* feat(studio): graphite colour system with a light theme that passes contrast
Studio's tokens take the Graphite values: a lifted dark, a light set behind
data-theme="paper", the player dark in both. Every text role clears 4.5:1 and
every icon and ring 3:1 on the fills it sits on, held by the contrast ratchet.
* fix(studio): every studio surface follows the light theme and passes contrast
The dock no longer inherits dockview's forced dark scheme, popovers sit on the
raised surface, stock white, red and amber text map to roles, the keyframe
diamonds and track headers use roles, and the contrast ratchet measures every
pair in both themes (150 pairs).
* fix(studio): stock white, red and amber stay colours; studio text and fills use roles
Round-two review: white, red and amber are colour content as well as text, so
they keep Tailwind's values and hosts keep their palettes. Text, toasts, the
feedback card, the FX popover, destructive buttons, category labels and the dock
take Graphite roles instead; the contrast gate rejects an unparseable colour.
* fix(studio): every light-theme ink, icon, dot and focus ring holds its contrast
Stock white, red, amber and file-type hex left in Studio chrome now use roles or a light-dark pair; the file tree icons carry a contrast test.
* chore(studio): drop two comments the names already carry
* fix(studio): red marks, teal outlines and colour curves hold contrast in both themes
Danger graphics use the danger ink, file tree outlines use the accent ink, the colour-curve well stays dark like the scopes, and six timeline token pins come back.
* fix(studio): invalid readouts, the curve focus ring and editor dots stay distinct in both themes
The invalid readout border is dashed as well as red, the curve graph's ring sits inside its dark well, the radius dots and ease handle centre follow the theme.
* refactor(studio): ease and radius editors use colour classes; the dashed invalid border has a test
Simplify pass: the ease editor's inline accent styles become stroke and fill classes, the radius dot class is computed once, the ratio check is one expression, and the alias table is named for what it holds.
* fix(studio): the timeline viewport gate judges p95 over every measured scroll step
* fix(studio): format the pooled p95 lines in the timeline viewport gate
* test(studio): pin the timeline viewport gate's pooled verdict with tests
* refactor(studio): share one Chrome launcher across the timeline and fade-handle gates
* fix(studio): undo and redo are no longer held behind thumbnail renders
* test(studio): composition card tests share one mount, and the bench's 2 s undo rule is tested
* fix(studio): a composition card and its timeline clip share one thumbnail render
* test(studio): bank opening a film at 15 thumbnail requests, one per composition
* fix(studio): composition cards load near the viewport and keep their last frame until the next
* fix(studio): a composition card keeps its last frame under cache pressure, and loads 200 px early
* fix(studio): a rotate GSAP does not own saves where you let go, as the element's own CSS rotate
* fix(studio): a rotate keeps a transform-centred element in place
* fix(studio): a plain rotate shows its angle in the panel and turns inline elements too
* fix(studio): a rotate turns a mirrored or stretched centred element with the cursor
* fix(studio): keep the rotate's panel readout within the comment budget after the move landed
* fix(studio): an inline element's turn saves the inline-block it was drawn with
The overlay passes the rotate target it read at press into the commit, so the
save draws and patches exactly what the draft did instead of re-reading an
element the draft already made inline-block. The GSAP-aware commit also routes
on that press-time decision.
* test(studio): bank the 27 rotate cases a GSAP-free rotate now passes
baseline.json is the edit-accuracy-gate artifact of the CI run at cb75cc37,
copied byte for byte: 530 to 557 passing, all rotate-none cases, none
regressed.
* fix(studio): a word an older Studio turned keeps its inline-block and whole angle on a new rotate
Clearing the legacy rotation marks puts the element back to inline and drops
their turn, so the save reads the target again instead of using the one the
press read before the clear. A test also pins that a turn the press drew as CSS
stays on the CSS writer.
* test(studio): bank the edit accuracy baseline from the rebased rotate run
* fix(studio): a move keeps the route it chose at press through drop and save
* fix(studio): a refused move drops the transform GSAP folded it into, so the layer goes back
* fix(studio): a group refuses a GSAP takeover before writing any member, and only a real fold refuses
* fix(studio): resize an element without GSAP by its own width, height and translate
A corner resize of an element GSAP does not size or position now saves plain CSS:
its inline width and height, the scaled crop, and the translate that keeps its
centre where it was, in one save and one undo step, with no animation read and
no preview reload. It used to write a GSAP script into a file that had none, and
the box snapped back to its old size on release.
The anchor that keeps the centre planted is the same plain px translate a move
writes, kept to a thousandth of a pixel, and a percent translate is held as px
from the press so a growing box cannot drag it along on the first frame.
* fix(studio): a resize leaves a box to GSAP when GSAP tweens anything the CSS box writer sets
gsapWritesBox now reads its channels from BOX_SIZE_STYLE_PROPS, the writer's own
property set, instead of width and height alone. A host-level test pins the
GSAP-free resize route through useDomGeometryCommit.
* fix(studio): a GSAP scaleX or scaleY tween also keeps a resize on the GSAP route
They are GSAP's names for parts of the one CSS scale the box writer sets.
* test(studio): bank the edit accuracy cases a plain-CSS resize now passes
baseline.json is the edit-accuracy-gate artifact of the CI run at 5ae51f1d,
copied byte for byte: 494 to 530 passing, the 36 resize cases without GSAP,
none regressed.
* fix(studio): fade handles show a real tooltip and a 24 px target
A sound clip's fade-in and fade-out dots now open Studio's tooltip
("Fade in 0.3 s") instead of a native title the clip's own title
competed with, take a 24 px hit box, and paint an opaque white dot that
reads on the audio clip in both timeline themes.
* fix(studio): fade handles sit at the clip's top corners, above its content
The fade-in and fade-out dots now pin to the clip's top-left and top-right
corners instead of the end of each fade, paint above the waveform and
thumbnail layers, and share a short clip's width without overlapping. The
fade ramps are clipped to the clip's rounded corners, and the tooltip
uses the same seconds format as the slider's value text.
* fix(studio): fade dots keep their 10 px size on a very narrow clip
On a clip narrower than two fade hit boxes the flex row squeezed each dot
below 10 px. The dot no longer shrinks and no longer takes the pointer, a
browser check measures its box on a 0.1 s clip in CI, and two drag test
comments now describe the corner handles.
* fix(studio): the Escape key cancels a fade drag wherever keyboard focus is
A press on a fade handle never focuses it, so Escape only reached the
handle's own key handler when something else had focused it, and the
release then saved the dragged fade. The drag now listens for Escape on
the window for as long as it lasts, restores the fade and saves nothing.
* test(studio): trim the fade handle browser check header to what the code cannot say
* fix(studio): a translate written with min(), max() or clamp() moves without a jump at press
* fix(studio): refuse out loud a translate Studio can't read, and resolve percent against the transform box
* fix(studio): one refusal per held arrow key, and the panel reuses the offset it already read
* fix(studio): open projects whose folder name has a colon
* fix(studio): open projects named with a colon or '#' in Studio and its preview
* fix(studio): refuse a Windows stream id reached through a session alias
One adapter rule, isServableProjectId, now guards the requested id, a
session's projectId and the folder listing. The bundler-failure preview
places its <base> inside an attributed <head>, and the media metadata
cache keys on the encoded URL so ids with ':' can't collide.
* fix(studio): the properties panel no longer hands GSAP an element it does not position
* refactor(studio): drop the comment the comment-share ratchet counts
* fix(studio): the panel also skips GSAP's transform-origin channels on an element GSAP does not position
* refactor(studio): delete the unused second list of GSAP transform channels
* feat(studio): hosts can apply the canvas menu's z-order through useDomEditZOrder
* style(studio): format useDomEditZOrder
* docs(studio): trim the z-order hook's moved comments
* fix(studio): a read-only preview never reorders or deletes, from any host control
The session's z-reorder and delete commits now refuse in a read-only preview, so the lane mirror is skipped too. useDomEditZOrder is off in read-only and for a detached element, apply reports false when nothing is saved, a same-class sibling keeps its selector index, and the hook returns a stable object.
* fix(studio): read-only stops canvas and host z-order only, and a reloaded selection
The session's z-reorder and delete commits no longer refuse in a read-only preview: the timeline's lane-to-z sync goes through the same commit and stays editable. The refusal stays in useDomEditZOrder and the overlay. A selection from before a preview reload, whose document no window shows, gets no z-order.
* fix(studio): one Cmd+Z pressed inside the preview takes one undo step, not three
* fix(studio): one undo per key press, tracked, with the preview's extra history listener removed
* fix(studio): start timeline drops on an empty lane at 0 and snap to clip edges
A file, asset, block or composition dropped on an empty lane, a new lane, or an empty timeline now starts at 0 instead of the pointer time. On a lane with clips the start snaps to the nearest clip edge within TIMELINE_SNAP_PX. The drop preview and the committed drop share the placement.
* fix(studio): keep the drop file's comment share flat
* fix(studio): a drop snaps to clip edges only while the timeline magnet is on
* fix(studio): undo and redo show on the next frame instead of after the server answers
* fix(studio): an undo pressed before the history caught up still restores in place
* fix(studio): undo and redo of an edit inside a nested composition apply in place
* fix(studio): undo of a nested edit reloads when its composition has a GSAP script
* refactor(studio): a nested restore with a GSAP script has no in-place scope
* fix(studio): an undo after an outside write shows the file the server restored
* feat(studio): a host app can mount the dock with its own panels, storage key and width
* refactor(studio): the dock keeps a host layout through Studio's own preference storage
* fix(studio): a dock reopens a panel beside the preview only when its usual neighbour is not in the dock
* fix(studio): move and nudge an element without GSAP by its own CSS translate
* fix(studio): route a GSAP-free move through the on-element writer and meet the file checks
* fix(studio): a drag press on a GSAP page leaves an element GSAP does not position alone
* fix(studio): a press reads a percent or calc translate without touching the element
* refactor(studio): import POSITION_CHANNELS from its owner instead of a re-export
* test(studio): share the drag press helper in the start gesture tests
* refactor(studio): split the translate length reader into term and sum
* test(studio): cover each translate form the reader resolves or hands to Chrome
* test(studio): bank the edit accuracy cases plain-CSS moves now pass
baseline.json is the edit-accuracy-gate artifact of the CI run at 2a84c103, copied byte for byte: 60 newly passing cases, none regressed. The commits since change no behaviour (a reader split and tests).
* fix(studio): a text edit saves to the element that was edited, or says why it could not
* fix(studio): saving a text edit keeps whatever else is selected as it was
* test(studio): gate edit accuracy against the base branch baseline
Work in progress: the ratchet and its tests; the CI job follows.
* test(studio): run the edit accuracy gate in CI across eight shards
Shares the built CLI from the Build job, re-runs twice each case that passes on the base branch and failed, and posts one sticky comment from the gate job.
* test(studio): mark the ratchet accuracy check as intentionally branchy for fallow
* test(studio): confirm newly passing edit accuracy cases 2 of 3 before banking them
The shards re-ran only regressions, so one lucky pass of a base-failing case failed the gate as unbanked and was never listed as unstable. Every case whose verdict differs from the base branch, either way, is now re-run twice and judged 2 of 3. A render error fails the gate like an unsettled snapshot, a malformed baseline.json fails the job instead of reading as empty, and the header is 3 lines.
* ci(studio): run the edit accuracy gate for every package the bench builds or imports
* ci(studio): also run the edit accuracy gate for lint, sdk and the Studio and CLI build inputs
* ci(studio): run the edit accuracy gate when a bench package's manifest or exports change
* test(studio): bank the 120 edit accuracy cases that sub-pixel dragging now passes
* fix(studio): keep sub-pixel position and size while dragging and on drop
* test(studio): expect three-decimal rotation in the manual edits rotation test
* fix(studio): keep resized boxes on whole pixels so a centred element stays put
* test(studio): measure render drift against a producer frame in the edit accuracy bench
Finds the target's box in a producer-captured frame by its flat colour, with sub-pixel edges from area moments, and compares it with the reloaded preview. The baseline gains the render column from the same full run on main.
* test(studio): mark the pixel box scan as intentionally complex for fallow
* test(studio): regenerate the edit accuracy baseline with render drift on main
Same full run on main 14e87005d as #4760's baseline, with the render column: render passes 618/660, 70/660 pass, 314/660 pass everything but smoothness.
* test(studio): fail only render drift when the producer cannot render a case
A producer capture failure used to turn the case into a harness error that wiped every accuracy metric. It now sets render to null with renderError, fails render alone, and is counted in the table.
* test(studio): shorten the render drift box comment to two lines
* test(studio-server): history prune tests allow a file time a moment past Date.now()
* ci: temporary probe for the history prune test flake
* fix(studio): timeline scroll-end timer no longer fires after unmount
* ci: remove the temporary history prune probe
* test(studio): check the virtualizer teardown with a spy and a 200 ms wait
Page-side shader transitions clone the outgoing and incoming scenes,
ids included, into staging layers. Color grading grades the copied
<video> too, and findInjectedRenderFrame looked its frame up with
getElementById, which returns the live scene's <img>. The live and the
copied grading entries then both kept their canvas right after that one
<img>, moving each other on every mutation, and the loop blocked the
page: a graded <video> inside a transition stopped the render at the
transition's first frame.
Look through the element's following siblings first, where the injector
puts the frame, and fall back to the id lookup only when it isn't there.
* fix(parsers): moving a clip on the timeline moves its inner animations too
* fix(parsers): a timeline move keeps chained and partly known tweens where they belong
* fix(parsers): moving a clip leaves chained outside tweens and template aliases in place
* fix(studio): a clip timing edit syncs its animations once, not twice
* fix(studio): a timing edit falls back to the server when the SDK wrote no animation change
* fix(studio): the SDK sync check reads scripts inside a template too
* fix(studio): the SDK sync check finds template scripts through the DOM walker
* fix(core): a clip retime writes only its own timed tweens, never outside ones
* style(core): format the clip tween import
* fix: readers never see an empty or half-written project file
* test: pin the create routes and block install to whole-file writes
* fix(sdk): a write through a link to a missing file creates that file
* fix: saves through a linked folder reach the real file, atomic writes handle long names
* test(sdk): check for the stray file, not the folder listing
* fix(core): creates work on Windows FAT drives, and links through folders resolve as the system does
* fix: agree the helper move in the deletion check, and match temp paths without a path regex
* test: check POSIX file modes only off Windows
* fix(studio): resizing a cropped element keeps the same part of it in view
Rebased onto main as one commit (was a0a86cc1..080dfecf). Merged with the crop
release rule from main: a drag that ends where it started saves nothing.
* fix(studio): the crop follows each axis GSAP does not tween, while dragging and once saved
* fix(studio): the live-tween check reads every timeline, not only the first after a soft reload
* feat(studio): hosts can add their own items to the clip right-click menu
* refactor(studio): let the clip menu item names carry their meaning
* fix(studio): a clip menu item that moves the focus keeps it after the menu closes
* test(studio): the clip menu focus hand-off holds under StrictMode
* docs(studio): state the menu focus rule without history
* fix(studio): save moves in GSAP-free files without waiting 600 ms for a warm parse
* fix(studio): read a GSAP-free file once instead of re-fetching it as a cold parse
* perf(studio): stop sorting the font list while the font picker is closed
* test(studio): keep the font list test off the network and clean up on failure
* test(studio): stub the font APIs for every font field test and clear the head after each
* test(studio): start an edit accuracy benchmark for manual editing (work in progress)
* test(studio): an edit accuracy benchmark for manual editing
Drives real pointer and keyboard gestures in the built CLI Studio on generated one-element projects and scores tracking, drop jump, reload drift, undo fidelity and smoothness in composition px.
* test(studio): move render drift out of the first benchmark PR
Render drift against a producer frame ships in the stacked follow-up so this PR stays near the size budget.
* test(studio): keep the edit accuracy bench clean under fallow
Builds the grid as a product of its axes, drops unused exports and marks the harness functions that fallow scores as complex.
* test(studio): report edit accuracy with the Studio and bench commits it measured
Moves scoring and the report files into report.mjs, stamps both commits at start, counts undo byte failures apart from the worst box, and survives Studio swapping the preview iframe mid-read.
* test(studio): bank the edit accuracy baseline measured on main
660 cases on main ae914334f: 0 pass all six checks, 304 pass everything but smoothness. Five cases that hit a harness error in the full run were rerun and spliced in.
* test(studio): state the edit accuracy build prerequisite in the runner header
* test(studio): keep the edit accuracy baseline one line per case under the formatter
* test(studio): store edit accuracy baseline values rounded up
Rounding to the nearest 0.01 stored seven 0.502 px tracking results as 0.5, which reads as a pass at the 0.5 px limit.
* test(studio): score edit smoothness on dropped frames and main-thread work per frame
* test(studio): measure edit accuracy on a still preview and score press jump
Snapshots wait for the preview frames and the box to hold still for 1 s, so Studio's asynchronous reload after a save is measured, not raced. A preview that never settles fails the metrics it feeds instead of erroring.
Adds press jump (box at press against the box before it, 0.5 px). Smoothness counts only dropped frames beyond a blank-page control driven the same way in the same Chrome; unknown trace data fails smoothness alone.
The runner takes the suite lock per chunk of 8 cases (--lock), reads the port Studio announces, keeps undo history and tmp projects inside the case dir and removes them on a signal, and stamps the CLI tree's commit and bundle hash.
* test(studio): stamp the edit accuracy build with the Studio bundle as well as the CLI
The CLI bundle hashed the same for two different Studio builds; Studio's index.html names its hashed assets, so the stamp now covers it.
* test(studio): regenerate the edit accuracy baseline on main after the settle fix
Full grid on main 14e87005d with the settle fix, press jump and the smoothness control: 70/660 pass, 314/660 pass everything but smoothness, 0 harness errors, 0 unsettled. Scored without render, which #4764 adds.
* test(studio): wait for each edit accuracy undo write before sending redo
Undo and redo each wait up to 15 s for their own write, like the settle; redo is sent only once undo landed, and a write that never lands fails undo with undoTimeout instead of scoring an untested redo. The reload waits for the files to hold still. The headline is now the accuracy count, with smoothness on its own line against the control. The settle compares each read with the window's first read, and a CLI outside a git tree no longer stops the run.
* ci: install a pinned static ffmpeg on linux instead of apt
* ci: run the jobs that use a shared action when that action changes
* ci: drop the ffmpeg cache so the install is bounded by the download alone
* ci: cap ffmpeg download retries so a Retry-After header cannot stretch the install
* fix(studio): a click after an empty-canvas press selects on the first try
* fix(studio): a shift press outside the box no longer sets a click guard that never fires
* fix(studio): a copied clip keeps its look and motion, and a duplicated section keeps its name
* fix(studio): a copy's tweens start where their originals do and stay in the timeline's scope
* fix(studio): a copy takes only tweens it can place exactly, and never moves the film's own
* fix(studio): a copy leaves out a tween that sits inside a callback or a guard
* fix(studio): a copy leaves out a set that runs only inside an if or an else
* chore(studio): keep the copy change under the comment ratchet
* fix(studio): name a repeated section by its authored id in the Design panel
* fix(studio): the timeline's page-read fallback also names a repeated section by its authored id
* fix(studio): undoing an agent's build takes its sections off the timeline too
* fix(studio): a reloaded preview's timeline is read from the new document, sections included
* fix(studio): a section still in the preview stays on the timeline, and trims are unchanged
* feat(studio): a host can keep only Select and Split in the timeline tool menu
* refactor(studio): drop the prop comment the name already carries
* feat(studio): select rightward includes clips crossing the playhead
* feat(studio): a popover can point at its trigger with an optional arrow
* fix(studio): keep a side arrow off the popover corner
* refactor(studio): name the side arrow corner clearance instead of commenting it
* refactor(studio): the arrow corner clearance applies on every side
* fix(studio): a family's saved font faces keep their own weight and style
* fix(studio): keep the font face change within the comment ratchet
* fix(studio): a saved font face cannot break the page, and only weights a browser keeps are written
Miguel Ángel