caption-emoji-pop and caption-neon-accent look a word up in KEYWORDS and
WORD_EMOJI after stripping everything outside a-z, and caption-glitch-rgb
counts a word's \w characters to decide on the aftershock. A Korean, Cyrillic
or accented word comes out empty or cut, so a keyword written in that script
never matches and a long word never pulses.
The three components now strip by Unicode letter (\p{L}) instead, which is
the same set for a lowercased English transcript.
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
* fix(deps): refresh zod, smol-toml, aws-cdk-lib and related resolutions in bun.lock
Lock-only re-resolution within existing ranges:
- zod 4.4.3 -> 4.6.7 (GHSA-mrhx-c7xc-m3w5)
- smol-toml 1.6.1 -> 1.9.0 (GHSA-7w5x-hrqm-74c2, GHSA-r4xh-jqrq-34v2)
- aws-cdk-lib 2.254.0 -> 2.260.0 (GHSA-vcrf-j523-4mrf)
- fast-xml-parser 5.7.3 -> 5.12.0 (GHSA-474x-hrwr-fgx4)
- @babel/core 7.29.0 -> 7.29.6 (GHSA-4x5r-pxfx-6jf8)
- postcss-selector-parser 7.1.7 -> 7.1.8
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(inspector): raise unpacked size budget for postcss-selector-parser 7.1.8
The inspector bundles postcss-selector-parser; 7.1.8 inlines its escaping
(drops cssesc) and adds the hardened parser paths, which grows the unpacked
inspector bundle by about 10 KB and tips it past the 1% tolerance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(deps): bump @mcp-b/global to 6.1.0 to clear @modelcontextprotocol/server advisories
@mcp-b/global 5.x and 6.0.x pin @modelcontextprotocol/server 2.0.0, which is
affected by GHSA-f897-62hf-6gxq, GHSA-46jr-4wx8-v453, GHSA-jm63-cq47-j5jp and
GHSA-rvq5-wwqv-78pq (>=2.0.0 <2.3.0). 6.1.0 is the first release that pins
server 2.4.0.
6.x installs the W3C WebMCP polyfill and passes { signal } to execute; the
real-package tests now drive the 6.x surface (secure context, MCP tools/call).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(studio): close the WebMCP polyfill after the real-package suite
6.x's BrowserMcpServer sends a tools-changed event on a timer after the
last tool unregisters. Without cleanupWebModelContext() it fired after jsdom
teardown and vitest failed the run on an unhandled rejection. Also scope the
transport test's header to the MCP transports it actually checks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
When one capture worker fails, the pool aborts its peers and each of them
reports "Parallel worker cancelled". The pool message listed every worker in
id order, so the one real cause sat between cancellations:
Worker 0: Parallel worker cancelled; Worker 1: Parallel worker cancelled;
Worker 2: Disk capture may need ~2712.6 MB ...; Worker 3: Parallel worker cancelled
The message now carries the failures that are not cancellations and counts
the peers that were stopped because of them. A pool where every worker was
cancelled keeps the full list.
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
* fix(engine): render refusals name the address an IPv6-mapped or NAT64 host carries
* chore(engine): raise the engine size budget for bounded downloads and refusal classes
* fix(producer): accept browser requests only under loopback, IP or listed host names
The standalone render server now refuses a browser request (one carrying Origin or
Sec-Fetch-Site) whose Host or Origin names anything other than an IP address,
localhost, a *.localhost name, or a host listed in the new allowedHosts option or
the comma-separated PRODUCER_ALLOWED_HOSTS variable. A same-origin verdict from
Sec-Fetch-Site no longer lets an unlisted name through. Requests without those
headers (curl, server-to-server) behave as before, and the bind address is
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(producer): trim host-check comments to the package comment budget
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(parsers): never look up media on another computer's share
A media src that names another computer's share (\\host\share\x.mp4,
//host/share/x.mp4, file://host/share/x.mp4) or a device path was checked
for existence before anything asked whether it was on the project's own
drive. On Windows that check alone connects to the host and signs in with
the user's credentials. The shared media resolver now treats such a src as
missing without looking at it, unless it is on the share the project itself
lives on. Render's containment filter, frame extraction, the audio mixer and
the CLI media helpers all resolve through it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(parsers): keep the resolver's comment share at its baseline
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
publish, cloud render and feedback --file-issue build their archive from a
walk of the project folder that only consulted .hyperframesignore, so a rule
re-including dotfiles also uploaded .env files. The walk now leaves out env
files, Git data and Studio's private records whatever the ignore rules say,
and reports each one as left out with the reason "private".
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(docs): check catalog link values by type and isolate the preview
A catalog page restores its variable values from the query string, so a
shared link decides what the preview composition is given. Values from a
link are now kept only when they fit the variable they name: a finite
number, a boolean, one of an enum's options, a colour, an image address,
or text with no angle brackets. Anything else opens at its default.
The preview player now runs the composition with an opaque origin
(sandbox-origin="opaque"), so a composition cannot reach the docs page.
The bootstrap already drives the player only through its API and
postMessage, and every docs asset is served with CORS, so playback,
seek-on-remount and edits behave as before.
chatgpt-exchange sets its editable text with textContent instead of
building innerHTML from it. A new check, check:registry-innerhtml, fails
any registry file that reads variables and adds markup writes; the files
that do so today are listed with their current counts and may only go
down.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(docs): keep a link's text out of copied commands until the reader sets it (#7785)
A catalog page restores tuned values from its link, and the install
command, the render command and the agent request carried those values
as if the reader had typed them. The commands quote only for bash and
zsh, and the agent request is pasted as the reader's own words.
The page now tracks which variables the reader set on it. A value that
only came with the link goes into a command when it is a number, a
boolean, or short text of letters, digits, spaces and plain punctuation,
and into the agent request only when it is a number, a boolean or one of
the item's own options. A page opened with values from its link says so
and offers "Reset to defaults", and the install command wraps instead of
scrolling so the whole line is visible.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
downloadToTemp now refuses any Content-Encoding other than identity, caps each
download at 20 GiB (HYPERFRAMES_MAX_DOWNLOAD_BYTES) on the declared Content-Length
and while streaming, and aborts a transfer whose bytes over the last 60 s average
below 32 KiB/s (HYPERFRAMES_MIN_DOWNLOAD_BYTES_PER_SEC, 0 disables).
Compressed and over-limit responses are not retried. A throughput abort is a
retryable timeout like the stall timer, which is unchanged. A sized body that
trickles one chunk every 100 s is now cut by the floor instead of finishing.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): start helper programs only from trusted PATH folders on every platform
At startup the CLI and its worker entry points rewrite their own PATH to
absolute folders outside the working directory, on every platform, so a
program planted in the project (a relative or empty PATH entry,
node_modules/.bin, or any folder inside the cwd) is never run by name.
Child processes inherit the trimmed PATH. media-use reuses the same helper
instead of its private copy.
doctor resolves docker with findTrustedExecutable and runs it by absolute
path with execFileSync, no shell. A docker found only in the project's
node_modules/.bin is reported as not found, with a hint that it was skipped.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): keep stale-install detection and Windows system folders after PATH trim
The stale hyperframes check for reverse start reads the PATH the process
started with, so an older copy in the project's node_modules/.bin is still
reported. It only reads the filesystem and never runs that copy.
On Windows the trim now keeps %SystemRoot%, System32, System32\Wbem and
the WindowsPowerShell folder even when the CLI starts inside them, so an
elevated prompt in C:\Windows\System32 can still run cmd, taskkill,
tasklist, powershell and tar. PATH entries wrapped in double quotes are
unquoted before the check instead of being dropped. PATH order is kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): resolve the empty-PATH placeholder only when it is needed
Computing it at import threw in test environments where import.meta.url is not a file URL.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio): menu row highlight follows the menu's corner
* fix(studio): keep the menu row's comment share flat
* fix(studio): export the menu row class so hosts reuse it
* fix(engine): NAT64 addresses count as non-public; producer exports the address rule
* chore(producer): drop the new section comment
* fix(engine): judge mapped and NAT64 addresses by the IPv4 address they carry
* test(engine): operator ranges hold through mapped and NAT64 addresses
scripts/upload-docs-images.sh synced docs/images/ with the AWS CLI's default
of following symbolic links, so a link committed under docs/images/ would
publish whatever it pointed to on the machine running the upload. The script
now stops before any AWS call when docs/images is not a real folder in the
checkout or when anything under it is a link or not a regular file, and the
sync passes --no-follow-symlinks as well.
A new check, scripts/check-symlinks.mjs, fails CI when a tracked symbolic
link is not one of the two the repo already means to have.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): a subfolder of an imported project keeps its received-files record
Publish, cloud render and feedback now read the received-files record at
the nearest folder at or above the one they stage, so staging a subfolder
of an imported project still leaves out the outside files its received
files reference.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): a received file counts under any spelling of the same file on disk
The received-files lookup now also matches a project file that is the
same file on disk as a listed path, so a name respelled in letter case on
a case-insensitive disk keeps its received status.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): cloud render prepares only the GIFs staging kept
The entry GIF prep before a cloud render upload now reads only the staged
copies it is handed. An image whose source staging left out keeps its
reference as written and is never read from disk.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli,studio-server): never hand a playlist named as media to ffmpeg
A media file whose first bytes are an HLS playlist, an XML manifest or a
concat script now stays out of ffprobe and ffmpeg: cloud render leaves it
in the zip as authored, the media probe reports it as not media, and the
proxy copy refuses it. Real media never starts with these bytes, so it
probes and transcodes as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): catch playlists with control bytes, and keep them from stills and audio probes
A text playlist or concat script is detected whatever bytes follow its
first line; only markup must be all text. A still frame and the audio
probe behind waveforms, peaks and loudness now refuse a file the probe
reads as not media, as the proxy copy already does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): a parent folder's state the CLI cannot read never marks a project received
The walk up for a received-files record now skips a parent whose record
cannot be reached. Only a record file that is there but cannot be read
or parsed counts every file as received.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): a drive that reports no file ids matches no received file by identity
Some network and virtual drives report a file id of 0 for every file, so
the same-file match now gives such a file no identity instead of
matching it to every listed file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(studio-server): read a media file's first bytes with a read stream
The playlist check reads the first 64 bytes through a bounded read
stream instead of a raw file descriptor, and the audio-track probe's doc
comment sits on its function again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(studio-server): raise the packed size budget for the media playlist check
Main packs @hyperframes/studio-server close to the 1% tolerance on the
168500 budget, and the playlist check adds about 320 bytes, which put
the Build over. Set the budget to 172000.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli,studio-server): open media only through self-contained media demuxers
Every ffprobe and ffmpeg input that cloud render offload, the publish
proxy bake and Studio's media helpers open now carries a demuxer
whitelist built from the supported media extensions: MP4/MOV/M4A/AVIF,
Matroska/WebM, MXF, AVI, MPEG-PS/TS, FLV, ASF, Ogg, WAV, MP3, AAC, FLAC,
AIFF and the still-image formats. Playlist, manifest and concat demuxers
can no longer open, whatever a file's bytes say. The proxy's own concat
step is left as it is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli,studio-server): keep media probe argv flat, and expect the demuxer list in loudness args
Two probe argv literals built from nested spreads are flat again, so
"--" stays the visible second-to-last entry the argv contract checks.
The normalize-audio test expects the demuxer list before the input.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): keep Chrome's sandbox on for website capture where the host supports it
`hyperframes capture` added --no-sandbox on every Linux host. It now uses the
engine's host check, the same one rendering uses: the sandbox stays on unless
the process runs as root, in a container or serverless runtime, or with user
namespaces restricted. HYPERFRAMES_CHROME_SANDBOX=0/1 still overrides, and a
launch Chrome refuses with the sandbox is still retried without it. When the
host check turns the sandbox off, capture prints a one-time warning.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): drop the new comment block from the capture launch args
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): keep captured page text out of generated agent instruction files
The AGENTS.md / CLAUDE.md / .cursorrules files that `hyperframes capture`
writes are loaded by coding agents as instructions. The heading now uses the
captured site's hostname instead of the page title, font names are kept only
when they read as plain names, and the file states that everything the capture
saved from the website is data, not instructions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): trim the font-name comment
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): start nvidia-smi, vm_stat, ps and tasklist from trusted folders only
The renderer resolves these helpers through the shared trusted lookup and runs the absolute path, so a program with the same name in a project folder is never started in their place.
* test(engine): memory and import tests follow helpers started by resolved path
* fix(producer): every video's size samples share one grid
Videos with offset starts each added a full set of seek times; on one shared grid the count follows the composition's length, not the number of video tags.
* fix(producer): keep the shown-box times comment to one line
* fix(aws-lambda): Lambda renders reach only public addresses unless the function opts out
* chore(aws-lambda): drop a comment the function name already says
* perf(lint): rules scan script, style and comment blocks in linear time
Several always-on rules matched open/any/close regexes over the whole
document, which rescan to the end from every unclosed opener, so lint
time grew with the square of the page on pages of unclosed <script,
<style, <pre or /* openers. Those scans are now forward passes in
lint/src/linearScans.ts that return exactly what the regexes did:
block-comment stripping, the malformed </script check, the style-tag
balance count, the visible-comment rule's protected blocks and tag
tracking, the structural line count, the script-src check, and the
slideshow island lookup.
A fuzz test compares each scan with its regex on 20,000 generated pages,
and 4 MB pages of unclosed openers finish in milliseconds. On a 140 KB
page of unclosed <script> openers lintHyperframeHtml drops from about
2.6 s to 45 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(lint): compare each fuzz page in one assertion so the run fits the default timeout
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(player): raise the packed size budget for the bundled slideshow island scan
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): refuse media playlists that name files outside their folder
* refactor(producer): media preflight decides refusals in one helper
* fix(engine): read a playlist once, bounded, instead of sizing it first
* fix(producer): shown-box sampling stays within the validated render duration
Sampling where each oversized video is drawn stepped through the authored
start and end in 0.25 s float steps with no bound, before the render
duration was checked. An enormous start never advanced the step, and an
unbounded end grew the sample set until memory ran out.
The render duration is now validated first, each window is cut at the
composition's end, a window that is not finite is skipped, and the loop
counts integer steps. Building the sample list happens inside the
measurement's try, so any failure falls back to full-size extraction.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix: render and capture file servers refuse env files and Git data
The producer render file server, the engine file server and the CLI
static project server served every regular file under the project to the
page, including `.env*` files and anything under `.git`. Compile already
refuses to inline those files, and Studio preview and play refuse to
serve them.
All three now answer 404 for them through the shared isVcsOrSecretPath
rule, checked on the requested name and on a link's target, with the
case and Windows spelling variants it already covers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(producer): inline only public HTTPS scripts, under the download URL policy
The compile step fetched every external `<script src>` with a bare fetch:
any scheme, any host, redirects followed, body size unbounded.
It now fetches through the engine's download policy, the one remote
stylesheets and media already use: HTTPS only, literal and resolved
addresses public, every redirect hop checked again, connect pinned to the
checked address, and at most 16 MiB per script. A new engine helper,
fetchPublicHttpsBytes, returns the raw bytes so integrity metadata is still
checked before decoding; fetchPublicHttpsText now decodes its result.
A script outside the policy keeps its `<script src>` tag, and the capture
browser loads it as before; the log says so at info level instead of
warning that the render may fail. Scripts under the operator's
HYPERFRAMES_GSAP_BASE_URL mirror, which may be loopback HTTP, are still
inlined. An integrity mismatch still fails the render.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(producer): the audio probe skips remote video URLs outside the public HTTPS policy
A video whose remote src compile could not download keeps that URL, and
the probe-only audio pass handed it straight to ffprobe, which fetches any
scheme and host. A src that fails the literal public-HTTPS check (plain
HTTP, loopback, private, link-local or `.internal` hosts) is now skipped,
the same outcome frame extraction already gives it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(producer): the GSAP mirror exception covers only its dist folder
Scripts from the operator's HYPERFRAMES_GSAP_BASE_URL mirror are fetched
outside the public-HTTPS policy, because the mirror may be loopback HTTP.
That exception now covers only plain script file names in the GSAP dist
folder the compiler itself points at, refuses redirects and stops at
16 MiB, like every other inlined script. Any other path on the mirror's
origin follows the public-HTTPS policy and is left for the browser.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): on a render service, https media inputs open only HTTPS
ffmpeg and ffprobe read some https sources in place. An https playlist
or manifest can name segments on any protocol, and the default protocol
list for an https input does not restrict them. With
PRODUCER_RESTRICT_PAGE_NETWORK=true, these inputs now pass
`-protocol_whitelist https,tls,tcp`, so segments and parts they name must
be HTTPS too. Local renders are unchanged.
A new test pins that local project files which are playlists, concat lists
or DASH manifests (with their own extension or named `.mp4`) are probed
and decoded without any request to the URLs they list: ffmpeg limits what
a local file input may open to file, crypto and data, and the concat
demuxer refuses URL entries in safe mode.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): renders reach only public addresses from every part of the browser
* fix(engine): renders reach only public addresses from every part of the browser
* fix(engine): renders reach only public addresses from every part of the browser
* fix(engine): renders reach only public addresses from every part of the browser
* feat(studio): press ? to open the keyboard shortcuts panel
Adds a rebindable `shortcuts` action bound to `?` (Shift+/) that toggles
the shortcuts panel; Esc still closes it. The key is ignored while focus
is in an input, textarea or contenteditable, and the panel lists `?`
under Panels.
Refs heygen-com/hyperframes#5580
* docs(studio): document the ? shortcut for the shortcuts panel
* fix(engine): look a host up again after a failed lookup instead of refusing it for the whole page
* fix(engine): name the failed-lookup reset instead of commenting it
* fix(engine): hold the render page to the asset download address rules
With PRODUCER_RESTRICT_PAGE_NETWORK=true the render page reaches only its own project server and public addresses, checked by host and by resolved address, the rule asset downloads already follow. Unset, nothing changes for local renders.
* fix(engine): keep the page network filter comment within four lines
* fix(producer): a clip whose file name has a space or accent is mixed into the export once
* fix(producer): compare a clip's src decoded, keeping the authored spelling either way
* fix(studio-server): save an edit Windows briefly holds without freezing the app
* fix(studio-server): keep later edits behind a held one, one backup per replay
* fix(studio-server): send a held cut again, and keep the sync retry for rollbacks
* fix(cli): an unknown flag's list leaves out hidden rename shims
The list pointed agents at init's --template, --video-legacy and --agent,
which only print an error. It now shows the flags --help shows.
* fix(cli): an unknown flag never suggests a hidden rename shim
The did-you-mean search still used every spelling, so '--templat' on init
was pointed at the hidden --template. It now uses the same visible flags
the list shows.
* fix(studio-server): check and render keep their history on a drive without hard links
The history lock had its own list of the errors that mean a drive cannot
hard-link, and it lacked EISDIR, which Windows returns on a Google Drive
folder. Every check report and render record there failed. The lock now
uses core's list, the one owner, which also gains ENOSYS from the lock.
* fix(engine): downloads and plan blobs publish on a drive without hard links
The producer's plan publisher kept a third list of no-hard-link errors
without EISDIR, and the downloader had no fallback, so both failed on a
Google Drive folder. Both now use core's cannotHardLink; the downloader
copies into an exclusively created file, so it still never overwrites.
* fix(engine): a download on a drive without hard links publishes atomically
Instead of copying into an exclusively created file, the fallback renames
the staged download into place under the cache path lock when nothing is
there yet, so a killed run never leaves a truncated file to reuse.