: null` reaches every sandboxed frame, whichever
page it sits in. The grant now lives on a per-project path instead:
`/projects/:id/preview/~<token>/...` serves exactly what `/preview/...`
serves, and only there do font loads (`Sec-Fetch-Dest: font`, a font file
served as `font/*`) and module-script loads (`script`, a .js/.mjs file
served as `text/javascript`) from an opaque frame get
`Access-Control-Allow-Origin: null`. The plain path gets no grant.
The token is an HMAC-SHA256 of the project id under a random per-server
secret, compared in constant time. The Studio reads it from
`/projects/:id/frame-token`, which refuses cross-site and same-site
requests, and the Compositions hover preview frames its composition at
the token path. Documents served there are re-based under the token and
carry `Referrer-Policy: no-referrer`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(studio-server): raise the packed size budget for the preview frame tokens
The per-project frame-token path adds server code to the package, and the
packed tarball measured 166941 bytes in CI against a 164800 budget. Set the
budget to the measured size plus about 1% headroom.
— Rames Jusso
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(skills): plugin launcher and media helpers run the pinned CLI, not a project copy
The plugin launcher now starts npx from the home folder, so the project's
node_modules, package.json bin and .npmrc cannot choose which CLI npm runs or
how it starts it. It names the project in HYPERFRAMES_PLUGIN_CWD and the CLI's
bin moves into that folder before anything else runs, then drops the variable
so nested runs keep the folder their caller gives them.
In --script mode the launcher hands itself over as HYPERFRAMES_CLI_ENTRY. The
media helpers (Kokoro voice, transcription, the resolve shim) run the CLI they
were handed, and stop with an actionable message when none was, instead of
resolving `npx hyperframes` from the project.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(skills): keep the project name and require a home folder for npx
The CLI bin now sets PWD to the project folder after moving into it, because
npm's shell resets PWD to the folder npx started in and preview names the
project from PWD. The plugin launcher no longer falls back to the temp folder
when there is no home folder; it stops with an error instead, and only when it
is about to start npx.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(cli): type the bin test environment
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Local Studio mode runs Vite in the resolved @hyperframes/studio folder, so the
project folder decided which config and binary ran. Use it only when the
project resolves to the same Studio the CLI itself resolves (a source checkout
running its own build) or when HYPERFRAMES_LOCAL_STUDIO=1 is set; otherwise
serve the embedded Studio.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli,parsers): take helper programs only from install locations, never the project folder
FFmpeg, FFprobe, whisper-cli, parakeet-mlx, Python and system Chrome are now
found through one shared rule: an absolute PATH entry outside the working
directory, or a fixed install dir. A relative PATH entry and any folder
inside the working directory are skipped, on every platform. Windows walks
PATH itself instead of asking `where`, which lists the working directory
first.
FFmpeg discovery no longer falls back to `.hyperframes/bin` or to an
installer package in the project's node_modules; an installer package
installed next to HyperFrames still works. The env overrides are unchanged.
The CLI skill reference stops suggesting the project-local folder.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(cli): give the PATH-lookup tests a fixed PATH on Windows
Windows now walks PATH instead of running `where`, so tests that faked
`where` output found the runner's real python and cmake. Both tests now set
PATH to a temp folder holding only the programs they expect.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix: report a missing helper program instead of running it by bare name
When no trusted FFmpeg or FFprobe is found, the engine, the producer's
probes and Studio's waveform, peak and audio-track helpers now report
"not found" with the HYPERFRAMES_FFMPEG_PATH / HYPERFRAMES_FFPROBE_PATH hint
instead of handing the bare name to the OS lookup. Callers that degraded
on a missing binary still do: runFfmpeg returns a spawn_error result, the
GPU and version probes report none, and Studio's audio helpers fail soft.
The Lambda handler names its bundled ffmpeg and ffprobe by path, since
they live under the task root, which is the working directory.
brew, git and cmake for the whisper.cpp install, and the clipboard tool
for `hyperframes add`, run by the absolute path the trusted lookup
approved. The media-use engine gets a PATH holding only trusted folders.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): fail a missing FFmpeg the way a failed spawn always has
Throwing from getFfmpegBinary turned every caller's missing-binary path
into a synchronous throw: the streaming encoder's "failure result, not a
throw" contract broke, and unit lanes that run without FFmpeg crashed at
module load. With no trusted copy the getters now return an absolute path
inside HyperFrames' own install that is never created, named for the
override to set. Spawning it fails with ENOENT exactly as before, and the
OS still never looks the bare name up. The call-site guards and fixture
changes that the throw needed are reverted.
The whisper install tests match git and cmake by name and skip the brew
case on Windows, where the lookup walks the real PATH.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(engine): pin the found FFmpeg before the HDR test looks it up as macOS on Windows
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(producer): fail a render whose variables the page does not declare
* chore(producer): keep the comment share and the dead-code audit clean for the variables check
Registry items may not write QWEN.md, CRUSH.md, WARP.md or CONVENTIONS.md at
any depth, in any case, alongside the AGENTS.md, CLAUDE.md and GEMINI.md
names already refused. The catalog's nested SKILL.md and README.md notes are
unaffected.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(core): fall back to a direct write when link() is denied with EACCES
On Android app data (Termux, SELinux enforcing), link(2) fails with
EACCES in a folder the temp file was just written to. createFileAtomically()
and the studio history owner lock treated that as fatal, so
`hyperframes init` left a half-scaffolded project and every render logged
a recordRenderSuccess failure. Both now take their existing exclusive
direct-write path, as they already do for EPERM.
Fixes#7473
* fix(share): treat EACCES as no hard links in the project share lock
Same Termux/SELinux link() denial as core and the studio history lock:
the share lock's link-then-exclusive-create fallback now covers it too.
Folds the Android note into the existing comment so no file's comment
share rises.
* fix(cli,studio-server): limit registry installs and Studio file writes to project content folders
Registry blocks and components now install only into the project's
compositions, components and assets folders (plus models/ and lib/, which
the catalog uses, and notes at the root). node_modules, agent instruction
files and programs or scripts are refused at any depth, matched
case-insensitively with either separator, and checked again on the path
the project's own links lead to.
Studio's file routes never read or write git metadata or env files, and
every route that writes, renames, deletes, copies or uploads a
request-named path refuses the project folder itself, node_modules and
hidden files or folders other than .hyperframes/ and .media/, on both the
spelled and the link-resolved path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(cli,studio-server): trim comments added by the write allowlist
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli,studio-server): trim path segments in linear time
The trailing dot/space and :stream trims used anchored regexes that
backtrack quadratically on long runs of spaces, dots or colons in a
request path. Trim with a plain scan instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): refuse example dependencies in add and allow only TEMPLATE.md at the root
- addToProject now refuses a block or component that depends on an example,
not only an example asked for by name. The installer lets an example write
at the project root, so a dependency must not carry that through.
- The only file a block may write at the project root is TEMPLATE.md, the
catalog's usage note. Any other .md or .txt name could be one a tool loads.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): run thumbnail captures in an opaque origin
The thumbnail renderer loads each composition as a top-level page, so its
scripts ran on the Studio's origin. The capture document (`?hf-capture`) now
carries `Content-Security-Policy: sandbox allow-scripts`, so those scripts get
an opaque origin.
An opaque page's CORS-mode loads fail without a CORS header. The renderer that
opened the page is trusted, so it adds `Access-Control-Allow-Origin: null`
itself, over that page's own DevTools session, to the loads a frame needs:
- fonts, module scripts and images;
- the `.html` scenes, `.json` and `.cube` files the runtime fetches.
Only GETs of the project's preview files qualify, matched by file type. Both
thumbnail renderers do this: the CLI's and the Vite dev server's.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): let an opaque capture wait for a video's preview copy
A video the browser cannot decode is drawn from its `?hf-proxy=` copy. The
runtime waits for that copy with a ranged fetch, preflighted when it carries
the priority header. In the opaque capture page both failed, so each such
thumbnail fell back to a still after about five seconds of retries.
The renderer now answers that preflight itself. It also exposes the copy's
status headers to the wait, for video files under the preview path only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(studio-server): keep capture-load comments within the package share
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): keep video pixels and project fetches in opaque thumbnail captures
- Media loads: the renderer now grants them, and a capture document loads the
project's own videos with `crossorigin="anonymous"`. Color grading and shader
textures can then read their pixels.
- Fetches: every GET of the project's preview files is granted, except a dotfile
or dot-folder path such as `.env` or `.git/`. This covers the models and fonts
that registry blocks fetch at runtime.
- Opt-in: the capture sandbox is now a `createStudioApi` option,
`sandboxThumbnailCaptures`, off by default. The CLI Studio server and the Vite
dev server turn it on, because their thumbnail renderers install the grant.
Another app keeps its current capture documents until its renderer does the
same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(shader-transitions): prepare transitions when IndexedDB is denied
An opaque-origin document, such as a sandboxed preview or thumbnail capture,
has no IndexedDB, and opening it throws. The snapshot cache now treats that
as no cache, so the transitions still prepare and draw.
Thumbnail captures also grant a CORS read of a missing preview file, so a
fetch() sees its status as a same-origin page would.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): leave a video with another host's <source> out of capture CORS
A video whose <source> child points to another host now keeps its plain
load in a thumbnail capture. Adding crossorigin to it would make a CDN
that sends no CORS headers refuse the video, leaving the clip blank.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The test checks that clips read from a URL are not held to the CPU-sized
prepare limit. On 32 or more cores that limit reaches the URL limit of 16,
so the test could not pass there. It now fixes the core count at 8.
* feat(cli): share rendered videos through unified reviews (#7135)
* feat(share): attach editable project sources to reviews (#7137)
* feat(share): measure local sharing outcomes (#7138)
* feat(cli): stage the sixty-day publish transition (#7136)
* feat(studio): share review pages from local preview (#7140)
Advertise the CLI sharing bridge before bootstrapping it so standalone Studio does not
request unsupported routes. Preserve origin and session checks.
Budget the intended sharing artifact: clean main packs at 8,060,519 bytes versus
8,162,979 with sharing (8,169,168 in CI). Required policy and Unicode tables, bundled
sharing code, and Studio controls account for the growth. Set the packed-only budget
to 8,577,627, following the 5% headroom convention; retain the unpacked cap and all
package checks.
* fix(share): persist refusals and improve CLI retry feedback
* fix(ci): restore render diagnostics and include sharing workspace
* fix(cli): keep sharing owner error codes stable
* fix(share): clarify update feedback and align audience changes
* fix(share): clarify updates and avoid scratch-file rerenders
* fix(cli): report active shares and preserve update channels
* fix(cli): resolve update channel when checking registry
* fix(ci): refresh sharing package budgets after rebase
* fix(share): require consent before uploading project files
* perf(engine): check still frames once per page and share the result with later sessions
* perf(producer): the probe skips the still-frame check unless its session captures
* chore(engine): keep the comment share of the deferred still-frame check files at their base
* fix(producer): the probe session's deferred still-frame check counts as stall-watcher activity
* docs(engine): the still-frame spot-check range is the task's span, not only its own frames
* fix(engine): a reused fast-capture probe session keeps still-frame reuse
* fix(engine): offset clips from a WAV with a wrong byte rate play the right sound to their end
A WAV header that states twice its real byte rate sent the mixer's input seek
to twice the clip's in-point: the clip played later words, then silence. Such a
source is now cut by decoded timestamps instead of by a byte seek.
* refactor(engine): let the byte-rate flag's name carry the seek rule
The review service can now refuse a new version with 429 export_hourly_limit
(too many versions in the last hour). It had no sentence, so the person read
"Couldn't add this version. Try again." and every retry was refused again.
It now says the hourly limit was reached and when to try again, from the
server's Retry-After, falling back to "in an hour" when none is sent. The
daily limit keeps its "tomorrow" fallback.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(parsers,lint,producer,cli): confine composition path lookups to the project
- The local asset resolver offers only files inside the project: a path
that is still absolute after its leading `../` is clamped (an encoded
`//`, a drive letter, a leading backslash) yields no candidate.
- Lint, the catalog-usage walk and `keyframes` read a
data-composition-src only inside the project. Lint and the catalog
walk resolve it with the render's own resolver, now shared from
`@hyperframes/parsers/asset-resolution`, and lint reports one outside
the project as the render pre-flight does.
- The render compile step does not look at a media, gsap script or image
path on another UNC or device root, and the GIF and HEIC preparation
treats a `//host/...` image src as a URL rather than a share.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(parsers,lint): keep Windows absolute asset paths that render copies in
The shared asset resolver now offers only files inside the project. On
Windows a drive-letter src such as C:\media\clip.mp4 used to reach the
real file through the old fallback, and render still copies that file in.
Lint now accepts such a file for media and for CSS written in the page,
so it raises no new missing_local_asset or texture_mask_asset_not_found
error. The media-treatment and normalize-audio commands find it again.
The check never looks at another computer's share, and on macOS and
Linux a src that starts with / is still a project URL, so nothing
changes there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(parsers): expect a drive-letter sub-composition src to be refused on Windows
On Windows an absolute src outside the project is a drive-letter path,
which resolveCompositionSrc refuses, as the render pre-flight already
does. On POSIX the same src is a project URL and resolves inside the
project. The test now expects each platform's result.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(lint): do not report a missing timeline registry a script outside the project may hold
Lint no longer reads a script outside the project. When a root
composition loads such a script, lint cannot know whether the script
registers the timeline, so it no longer reports missing_timeline_registry
for that composition. Render still loads the script. A composition whose
scripts are all in the project is checked as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(capture): validate the Vertex region and model id before captioning
Vision captioning passes the region and model id to the Google SDK, which
builds the request host from the region and the request path from the model
id. A region that is not a plain Google Cloud region name, or a model id with
URL delimiters, now skips captioning with a warning instead of building a
request. Every real region (including "global") and model id still works.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(capture): accept colons inside Google model id segments
Vertex endpoint paths in domain-scoped projects contain a colon
(projects/example.com:proj/...). The colon cannot change the request host,
so allow it inside a segment. Leading slashes, empty segments, percent
signs, query and fragment delimiters stay rejected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* test(cli): give each test file its own home so renders never wait on another file's lock
* test(cli): pin the cli test config after adding the per-file home setup
Long caption overlays plan thousands of verification screenshots, so the
verification budget clamped at its 300 s ceiling in every render worker. That
equals the producer's 300 s first-frame and session-init stall floor, so the
stall always won and the render failed before its first frame. Verification
fails closed on exhaustion (dedup is disabled for unverified runs and normal
capture proceeds), so a lower ceiling costs only some frame reuse.
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* feat(engine): decouple transparent background from png capture format
PNG capture used to imply a cleared page background. A new
CaptureOptions.transparentBackground lets lossless opaque PNG capture
(4:4:4 MP4 output) keep the page background. Absent, it still follows
format === "png", so alpha output and motion blur are unchanged.
Opaque PNG skips drawElement, whose verification and worker-encode
paths are JPEG-only.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(engine): add 4:4:4 chroma option to getEncoderPreset
getEncoderPreset takes an optional chromaSubsampling ("420" by
default). "444" returns yuv444p for SDR MP4, so libx264 encodes the
High 4:4:4 Predictive profile and odd sizes need no even padding.
WebM, MOV and HDR presets ignore it. The ChromaSubsampling type is
exported for the producer.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(producer): chromaSubsampling render option with lossless png capture
RenderConfig.chromaSubsampling ("420" by default) opts an SDR MP4
render into 4:4:4. "444" encodes libx264 yuv444p and captures every
frame as opaque PNG: Chrome's quality-100 JPEG is already 4:2:0 on
beginFrame and screenshot capture, and drawElement can fall back to
screenshot mid-render, so PNG is the only capture that keeps the
colour. drawElement is turned off for these renders.
The probe and main capture options share resolveCaptureSurface, so
they cannot disagree on format, quality or background.
Formats other than mp4, GPU encoding and HDR output are rejected
before any work starts, auto-detected HDR sources fail before capture,
and distributed requests refuse 444. The default render is unchanged.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(producer,studio-server): accept chromaSubsampling on render routes
The shared parseEncoderOptions validates chromaSubsampling and answers
400 for "444" with a non-mp4 format, gpu, or force-hdr, so a request
that would come back 4:2:0 never starts a render. The Studio render
route, both Studio adapters and the producer HTTP server forward it.
The producer server also refuses "444" with outputDynamicRange "hdr",
which parseEncoderOptions cannot see.
No Studio UI control; the option is route-level like crf and hdrMode.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(cli): add --chroma-subsampling flag
`hyperframes render --chroma-subsampling 444` opts an SDR MP4 render
into full-resolution chroma (yuv444p with lossless PNG capture). The
default stays 420. The plan rejects 444 with a non-mp4 format, --gpu
or --hdr before anything renders, with a usage error that names the
conflicting flag.
The value reaches the local render request and, only when it is 444,
the Docker argv, so default Docker invocations are unchanged.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: document 4:4:4 chroma subsampling
Add a "4:4:4 masters" section to the rendering guide: when to use
--chroma-subsampling 444, why it captures PNG, the measured chroma gain
(large for text and UI, none for footage and gradients), its limits,
and a player compatibility table to be filled in from testing.
Document the flag in the CLI reference, the built-in rendering docs and
the hyperframes-cli skill reference (with the regenerated skills
manifest), and the chromaSubsampling option in the producer docs.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: add 4:4:4 player compatibility results
Fill the 4:4:4 compatibility table from testing H.264 High 4:4:4
Predictive output. Chrome 155, Edge 155, VLC 3.0, Safari on iOS 26.6
and Safari 26.6 on macOS play it, and so does AVFoundation on macOS
26.6, the framework QuickTime Player plays through. Firefox 157 on
Windows and both Windows media players do not.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: note the measured render cost of 4:4:4 output
Replace the vague "slower capture" note with measured numbers: a text
and UI composition rendered in about twice the default time, full-frame
footage in about three and a half times, because PNG encoding cost grows
with image detail.
Refs #5555
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(producer): reject chromaSubsampling 444 in distributed render config
Distributed configs have no chroma field and validateDistributedRenderConfig
ignored unknown keys, so a JS caller passing chromaSubsampling: "444" to the
Lambda or Cloud Run SDK got a 4:2:0 file with no error. Reject it, and make
the rendering guide say so precisely.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(producer): check a 444 render reaches the encoder as yuv444p
Every helper had tests, but nothing checked the orchestrator passes
chromaSubsampling through, so dropping it from the encoder preset or capture
surface would turn 444 renders back into 4:2:0 with all tests green. Render a
small fixture with "444" and with the default and check pix_fmt with ffprobe.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(producer): move the 444 drawElement gate into a helper
Keeps runCompileStage under the fallow complexity threshold.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore: keep comment share at or below the merge-base
comment-ratchet failed on 11 files this PR touched. Keep the 4:4:4
rationale once in chromaSubsampling.ts, move the distributed-render
reason into its error message, and cut the rest.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(producer): check a 444 render captures full-resolution chroma
The encoder check alone passes when chromaSubsampling is dropped from the
capture surface: frames are captured as JPEG and encoded yuv444p with
4:2:0 colour. Put a red/blue edge at an odd x, inside a 4:2:0 chroma
pair, and require the two pixels to differ in Cb.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(producer): trim the chromaSubsampling.ts header to the package comment share
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix: finish switching to capturesAlpha from #6262
The merge that adopted main's transparent / capturesAlpha left these files
uncommitted, so resolveCaptureSurface still returned transparentBackground.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(engine): use transparent / capturesAlpha from #6262
main added CaptureOptions.transparent and capturesAlpha (#6262), which do
what this PR's transparentBackground did. Use them, keep the PNG check in
drawElementWorkerEncodeEligible, and re-pin test-classification.mjs and
the skills manifest against current main.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(producer): cut the 4:4:4 comment blocks to one to three lines
The reasoning lives in the PR description with the measurements.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor: reconcile 4:4:4 chroma with latest main
Keep main's video-layer capture overrides, template entry resolution,
ProRes profile parsing and blur-differential tests alongside the chroma
option, and fold the encoder sub-parsers into one loop.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore: keep main's compile-stage tests and regenerate the skills manifest
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>