* test: a concat list refuses paths whose folder name holds a line break
* fix(engine): write every concat list through one writer that refuses control characters
* test: name the line-break folders without slashes; skip them on Windows
* Revert "fix(engine): write every concat list through one writer that refuses control characters"
This reverts commit 97d482340a.
* Reapply "fix(engine): write every concat list through one writer that refuses control characters"
This reverts commit 7d723fd276.
* test(engine): a chunk concat list ends with a newline
* fix(studio): a sealed preview keeps Studio's size, move and turn edits after a seek
* fix(studio): a sealed preview reapplies Studio's edits on each frame it seeks or plays to
* fix(studio): build the sealed reapply script from constants, never a caller's value
* fix(studio): a sealed preview reapplies Studio's edits after a redraw at the same time
* fix(studio): serve the seek reapply only into sealed preview documents
* chore(core): shorten the redraw count's doc line
* fix(studio-server): proxy piece list names only the pieces, not their folder
The chunked proxy join wrote each piece as an absolute path into a concat
list in the system temp folder. A line break in a folder name above the
project ended an entry early, so the rest of the path could be read as
further list lines naming other local files.
The list now sits in the cache folder beside the pieces and names each
piece by its own generated file name, and ffmpeg runs the join from that
folder. No project or folder name is written to the list, so folder names
with line breaks, quotes, '#' or '?' all join only the copy's own pieces.
The CLI publish bake uses the same join and is covered too.
* fix(studio-server): a proxy copy lock dated ahead of the clock no longer holds waiters
A copy waits for another process's lock while that owner's lock, temp and
piece files keep changing. The newest modified time was taken as is, so a
lock or temp file dated in the future looked like fresh progress forever,
and the waiter held its transcode slot without end. Preview and the
publish bake could then wait with nothing running.
Each file's modified time now counts as no later than the moment the
waiter first saw that time. A file dated ahead of the clock ages from
then, so the waiter gives up within the stall window and frees its slot.
An owner that keeps writing still sets a fresh time on each change and is
waited for exactly as before, even beside a future-dated file.
* fix(studio-server): keep the absolute piece list and default folder on Windows
On Windows the working folder is part of the DLL search, and a long cache
folder path may fail to start a process from it. Windows paths cannot hold
a line break, so the piece join there lists absolute piece paths again and
starts ffmpeg from the default folder. Elsewhere it still runs from the
cache folder with bare piece names.
The line-break test now injects an unquoted list line, so the old join
really pulls in another file, and it checks that the copy holds none of that
file's frames. A spawn test checks the join's folder and bare list name.
* fix(producer): resume checks the segment file the join reads, not the manifest's path
A resumed segmented render now validates each reusable segment at its own
computed path in the segment folder. It must be a plain file, not a link, and
match a slice of this render's plan. The path a manifest records is never
opened or probed.
* fix(engine,producer): concat joins open only the list and encoder containers
Both concat-demuxer joins (the engine's chunk and segment concat and the
distributed assemble join) now pass a format whitelist of concat, mov and
matroska. ffmpeg applies it to every file the list opens, so a listed file
that is a playlist or manifest cannot pull in other files, while mp4, mov and
webm joins work as before.
* fix(producer): resume reuses only segments this install recorded
Segment manifests are now signed with an HMAC keyed by a per-install secret
kept in the HyperFrames config folder, outside every project. Resume trusts a
manifest only when its signature checks out, so a segment folder copied from
another machine or shipped inside a project is captured again, while a
render's own interrupted run resumes as before.
* docs(producer): restore the segment path ordering comment and rewrap resume docs
* fix(producer): replace an empty or short resume key file
A key file too short to use is now replaced by a fresh key instead of
disabling resume for good. A missing key is still linked into place, so
renders that create it at the same time share one key; a test runs eight
processes against one path to check that.
* fix(producer): warn under --resume when no resume key is available
When the resume key can be neither read nor created, for example on a
read-only home folder, a file system without hard links or a serverless
runtime, --resume now logs one warning that every segment will be captured
again instead of only reporting zero complete segments.
* fix(engine): fail a concat join at a list entry ffmpeg refuses
The concat demuxer ends the stream at a listed file it cannot open and ffmpeg
still exits 0, which cut the output short without an error. -xerror on the
concat input makes the join fail instead. mp4, mov and webm joins still exit
0.
* fix(producer): resume checks each reused segment's SHA-256
Each finished segment's SHA-256 is now recorded in the signed manifest. The
hash is taken once when the segment closes, and on resume only the files
about to be reused are hashed again, so a file swapped for another of the
same size is captured again.
* docs(cli): note that resume re-checks segment hashes and recaptures older segments once
* fix(studio-server): proxy joins only its own pieces, through the shared concat input
Studio's chunked proxy copy now names its pieces with a tag keyed by the
user's preview token, so pieces that arrive inside a project folder sit at
names it never looks up and every piece is encoded fresh. A stopped copy of
the user's own still continues from its finished pieces.
The piece join uses concatListInputArgs, now in @hyperframes/parsers so the
engine, producer and Studio share it. It allows only local files in the
concat, mov or matroska formats and fails at a refused entry. The cache sweep
recognizes both tagged and older piece names.
* test(parsers): prove the concat protocol whitelist with a data: entry
ffmpeg already refuses http:// in a concat list opened from a file, so the
test passed without the whitelist. A data: entry opens by default; the test
now asserts it is refused as not on the 'file' whitelist.
* fix(studio-server): only the copy's own tagged pieces count as lock progress
The copy lock counted any .part-<key>- file as the owner's progress, so a
piece shipped with a future modified time kept a stalled lock looking active.
It now counts only names under the copy's own tagged piece prefix. The cache
sweep removes untagged pieces past the usual age even while that copy runs,
and measures age in both directions so a future-dated piece is swept too.
* fix(studio-server): say once when proxy pieces cannot be keyed to the user
Without a preview token, piece names use a per-process secret, so pieces are
reused only within that Studio process. Studio now logs that once, so a copy
restarting from its first piece after a restart has a visible cause.
* test(studio-server): name the proxy join test for what it checks
* fix(producer): derive the segment manifest key from the preview token
Sign segment manifests with an HMAC key derived from the per-user preview
token instead of a separate key file. The token is created with an exclusive
open, so concurrent renders never end up with different keys, and it is only
used when the file is private to this user.
Resume now checks a segment's SHA-256 before probing it, and a real-ffmpeg
test documents how the concat join treats an entry that is itself an ffconcat
list.
* docs(producer): describe the manifest key per user and the no-token case
* fix(studio-server): an inserted scene names its file from the project root, as the renderer reads it
* test(studio-server): keep a test comment under the line limit
A folder move or delete in Studio checked each file inside only by its
name. A script that package.json runs from a subfolder could be moved
aside or deleted with its folder. A move could also land a file on a
path a direct write refuses, such as a script package.json runs or
mise/config.toml, by moving a folder onto the right name.
Now each entry in the folder is checked at its current path with the
same rules a direct delete or rename uses, including the package.json
scripts. For a move, it is also checked at the path where it would
arrive. Ordinary folder moves still work.
npm, pnpm, yarn and bun run `node-gyp rebuild` on install when a binding.gyp
is present and package.json has no install script, and a .gyp file can include
any .gyp or .gypi. npm, pnpm and yarn run `node server.js` for start when
package.json has no start script. A project from `hyperframes init` has
neither script, so both defaults apply.
toolTrustedFileKind now classifies .gyp/.gypi files and server.js, in every
spelling projectPathSegments folds, so Studio's file routes and the registry
installer refuse to write them. Rename and duplicate refusals now report
why: protected_path when the destination is inside the project, as POST and
PUT already do.
* feat(studio-server): let a host count the requests the API refuses with 403
* fix(studio-server): a refusal names the endpoint asked for, not the middleware that refused it
* style(studio-server): refusal reports carry their meaning in names, not comments
* fix(studio-server): a refusal the API sends through itself again is counted once
* fix(studio-server): a request sent through the API again is known before a middleware swaps it
A media-proxy request with neither Sec-Fetch-Site nor a Referer naming the
host was treated as same-site. Play, the static project server and Studio now
require a positive signal and refuse the request otherwise. The check
command's own re-request of failed media sends a Referer for its origin so it
keeps working.
— Rames
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): the render file server never serves a project file as a service worker's script
* fix(producer): render, capture and inspector servers never serve a service worker's script
* fix(cli): cloud render --json reports re-encode progress and times out a stalled cache check
* refactor(parsers): share the ffmpeg progress reader from ff-binaries
The reader lived in a new studio-server entry point, which pushed its packed size over budget and left package.json out of step with package-subpaths.json. Both callers already depend on parsers. The re-encode now gets a progress callback only when someone listens.
* style(cli): format the offload test
* fix: refuse media paths that ffmpeg's image2 demuxer would read as a pattern
ffmpeg's image2 demuxer reads %d, %03d, %% and a % before a wildcard in an
input path as a pattern of other files, folders included. Every project,
private-file and received-file check approves the literal path, so a
composition could name a placeholder whose pattern reaches images outside
the project and get them back as video frames.
Every ffmpeg and ffprobe input built from a project file now goes through
assertLiteralMediaInput, which refuses such a local path before ffmpeg runs.
The engine's urlInputArgs, the parsers' mediaInputFormatArgs and the
studio-server's new mediaInputArgs all take the input path, so the CLI
snapshot and cloud offload, lint, core grade analysis and studio-server
callers are covered too. URLs and other percent signs, such as %20, are
read as before.
* build(parsers): declare the literal-media-input subpath
* build: raise the parsers packed-size budget for the literal-media-input subpath
The new subpath adds its own built entry, type file and export map rows,
about 2.7 KB packed. It is a separate module so test suites that mock
ff-binaries still run the real path check.
* test(parsers): a harmless name shaped like a pattern is refused too
* fix(studio-server): cache reads never follow a link the project ships
The waveform and peaks caches, the thumbnail cache, the stored preview
documents and the preview proxy copies all live in folders inside the
project, which its author controls. Their reads opened entries with
link-following calls, so a project could ship a link at a predictable cache
name and have Studio return the file it pointed at.
Each of these reads now goes through readProjectCacheFile or
isProjectCacheFile: the entry must be a plain file, reached with no link
from the project folder down, and is opened without following links.
Anything else is a miss, and the entry is rebuilt.
- Waveform and peaks entries must also parse as an array of numbers.
- A proxy copy, its fast and content index files, and the pieces of a
chunked copy count only as plain files.
- Stored preview documents carry an HMAC under the user's private preview
token, so a document the project shipped is never served. Without a
token, nothing is stored.
* fix(studio-server): the proxy copy lock treats a link at the copy name as not made
The copy lock tested "the copy already exists" with a link-following check,
and the transcoder returned the copy path unchecked when the lock said
another process had made it. A link at the copy name was then served.
- acquireCopyLock takes the caller's "copy made" test; the transcoder
passes isProjectCacheFile and re-checks the path before returning it.
- Copy index files are written by rename, so a link at an index name is
replaced instead of written through.
- Chunk pieces are checked from the project folder down, like every other
cache entry.
- A cache entry with more than one name (a hard link) is a miss.
* fix(engine): check proxied download hosts against local DNS; keep raw GIF downloads private
Behind an env proxy (HTTPS_PROXY with NODE_USE_ENV_PROXY=1 or
--use-env-proxy), composition downloads skipped the resolved-address check
and relied on the literal-host check alone. An internal DNS name passed, and
NO_PROXY hosts were dialled directly.
Now the downloader resolves the host on this machine first and refuses it
when any answer is non-public. A name this machine cannot resolve is still
left to the proxy, so proxy-only networks keep working.
Studio's remote GIF prep now picks only URLs whose path ends in .gif,
discards a downloaded body that does not start with a GIF signature, and
the raw downloads folder is no longer served to the page; only the converted
output is.
* fix(cli): read the downloaded GIF's header through a bounded read stream
CodeQL read the openSync call as a temp-file creation. The file sits in the
project's .hyperframes folder and is only read, but a six-byte read stream
does the same check without a file descriptor of our own.
* fix(cli): a received folder's record is read wherever Studio moves it
The search for received-files records now walks every folder a render,
preview or publish can load files from, including renders/, dist/, .next/,
coverage/ and Studio's own folders in the project's .hyperframes/. A
received folder moved there by Studio keeps its files received.
Studio's folder delete and move also refuse a folder holding a
received-files record that lists a file outside its own folder, since
that file would stay behind with no record. A received folder whose
record lists only its own files can still be deleted or moved.
* fix(cli): a renamed or moved received file stays listed under its new name
The CLI-hosted Studio now follows every rename, move or delete it
completes in the received-files records, whichever client sent it. A
renamed file, or every file in a moved folder, stays listed under its
new path; a file moved out of a received folder is listed in the
project's own record. Changes are made under the record's lock, and a
record that cannot be changed is left as it was.
The per-file readers used by publish, cloud render, feedback and add now
count every file as received when a record lists a file that is no
longer on disk, since that file may be there under another name. This
matches the local render check. Projects with no record are unaffected.
* fix(cli): received content that leaves the record still counts as received
When a Studio rename, move or delete takes a received file off a listed
path, the sha256 of its content is kept in .hyperframes/received-content.json,
the file and format the desktop app uses; other keys in it are kept. A
file holding the same bytes counts as received for publish, cloud render,
feedback, add and the local render check, so content brought back by undo
or renamed outside Studio stays received. Studio's own backups do not count.
A listed file that is gone from disk now makes only pages, stylesheets,
scripts, SVG and links out of the project count as received, and the
upload warning names the gone file.
The record follows a request spelled in another letter case or Unicode
form by matching the name the disk holds, and waits for the record's lock
without blocking the server. Studio's folder delete and move no longer
refuse a folder whose record lists a file outside it; that file's content
is kept as received instead.
* fix(cli): a folder delete or move keeps what a record inside it lists elsewhere
Before Studio deletes or moves a folder, every file that a received-files
record inside the folder lists outside it is listed in the project's own
record by its path from the project folder. Such a file stays received
after it is edited, and a listed link stays received too. When the
project's record cannot be written, the delete or move is refused.
Received content that cannot be read is never written over, so the
desktop app's keys in it are kept. The warning about a listed file that is
gone now says how to clear it.
* fix(cli): received-content checks skip large non-code files and name a skipped search
A file larger than 16 MB that is not a page, stylesheet, script or SVG is
no longer read to compare its content with received content, so a large
archive is not read on every check. When the project has too many folders
to search for received-files records, the upload warning now says so.
The case-matching publish test now expects the gone-file rule on a disk
that tells letter case apart, and a comment no longer names a file path.
* fix(cli): name a gone received file from the project folder on every system
The gone-file note now names the file from the same project path the
record search starts at. On Windows, a short (8.3) folder name in the
temp path made the two spellings differ, so the note showed a long
relative path instead of the file's name.
Content hashing opens a file once, checks it through that handle and
reads from the same handle, and the received-content file is read before
it is classified, so nothing is checked by path and then used by path.
* fix(cli): received files stay received through undo, saves, and partial folder deletes
An undo, redo or restore lists again every file it writes that holds received
content, and a Studio save lists such a file before it changes it. So a Studio
delete followed by an undo and an edit leaves the file received.
A Studio rename, move or delete now records the received side first, under the
record's lock, and is refused when that fails: the content leaving listed paths
is kept, a renamed file is listed under its new path, and every line of a
record inside a moved or deleted folder is copied into the project's record.
Once the files have moved, only lines whose file was there and whose content
was kept are taken off. Lines at a rename's destination and lines whose file was
already gone stay. The search for records inside the folder has no size limit,
and nothing is locked when nothing is listed.
* fix(cli): unreadable received content counts every file as received
The readers treat a received-content file they cannot read the way they treat
an unreadable received-files record: every file counts as received, and the
local render check names the file it could not read.
A record line that points outside the project is ignored, so a record in a
received folder cannot make the readers look at or report files outside the
project.
* fix(cli): the CLI's own history undo and restore keep received files listed
`hyperframes history undo` and `restore` with no Studio running now list again
every file they write that holds received content, the same way Studio's
history routes do. A delete in Studio, then an undo or restore from the CLI,
then an edit, leaves the file received.
A Studio edit that names its file in the request body, such as a freeze-frame
edit, now lists that file first when it holds received content.
* test(studio-server): a loopback page with no fetch metadata is marked unless it is the app's own
* fix(studio-server): a loopback page with no fetch metadata is marked unless the app framed it
* test(studio-server): a page marked by its Referer names Referer in Vary
* fix(studio-server): a page marked by its Referer tells caches it varies by Referer
* fix(studio): the copied render command carries no composition text a shell can run
The Variables panel's "Copy render command" put every variable value inline,
quoted for POSIX shells only. A shared project's author chooses those values,
so in PowerShell, cmd.exe (including npx.cmd's re-parse) or fish a default
such as `x'; calc.exe; '` ran as commands when the command was pasted.
The command is now fixed text: `npx hyperframes render -c <composition>
--variables-file .hyperframes/render-variables.json`. Copying it saves the
previewed values to that file through a new PUT /projects/:id/render-variables
route, which refuses a link standing in for .hyperframes and replaces, rather
than writes through, a link at the file itself. A composition path that is not
one plain word in every shell (letters, digits, `_ . / -`) gets no command.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(scripts): the comment checker skips a template nested in a substitution
stripLiterals ended a template literal at the first backtick, including one
that opens a nested template inside `${...}`. On a line such as
`'${value.replace(/'/g, `'\\''`)}'` that left a backtick unmatched, so
scanCLike read the rest of the file as one template and counted none of its
comments. The ratchet then measured VariablesPanel.tsx on main at 1 comment
line instead of 31 and failed any change that removed that line.
A `${...}` is now scanned as code: strings, regexes and templates inside it
are skipped, and its closing brace returns to the outer template.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(studio-server): read the replaced render-variables record through one no-follow descriptor
The link test checked the record with lstat and then read it by path, two
lookups CodeQL reports as a file-system race. Opening it once with O_NOFOLLOW
fails if a link is still there and reads the file that open checked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): audio decodes are bounded, stop when nobody waits, and their cache is capped
The waveform and peaks routes each started a full-file ffmpeg decode per request, with no limit on how many ran
at once, no sharing on the waveform route, and no way to stop one after its request went away. Their cache
folder also grew without end.
- Both routes now decode through one shared queue of 4 slots, one decode per cache entry, shared by every
request that wants it. When the last request waiting on a decode leaves, its ffmpeg is killed (or it leaves
the queue).
- `.waveform-cache` keeps at most 500 entries; a write past that drops the oldest.
— Rames
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): trim comments to the package's comment share
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(studio-server): raise the packed size budget for the audio decode limits
The bounded decode queue, the abort when nobody waits and the capped cache
add about 485 bytes to the packed tarball. Main had already reached 178,489
bytes against a 178,000 budget, so together they pass the 1% tolerance
(179,877 bytes). The new budget is the measured size plus the script's 5%
headroom.
— Rames
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): waveform decodes keep a fixed-size peak buffer, time out, and the cache is byte-capped LRU
- decodeAudioPeaks streams samples into a 65536-entry max-abs buffer that merges
neighbouring runs when full, instead of buffering and concatenating the whole decode.
- A shared decode is stopped after 10 minutes and its slot freed.
- .waveform-cache is capped at 256 MB as well as 500 entries; a cache hit marks its
entry used, so eviction drops the least recently used.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(studio-server): raise the unpacked size budget after rebasing onto main
Measured with pnpm pack after the rebase: 697160 unpacked bytes, over the
690000 budget by more than 1%. The new budget is the measured size plus 5%.
— Rames
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): files from outside the project are uploaded only once approved
publish, cloud render and feedback --file-issue copied every image, video, audio,
font and CSS file a composition named from outside the project folder, and every
project file that links outside it, and only listed them. A project that came from
someone else could pull a file from anywhere on the uploader's disk into an upload.
Now each such file needs approval. In a terminal the command lists them and asks,
defaulting to no; a yes is remembered per project in the CLI's config folder
(~/.hyperframes/outside-files.json), never in the project. --yes, --json, --dry-run
and unattended runs leave them out with a warning naming each file and saying how
to include it. --include-outside-files uploads them without asking.
— Rames
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(cli): check the approvals file mode only where files have POSIX modes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(studio-server): bound preview stills and keep proxy work to the Studio's own pages
- Queue at most 16 stills behind the 4 being made; past that, answer 503 with Retry-After.
- Refuse a still past the end of the source before starting ffmpeg (404).
- Refuse a ?hf-proxy request on a plain preview path when another site asks for it. The
Studio's own opaque frames already load through the frame-token path, and sandboxed
thumbnail captures now do too.
— Rames
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* chore(studio-server): raise packed size budget for the still-queue and cross-site checks
The bounded still queue, the past-end refusal and the cross-site proxy
check add about 2.2 KB to the packed studio-server (180252 bytes in CI).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A local render brought in .js and .mjs files that markup or a project link
reached outside the project, so an imported composition could copy a script
from elsewhere on disk into the page it runs and read it. Images, video,
audio, fonts and CSS still come in. Scripts are left out with a warning that
names each one and how to include it: move it into the project, or pass the
new --include-outside-files to hyperframes render.
The policy covers the compiler's outside-asset copy, the file server's
linked files and the distributed plan's project copy.
— Rames
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
History undo, restore and Cmd+Z now leave alone every file that Studio's
file routes refuse to write. This covers tool-trusted, private, hidden and
node_modules paths, and scripts the project's package.json runs. Cmd+Z
passes over an edit that only changed such files. A claim no longer takes
such a file's pending change.
A claim's `overwrote` version now counts only when the server itself saw
that version at that path. A version stored for another file can no longer
become a file's earlier state.
The write rule also refuses more files that tools and agents load on
their own: opencode and crush settings, compose override files,
Dockerfile.* and Containerfile, every lefthook spelling, mise, nodemon,
devbox, wrangler, vercel and netlify settings, *.config.jsonc,
tsconfig.*.json, Taskfile.dist, simple-git-hooks, Ruby build files, and
.nix and .fish scripts.
* fix: hidden and credential files stay out of Studio, history and lint
- The private-file rule now refuses every hidden file or folder in a project,
except Studio's own files, the root .media folder, and pictures, video,
sound, fonts, stylesheets and scripts a page can load. It also refuses
Terraform state and variable files and more key-store types.
- Project history never tracks, lists or serves a private file, and the blob
route serves only bytes a listed file had, so copies kept by an older
version are no longer readable.
- Lint reads a linked stylesheet, script, import map or sub-composition only
when the private-file rule allows it and the file is the kind its link
names, so no finding quotes a private file.
- The rule moves to @hyperframes/parsers/private-path so lint can use it;
@hyperframes/core/safe-path re-exports it unchanged.
* fix(studio-server): a refused history blob reads as not a history blob
* fix(core): keep realpath in core, so its file-system callers and tests see one module
* fix(studio-server): no write route touches files other tools run or trust
A composition previewed in Studio runs same-origin with the API, and the write rule
refused only private paths, node_modules and hidden names. Its script could create,
overwrite, rename or delete package.json, lock files, agent instruction files and
scripts, which a package manager, build tool or coding agent then runs with the
developer's rights.
Core now names these files in one predicate, toolTrustedFileKind, read in every
spelling a case-insensitive file system or Windows opens. Studio's write rule and the
registry installer's target rule both use it, so every write route (PUT, POST, PATCH,
DELETE, upload, duplicate-file, file-mutations and gsap-mutations) refuses such a
target. Deleting or moving a folder that holds one is refused too. Ordinary
composition, stylesheet, script-asset and media writes still succeed.
* fix(core): fold look-alike letters before matching files other tools trust
A case-insensitive macOS volume opens "AGENTſ.md" (long s) as AGENTS.md
and "Makefile" (fi ligature) as Makefile, but lower-casing alone does not
map either letter. Project path segments now fold names the way such a
volume does: NFKC, upper-then-lower case, and invisible characters
dropped. Every rule built on the segments gets the fold.
Also:
- name JSON, YAML and TOML tool config, compose files, turbo.json,
nx.json, pyproject.toml and gulp or grunt files
- treat a ~N name as a short name only when it has the 8.3 shape and an
extension a trusted file's could shorten to, so take~2.mp4 stays
writable
- refuse to delete or move a folder holding node_modules or .git, walking
it without the listing's ignore rules
- report an upload refused by the write rule in `skipped`
* fix(studio-server): a host that built a page on the build worker can exit
* test(studio-server): a dead build worker's exit code is left to the platform
* refactor(studio-server): every build-worker queue change sets its ref in one place
* fix(core): a .hyperframes folder in a subfolder stays private too
The private-file rule treated .hyperframes as the app's records folder only at
the project root. A project copied into a subfolder keeps its own
.hyperframes/, and its chat history, hand-off and other records there were
listed in the file tree and served by Studio's file and preview routes and by
the other file servers that share the rule.
A .hyperframes segment at any depth is now private, in every spelling Windows
opens. Studio's own files and the prepared-assets and requests folders stay
servable only at the root, the one place Studio reads and writes them. The file
tree also skips .hyperframes/backup at any depth.
* fix(cli): the capture --local-file page server answers only its own host
The page server that capture --local-file starts on a loopback port answered a
request whatever its Host header said. A site whose name resolves to 127.0.0.1
could then read the files it serves from the browser.
The server now refuses, before any file lookup, a request whose Host is not its
own loopback name and port, with the same isOwnLoopbackHost rule the static
project server and the engine file server use. Every response carries
X-Content-Type-Options: nosniff. The capture browser loads the page by
127.0.0.1 and its port, and fetchLocalPageUrl reads from disk, so both work as
before.
* fix(skills): a plugin manifest in the project pins no CLI release
The skills launcher read a plugin manifest from the folder above skills/
wherever the skills were installed. For a project-level install that folder is
the project itself or its .agents or .claude folder, so a plugin.json shipped
with a project could pin an old hyperframes release and turn off update checks
and standalone skill refreshes.
The launcher now reads a manifest only from a bundle of its own: a folder whose
name does not start with a dot and that is neither the folder the CLI works in
nor one that contains it. Plugin managers install bundles into their own
folders, so those keep their pin. A manifest anywhere else leaves the call the
same as with no manifest. A manifest that pins a release older than 0.9.34, the
launcher's floor, is refused. The agent-facing plugin rules and the skills
guide say the same.
* fix(skills): a plugin bundle inside the working folder pins only behind a dot folder
A package inside a monorepo could still pin the CLI: with the agent working at
the monorepo root, <mono>/packages/video/plugin.json sat beside a skills folder
the launcher took for a bundle.
The launcher now also refuses a bundle inside the working folder unless a
folder whose name starts with a dot lies between them, as .claude does in
~/.claude/plugins/cache when the agent works in the home folder. Bundles in
~/.claude/plugins, ~/.cursor/plugins and ~/.gemini/extensions keep their pin
from the home folder or a project. The plugin rules and the skills guide now
state the three conditions the launcher checks.
* docs(cli): a .hyperframes folder in a subfolder is left out and not served
The private-file rule now treats a .hyperframes/ folder at any depth as the
app's records. Files stored in one inside a subfolder, assets included, stay out
of uploads and are not served by Studio, preview or render. The cloud render
section now says so beside the rule it already describes.
* fix(cli): capture pages keep loading scripts with no or unlisted extensions
With nosniff on every response, a script the capture page server did not
know the type of (a .cjs file, or a file with no extension) was served as
application/octet-stream, and the browser refused to run it.
The server now names .cjs, .wasm, .xml, source maps, web manifests, CSV,
WebVTT, glTF and more audio, video and image types. A file whose type is still
unlisted is served without nosniff, so the browser sniffs it as it did before
and such a script runs. Every listed type and every empty error response keeps
nosniff.
* fix(share): sync and share leave out a .hyperframes folder below the root
The .hyperframes rules in the source path policy matched only at the project
root, so a copy's records in a subfolder, such as
sub/.hyperframes/agent-runs.jsonl, fell through to the default rule and
travelled with share and sync.
A new nested-hyperframes-state rule leaves out a .hyperframes folder at any
depth below the root. It sits before the first allow rule, as the contract's
rule order requires. The root folder's Studio files and other portable files
travel as before. Stored snapshot reads skip the new rule through
stored_manifest.legacy_rule_regexes, the shape the server and Desktop copies
already use, so snapshots and bindings captured earlier still read.
* test(studio-server): a sub-composition build past its deadline is refused off the request thread
* fix(studio-server): sub-composition previews build off the request thread within a deadline
* fix(studio-server): every parse of a sub-composition page runs on the build worker
* test(studio-server): a page built inside its deadline keeps the request thread free
* test(studio-server): a refused page leaves the next one built whole, hf-ids and all
* chore(studio-server): the build worker's comments fit its package's share
* fix(studio): recovery drafts are kept by each project's own server, never the browser
* fix(studio): a restored draft stays offered when Studio settles which composition is open
* fix(studio): a draft read never waits on a pipe or loads an oversized record
* test(studio): the draft read's child fixture is an entry point, and its note sits in the test
* chore(studio-server): size budget covers the recovery-draft route
* chore(cli): unpacked size budget covers the recovery-draft route
On Windows even a stat or realpath of a \\host\share path connects to that
host and signs in as the current user. Three readers looked such paths up
before refusing them:
- `hyperframes compositions` resolved a data-composition-src with
path.resolve, so an absolute or share src reached readProjectFile. It now
uses resolveCompositionSrc, the resolver preview and render use.
- The timeline reader (`timeline`, `audio`, measure) realpath'd a
sub-composition before its containment check. It now refuses a path on
another network root first.
- Artifact bookkeeping for render, check and snapshot looked up the paths
recorded in .hyperframes/video-state.json. A recorded path must now be
spelled inside a .hyperframes/artifacts/<kind> folder on the project's own
root before cleanup looks at it, and a record naming another share counts
as gone without a lookup.
* fix(studio-server): history restores stay inside the project and use only intact copies
* fix(studio-server): a history path's trailing dots and spaces are trimmed by a scan, not a regex
* wip: paused before review
* test: cover credential files on the preview routes and the deploy notice
* refactor: name the inspector's file lookup for what it serves
* fix(cli): a link to a private file outside the project keeps its own left-out reason
* fix: escape the upload notice and skip private files from old manifests
* fix(cli): history turn ids from the project folder are checked and sent encoded
* test(cli): history turn fixtures use the id shape the CLI mints
* fix(cli): a turn's label and agent name are kept as written; only ids are checked
* style(cli): format the turn reader
* refactor(cli): the turn reader checks a marker through small predicates
* chore: release v0.9.55
* chore: release v0.9.55 notes include #7830
* chore: release v0.9.55 notes include #7859 and #7867
* chore: release v0.9.55 notes name the docs change and cloud render
* chore: release v0.9.55 notes include #7866 and #7801
* chore: release v0.9.55 notes match the style bar