transcribe.cjs: resolveUvx() resolves the WhisperX runner — checks `uvx` on PATH then
~/.local/bin/uvx; if absent AND EC_INSTALL_UV=1, installs uv via the official standalone
installer (single binary, no Python/npm) and re-probes; otherwise prints how to enable it
and returns null so the caller uses the bundled whisper.cpp. The whisperx gate is now
`if (UVX)` (was `if (wantWx)`), so a missing uv degrades to whisper.cpp instead of erroring.
SKILL.md: drop the misleading "no manual installs" — uv is NOT bundled and a stock
hyperframes install usually lacks it; document the whisper.cpp fallback + the EC_INSTALL_UV
opt-in.
Verified: node --check; resolveUvx returns the path when uv present, prints the fallback
message + returns null when absent (not opted-in).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Remove modes/cinematic/_archive/ (champion / memory-wall / portrait-header
spec.md + template.html) — retired pre-DNA template shells that shipped to
every install as dead weight, redundant with git history, and out of step
with the skill's own convention that retired material lives OUTSIDE the skill
(~/Downloads/embedded-captions-archive/). Recoverable from git history.
Scrub the only two references to the _archive path:
- modes/cinematic/README.md: drop the 'archived templates in _archive/' clause
- scripts/make-composition.cjs: legacy-path comment no longer points at _archive
The memory-wall / champion design-pattern citations across references/* are
left intact — they describe live patterns + files in references/example-renders/,
not the deleted dir.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
test-set.md was reachable by nothing — no doc links it and no script
reads it, so an agent could never open it — AND its content is stale:
it documents the retired 54-template system (TEXT_SHADOW / BLEND_MODE /
CAP_COLOR placeholders, a 2026-04-23 bug log) and points at local-only
clips (~/Downloads/heygen_relevant_videos/) that aren't in the repo.
Recoverable from git history if the regression-corpus idea is revived.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
38e8d9f4 (integrate 12 identities) rewrote this section from a base
without b6d3a9c4's fix, restoring the stale "ten/six visual languages"
+ "Both modes draw from dna/" text — leaving it inconsistent with the
rest of the now-modernized file (48 identities, two engines, Standard
retired).
Restore the pointer-style section, aligned to the current vocabulary:
name both registries (dna/ = 10 column-flow Cinematic, themes/ = 38
themed), reinforce CATALOG.md as the single routing surface, keep the
hero-three-act note. The full DNA table stays in dna/README.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The description is a plain (unquoted) YAML scalar that contains ": " (e.g.
"two engines: column-flow", "Pipeline: transcription"), which a strict YAML
parser rejects with "mapping values are not allowed in this context". Wrap the
value in single quotes (no apostrophes inside, so no escaping needed). Verified
the frontmatter now parses and the description value is unchanged (quotes
stripped). Pre-existing on main; surfaced after the recent edits.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The skill is now fully CJK-free — verified across both repo copies over all text
files (kana, ideographs, halfwidth + fullwidth forms). All changes are cosmetic
except one intentional behavior change (the brush seal).
- CATALOG.md / SKILL.md: chinese trigger aliases and shortlisting heuristics
translated to english keyword equivalents (explode/VFX, terminal/console,
ink-wash/brush, ...). chinese-prompt routing is dropped by design.
- themes/{arcade,aurora,biolume,ransom,spectrum,vhs}.json: `when` aliases -> english.
- make-theme.cjs:
- one comment translated to english.
- brush seal default glyph (a kanji) -> a monogram of the climax word:
(heroDisplay.trim()[0] || "*").toUpperCase().
- cpslam setpiece default tag (katakana) -> "CYBERPUNK" (latin); cpslam is a
registered-but-currently-unused setpiece, so no shipped theme is affected.
- themes/brush.json: drop the explicit kanji sealGlyph so it uses the monogram default.
Behavior: only the brush seal changes (now e.g. "R" for "remarkable"); every
other shipped theme produces byte-identical output. Verified: make-theme.cjs
syntax ok, all 7 theme JSONs valid, brush + nightcity compile, brush seal renders
a latin monogram, zero CJK in generated HTML.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The dir held the shared font bundle (fonts.css + woff2 + build-fonts-css) and the
_motion/_anatomy reference catalogs — all used by BOTH the Cinematic and Theme
engines, so 'standard' was a misnomer after the Standard engine's retirement.
Pure rename (git mv, 48 files, fonts.css byte-identical) + path updates in the 5
real consumers (inject-fonts, make-theme, preview-frames, measure-layout font
embedding) and 3 comments/meta. build-fonts-css uses __dirname-relative paths so
it needed no change. Verified: make-theme still embeds @font-face (3 blocks, 0
'not bundled' warnings), inject-fonts + preview-frames resolve the new path.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Standard mode was retired 2026-06-12 (anchor replaced it) but stale references
lingered in ~15 files. Rewrote all misleading ones to current reality:
- routing advice that said 'use Standard mode (opaque rail)' for bright scenes →
the anchor theme / ink (references: bespoke-vs-presets, test-set,
typographic-moves, motion-vocabulary, direction-catalog, composition-craft,
aesthetic-principles)
- CATALOG no longer offers 'Standard or Cinematic or Theme?' as a non-question;
two-engine framing throughout
- gate/script comments + runtime messages: 'Standard mode only' → 'rail-surface
identities'; dangling PIPELINE.md citations removed (it was archived)
- make-composition mode=standard guard now states the retirement + points to
make-theme/anchor instead of the deleted make-standard.cjs
- themes/README + anchor.json + build-fonts-css wording
Remaining 'Standard' mentions are only retirement notes, the NBS Hershey-font
license, and the modes/standard/ dir path (kept — renaming churns many font
references). Lint clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Per user request: drop the ffmpeg noise= pass from _postfx.sh. Captions/footage
are no longer unified by a shared grain field. Themes whose ONLY plate effect was
grain now emit NO _postfx.sh — final_fx becomes a clean lossless copy of final.mp4
(no needless second encode). punch (windowed) and pushIn paths keep their camera
move, just without grain. plate.grain in the DNA JSONs is now inert.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
final_fx was softer than final.mp4 because the plate-reaction zoompan ran a
scale-up→resample→scale-down roundtrip on EVERY frame (a mild low-pass even at
z=1), yet a punch only moves the camera in a ~0.5s window. Now the punch path
splits the clip and runs zoompan ONLY on the active window; the head/tail pass
through with grain only (no geometric resample) so the static 95% stays as sharp
as final.mp4. Handles 2-seg (apex at clip start/end) and 3-seg (apex mid) cases;
falls back to whole-clip when the window spans nearly all of it. pushIn
(continuous zoom) and grain-only paths unchanged. Verified: passthrough region
now pixel-matches final crf11; apex window still carries the punch/shake.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Brings the branch skill to the full 48-identity catalog (10 column-flow + 38 theme):
interface family (breaking/finance/cover/blueprint/xray/glass/magnet) + uncanny
family (constellation/seance/mirror/brand/redline), their paradigms/setpieces in
make-theme.cjs, SKILL.md count + retired-Standard doc cleanup, 10 dead decls
removed. 12/12 new themes compile in-repo; 16/16 fixture regression green.
make-theme.cjs is now 702 KB (38 themes' worth of generators) — over the 500 KB
non-LFS gate; committed with HF_MAX_NONLFS_KB override (it is reviewable source,
not a binary; LFS would break diffs). Candidate for a future per-family split.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The section duplicated dna/README.md's table and carried claims from
before the Theme engine existed:
- title said "ten" but the body said "six"; "Both modes draw from
dna/" is wrong — Theme loads from themes/, not dna/
- it presented dna/ as the only DNA registry, omitting themes/ entirely
Replace the duplicated table with a pointer naming both registries
(dna/ = Cinematic, themes/ = Theme), reinforce CATALOG.md as the single
routing surface, and keep the hero-three-act note. No content lost —
dna/README.md already holds the full, maintained table.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
set-version's stable-release next-steps printed `git push origin main --tags`,
which pushes every local tag and fails the whole push on any pre-existing tag
(it broke the v0.6.107 release). #1517 fixed CONTRIBUTING.md + annotated the tag
but missed this console message. Now prints `git push origin main` +
`git push origin v<version>`, matching the pre-release branch.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(lint): promote rules to errors with registry exemptions and false-positive fixes
- Export isRegistrySourceFile/isRegistryInstalledFile from composition.ts
- Add registry exemptions to google_fonts_import and font_family_without_font_face
- Add registry exemption to requestanimationframe_in_composition
- Fix timed_element_missing_clip_class: data-track-index alone no longer triggers
- Fix caption_transcript_parse_error: balanced-bracket scanner replaces non-greedy regex
- Fix missing_timeline_registry: skips sub-compositions and template-wrapped files
- Fix scene_layer_missing_visibility_kill: strip JS comments before pattern matching
- Fix gsap_css_transform_conflict: exempt from() alongside fromTo()
- Fix gsap_from_opacity_noop: only fires when opacity value is actually 0
- Add regression test for data-track-index-only elements
* test(lint): add regression tests for false-positive fixes
Covers the 7 missing negative-case assertions flagged in PR review:
- registry marker suppresses google_fonts_import + font_family_without_font_face
- registry marker suppresses requestanimationframe_in_composition
- isSubComposition suppresses missing_timeline_registry
- scene_layer_missing_visibility_kill: fires, commented-kill fires, real kill suppresses
- gsap_css_transform_conflict: from() exempt alongside fromTo()
- gsap_from_opacity_noop: non-zero opacity (e.g. 0.5) is a valid reveal, not a noop
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(examples): fix warm-grain template to pass promoted lint rules
- index.html: remove undeclared "Lexend" from font-family stack
- intro.html: replace Google Fonts @import with bundled Inter font
- captions.html: quote TRANSCRIPT keys for valid JSON + use Inter font
Fixes CLI smoke CI failure after google_fonts_import, font_family_without_font_face,
and caption_transcript_parse_error were promoted from warning to error.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(cli): resolve warm-grain from repo registry in dev mode + bundle at build
getStaticTemplateDir now falls back to registry/examples/<id> in dev mode
so CI smoke tests use the PR-branch copy instead of fetching from main.
build-copy.mjs copies warm-grain to dist/templates/warm-grain at build time
so packed CLIs can scaffold it offline.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(examples): remove trailing comma from warm-grain TRANSCRIPT array
JSON.parse rejects trailing commas (valid JS, invalid JSON).
caption_transcript_parse_error was still firing because of the comma
on the last entry after quoting all keys.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The acorn addKeyframeToScript mixed ms.overwrite + ms.appendLeft on the
same _auto endpoint node, crashing MagicString ("Cannot split a chunk
that has already been edited") whenever an interior keyframe adjacent to
an _auto 0/100 endpoint introduced a new backfilled prop — the common SDK
path. It also replaced (not merged) existing keyframes, dropped ease and
_auto markers, corrupted commas on multi-prop backfill into empty {},
used a <0.001 percentage tolerance instead of recast's PCT_TOLERANCE=2,
and silently no-op'd on flat (non-keyframe) tweens.
Rebuild the node model to mirror recast: compute the FINAL property record
for every changed keyframe value node (target merge, _auto endpoint sync,
backfilled siblings) against the original AST, then emit exactly one
ms.overwrite per changed node (one insert for a brand-new key). No node is
ever both overwritten and appended into, so splices can never overlap.
- Merge: re-touching an existing keyframe merges new props over the
existing record, preserving untouched props, existing ease, and _auto.
- Convert-flat: first keyframe-add on a flat to()/from()/fromTo() tween
rebuilds its vars object to percentage keyframes (ease->easeEach,
ease:"none", from/fromTo->to) matching recast, then re-locates via the
-from-/-fromTo- -> -to- id fallback.
- Tolerance: PCT_TOLERANCE=2 for existing-keyframe detection.
- Shared serializeValue/safeJsKey for keyframe values (recast parity); the
tween-statement path keeps its local serializer for object/boolean extras.
- keyframeBackfill: only backfill props with a real numeric default; skip
unknown/string props so color:0 / filter:0 are never emitted.
- setGsapKeyframe move-path threads the same backfill defaults as the add
path so both entry points behave identically.
Differential tests (acorn vs recast parsed keyframe arrays) cover the
crash (2-endpoint + 0/25/100), empty-{} multi-prop backfill, merge with
extra props + ease, flat to()/fromTo() convert, "50.0%" non-byte-equal
key, near-% tolerance, and _auto-marker preservation onto an endpoint.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(cloud): add managed cloud rendering guide + fix flag reference
Add a dedicated guide for the managed `hyperframes cloud render` path
(HeyGen-hosted, zero-infra) at docs/deploy/cloud.mdx, covering auth/setup,
the zip→upload→render→download flow, templates via --variables, webhooks /
fire-and-forget, render management, and idempotent retries. Register it at
the top of the Deploy nav group and link it from the local Rendering guide.
Also fix a stale flag reference in the CLI docs: the `cloud render`
`--resolution` row listed the local-render presets (landscape/portrait/...)
but the cloud command only accepts `1080p`/`4k`, and `--aspect-ratio` was
missing. Verified against `hyperframes cloud render --help`.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(cloud): correct aspect-ratio wording and flow-diagram status
Two accuracy fixes from review:
- `--aspect-ratio` is only auto-detected for a local project dir; for
`--asset-id`/`--url` there is no local composition, so detection is
skipped and the server defaults to 16:9. Reword both the guide and the
CLI-reference rows to say so.
- The flow diagram showed status `done`, which is not a real value
(HyperframesRenderStatus is queued | rendering | completed | failed).
Use `completed` and re-align the box.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- set-version: create the release tag with `git tag -a -m` instead of a
lightweight `git tag`, which fails ("no tag message?") when a contributor has
tag.forceSignAnnotated / required-annotation set globally — it silently broke
the v0.6.107 tag step.
- CONTRIBUTING: replace `git push origin main --tags` (pushes every local tag →
whole push rejected on any pre-existing collision) with pushing the specific
tag, and document the monotonicity guard (stale higher tag blocks tagging).
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(studio): serialize GSAP script commits per file (shadow request race)
Rapid GSAP edits (ease/duration/keyframe/property) fired overlapping
read-modify-write POSTs to one script file — coalesceKey only dedupes edit
history, not requests. The gsap_fidelity shadow then diffed an op against
whichever POST's scriptText resolved, which could predate that op → false
"expected null, actual power2.out" mismatches. Server persists correctly; a
pure client request-pairing race.
Adds createKeyedSerializer (per-key promise chain, rejection-safe, self-
cleaning). commitMutation now serializes every GSAP-script commit per target
file by default (key `gsap-file:<path>`) — covering all op types and all
animations, not just one meta family — so same-file POSTs can't interleave.
Distinct files run concurrently; an explicit serializeKey still overrides.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(studio): dedup shadow numeric-equal + GSAP script extraction
Code-review cleanup (no behavior change):
- Extract the relative-epsilon float compare into shared sdkShadowNumeric.relEqual,
used by both timing parity (sdkShadow) and GSAP value fidelity (numericEqual) —
was duplicated verbatim, risking divergent tuning.
- Export extractGsapScript from sdkShadowGsapFidelity and import it in the keyframe
shadow instead of the byte-identical clone (the regex + marker set must stay in
sync with document.ts; one copy is safer).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Rapid GSAP edits (ease/duration/keyframe/property) fired overlapping
read-modify-write POSTs to one script file — coalesceKey only dedupes edit
history, not requests. The gsap_fidelity shadow then diffed an op against
whichever POST's scriptText resolved, which could predate that op → false
"expected null, actual power2.out" mismatches. Server persists correctly; a
pure client request-pairing race.
Adds createKeyedSerializer (per-key promise chain, rejection-safe, self-
cleaning). commitMutation now serializes every GSAP-script commit per target
file by default (key `gsap-file:<path>`) — covering all op types and all
animations, not just one meta family — so same-file POSTs can't interleave.
Distinct files run concurrently; an explicit serializeKey still overrides.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(sdk): setStyle removes hyphenated properties (was kebab/camel key mismatch)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(sdk): agree removeElement/getElement on duplicate bare ids
A bare hf-id duplicated across a sub-composition element and a top-level
element resolved to different instances: removeElement → resolveScoped →
querySelector (document-order-first, the inner sub-comp dup) while getElement
preferred the canonical match (scopedId === id, the top-level dup). So
removeElement(bareId) removed the inner instance and getElement(bareId) still
found the surviving top-level one — they disagreed.
resolveScoped now resolves an ambiguous BARE id to the canonical (top-level)
instance via isCanonicalScope (walks ancestors for isNewHostBoundary), falling
back to document order when no canonical match exists — matching getElement.
Fully-scoped paths (hf-host/hf-dup) and non-duplicated bare ids are unchanged.
Surfaced by SDK shadow parity (op:delete expected removed, actual present).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(studio): suppress shadow-parity false positives in timing + text
runShadowTiming: compare start/duration with a relative epsilon (1e-6)
instead of exact equality so float-precision drift (3.1 vs
3.0999999999999996, 21.36 vs 21.360000000000014) no longer flags; a real
difference (3.1 vs 3.5) still flags. trackIndex stays exact.
property:text resolver: trim both sides (snapshot.text is already trimmed)
and collapse empty-string vs absent (null) text so trailing-whitespace and
empty-vs-null no longer flag. Genuine text differences are unaffected; the
per-keystroke length lag is a caller-side debounce concern.
Adds tests for both fixes plus regression tests documenting two REAL SDK
divergences the shadow correctly surfaces (transform-origin removal no-op;
duplicate-bare-id delete resolution) — flagged, not fixed here.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(studio): shadow telemetry for GSAP keyframe ops (gsap_keyframe)
Wire the SDK shadow-parity telemetry to cover GSAP keyframe add/remove,
the primary unwired cutover signal, plus a defensive unmapped-PatchOperation
guard.
New packages/studio/src/utils/sdkShadowGsapKeyframe.ts:
- ShadowKeyframeOp + keyframeOpToEditOp: maps studio percentage-based keyframe
ops to SDK EditOps. add -> addGsapKeyframe{position:percentage}; remove ->
removeGsapKeyframe{keyframeIndex}, resolving percentage -> index against the
pre-op script with ~0.001 tolerance and a no-op-on-ambiguity guard for
duplicate-percentage keyframes (PR #1498 landmine).
- gsapKeyframeFidelityMismatches: reuses gsapFidelityMismatches for the
tween-level diff and layers a keyframe-array comparison (which the base diff
doesn't inspect), matched by GSAP animation id.
- runShadowGsapKeyframeFidelity: serialize-diff runner emitting op tag
gsap_keyframe (no keyframe reader on ElementSnapshot, so no existence path).
useGsapKeyframeOps synthesizes shadowKeyframeOp for addKeyframe /
addKeyframeBatch / removeKeyframe; the commit chokepoint dispatches the
keyframe-fidelity diff alongside the existing tween-fidelity path.
sdkShadow.ts: runShadowDispatch now emits dispatched:false reason:unmapped_type
if a future PatchOperation type ever escapes patchOpsToSdkEditOps, so the gap
surfaces in telemetry instead of vanishing.
Tests: sdkShadowGsapKeyframe.test.ts (18) covers index resolution, op mapping,
the ambiguity guard, the keyframe-aware diff, the runner, and the unmapped-type
guard.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
runShadowTiming: compare start/duration with a relative epsilon (1e-6)
instead of exact equality so float-precision drift (3.1 vs
3.0999999999999996, 21.36 vs 21.360000000000014) no longer flags; a real
difference (3.1 vs 3.5) still flags. trackIndex stays exact.
property:text resolver: trim both sides (snapshot.text is already trimmed)
and collapse empty-string vs absent (null) text so trailing-whitespace and
empty-vs-null no longer flag. Genuine text differences are unaffected; the
per-keystroke length lag is a caller-side debounce concern.
Adds tests for both fixes plus regression tests documenting two REAL SDK
divergences the shadow correctly surfaces (transform-origin removal no-op;
duplicate-bare-id delete resolution) — flagged, not fixed here.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
## What
Kills two false-positive classes in the SDK shadow GSAP value-fidelity diff (`sdkShadowGsapFidelity.ts`).
1. **Float precision** — `numericEqual` compared exactly, so SDK-computed `3.0999999999999996` vs server `3.1` flagged as drift. Now a relative epsilon (`abs(a-b) <= 1e-6 * max(1,|a|,|b|)`); real `2` vs `1` still flags.
2. **Selector-form divergence** — `[data-hf-id="X"]` (SDK writer) vs `.class`/`#id` (server writer) for the same element produced phantom `present`/`absent` pairs. `makeSelectorResolver` now keys tweens by resolved element (incl. nodes with no `data-hf-id`), unifying the forms.
## Why
Surfaced by production SDK-shadow parity telemetry — `gsap_fidelity` was the noisiest real-traffic op; both are diff-harness artifacts, not SDK drift.
## Tests
Epsilon (clean + real-drift) + selector-unification for `#id`/`.class`/`[data-hf-id]`.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
* feat(core): add param-substitution utility for GSAP timeline inlining
U1: clone + shadow-aware identifier substitution over acorn ESTree, plus
provenance tagging and a GsapProvenance type. Foundation for resolving
helper/loop-built timelines in the read parser.
* feat(core): inline helper-built and bounded-loop GSAP timelines
U2: expansion pre-pass that rewrites the analysis AST so a helper called N
times, a literal-bounds for-loop, a for-of, or a forEach over an inline array
each become concrete per-call/per-iteration tl.* statements with substituted
positions and provenance tags. Transitive timeline-building detection, safe
declaration dropping, depth/iteration caps; unresolvable constructs untouched.
* feat(core): resolve computed GSAP timelines in the read parser
U3: parseGsapScriptAcorn runs the inlining pre-pass before analysis, so
helper-built and bounded-loop timelines resolve at true positions with
motionPath arcs recognized; each tween carries provenance. Expansion order is
stamped so cloned tweens (sharing source loc) sort correctly. Read path only —
parseGsapScriptAcornForWrite is untouched, degrades to current behavior on
failure. The add-to-basket addCycle case now yields 7 resolved animations.
* feat(studio): runtime-authoritative keyframes for dynamic timelines
Phase 2 (U4-U6): the live-runtime scanner returns tween-relative keyframes
with per-tween timing and converts them to clip-relative when given clip dims,
fixing the timeline-vs-clip-relative bug; it extracts motionPath into arcPath
(shared buildArcPath) so the Arc Motion panel activates for data-driven arcs;
the cache leaves statically-unresolvable tweens to the runtime scan. Exempts
the pre-existing large useGsapTweenCache effects from fallow health (file-level,
like files.ts) rather than suppression comments.
* feat(studio): surface keyframe editability from provenance
U9: editabilityForProvenance(provenance) -> direct|unroll|override (core,
re-exported from the acorn subpath). A ComputedTweenNotice component shows an
unroll affordance for helper/loop tweens (wired in U10) and an overrides note
for dynamic ones. Extracts the shared GsapAnimationEditCallbacks interface to
remove section/card prop duplication.
* feat(core): lint understands computed timelines (acorn parser)
U7: the GSAP lint rule now loads parseGsapScriptAcorn (which inlines helpers
and bounded loops) instead of the recast parser, so overlapping_gsap_tweens and
related findings reflect true resolved positions for computed timelines — and
keeps recast out of the lint graph entirely. Literal compositions are
unchanged (parity), all 182 lint tests pass.
* docs: document the computed-timeline keyframe editing model
U8: keyframes.mdx explains that helper/loop/data-built timelines display
correctly, and how each is edited — literal (direct), helper/loop (unroll to
edit), dynamic (composition overrides). Nothing is permanently locked.
* feat: unroll computed timelines into literal tweens (U10)
Adds unrollComputedTimeline (core): serializes a parsed timeline's resolved
animations back to literal tl.* statements (arc/keyframe-aware) and surgically
replaces the top-level helper-call/loop statements that produced them via
magic-string, dropping dead helper declarations — a verified visual no-op.
Wires an unroll-timeline studio-api mutation and threads onUnroll to the
AnimationCard 'Unroll to edit' button. Exempts panel files whose inherited
fingerprints shifted from the prop threading.
* feat(runtime): declarative keyframe override layer for dynamic tweens (U11)
Adds applyKeyframeOverrides: fetches a gsap-overrides.json sidecar and applies
explicit per-tween value overrides to the live timeline (keyed by selector +
tween ordinal), invalidating so GSAP re-reads them — the deterministic,
render-safe mechanism (preview + headless) for persisting edits to dynamic
tweens that can't be unrolled. Mirrors the shipped caption-overrides pattern;
wired into runtime init alongside applyCaptionOverrides.
* refactor: drop the keyframe override layer; rely on unroll + source
Removes the gsap-overrides.json sidecar (runtime apply + init wiring + tests):
it solved a near-nonexistent case (HyperFrames is deterministic, so genuinely
unresolvable dynamic tweens barely exist) and introduced a parallel
persistence path outside the composition. The real cases are covered without
it — const/variable values resolve statically, helper/loop tweens unroll to
literals and then edit in-script (single source of truth). Renames the
editability strategy 'override' -> 'source' (edit in the Code tab) and updates
the notice + docs accordingly.
* fix(studio): drag outside tween range creates new keyframe, picks nearest tween
Fixes the GSAP drag intercept to pick the position tween closest to the
playhead (not the one with the most keyframes), and when dragging outside all
tweens' ranges, creates a brand-new keyframed tween instead of destructively
extending/replacing the nearest one. Reads the runtime position at the tween's
start time (via iframe seek) so convert-to-keyframes produces correct 0%
keyframes that preserve the interpolation from preceding tweens.
* fix(studio): drag outside tween range creates new keyframe, picks nearest tween
Also reverts all fallow health.ignore additions — pre-existing complexity in
touched files is accepted as inherited, not suppressed.
Moving sharp and onnxruntime-node to optionalDependencies (in the earlier
capture/native-module hardening) regressed `remove-background` from ~7% to
~97% failure starting at 0.6.101: the command genuinely *requires* both native
modules, but as optional deps they're skipped on most installs, so it hits the
guarded "module not available" error and fails for nearly everyone.
The capture crash that motivated the optional move is already fixed by the
lazy, guarded `await import()` in contentExtractor / inference — that holds
regardless of dependency classification. Making the modules optional was the
over-correction; the lazy import alone was sufficient. sharp ships its own
platform binaries as optional sub-deps, so it installs cleanly as a hard dep
without failing installs on unsupported platforms (it was a hard dep at 0.6.99
with remove-background at a healthy ~7%).
- Move sharp + onnxruntime-node back to `dependencies` (so they install for
everyone again). `@google/genai` stays optional — genuinely optional, lazy,
and not part of the regression.
- Keep the lazy guarded imports — they remain the crash-safety for capture.
- Add trackCommandFailure to remove-background's catch: it self-exits, so the
dispatch wrapper never saw it (the reason stream was blind). Now its failures
carry a reason, closing that command from the wrapper-blind follow-up.
remove-background tests + background-removal suite pass; tsc clean; build green.
* fix(engine): use ANGLE-EGL for Linux GPU path, bump NVENC probe size
Chrome 131+ rejects --use-gl=egl in headless shell; the GPU process
exits and the renderer silently falls back to SwiftShader. Switch to
(gl=angle, angle=gl-egl) which is on the headless-shell allowlist,
and add --ignore-gpu-blocklist + --disable-software-rasterizer so
data-center GPUs (L4/T4/A10) are not blocked.
Also bump the NVENC probe frame from 16×16 to 320×240 — NVIDIA
data-center cards require ≥257 on each dimension and reject the
smaller size with "Frame Dimension less than the minimum supported
value", causing the encoder probe to silently fall back to libx264.
Closes#1493
* fix(engine): address review feedback — probe test, observability, comments
- Export getProbeArgs and add test pinning 320×240 probe dimensions
across all 5 GPU encoder backends (nvenc/videotoolbox/vaapi/qsv/amf)
- Add driver/SKU rationale comment on the probe size constant with
context about NVIDIA data-center card behavior vs documented minimums
- Add rationale comment on --ignore-gpu-blocklist (operator opted into
hardware mode explicitly)
- Log resolved GL flags at browser launch for GPU fallback observability
* fix(studio): open SDK shadow session in master view (was never opening)
useSdkSession(projectId, activeCompPath) received activeCompPath=null in the
master/entry view — the studio's convention where null means index.html
(isMasterView = !activeCompPath || activeCompPath === "index.html"). The hook's
guard `if (!projectId || !activeCompPath) return` then bailed, so the SDK
session never opened in the default editing surface. Result: sdkSession was
null there → every shadow tap (onDomEditPersisted, onElementDeleted, timing,
gsap) was undefined/no-op → zero sdk_shadow_dispatch telemetry for master-view
edits (the common case). Shadow only fired when a sub-comp was explicitly
opened (which sets activeCompPath).
Resolve null → "index.html" (matching the existing convention used by
isMasterView and blockInstaller) so the session opens in master view.
Verified live: instrumenting the hook showed phase "skipped_no_ids"
(activeCompPath null) before, "opened" after.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(studio): shadow property parity — read camelCase style key, not kebab
The inline-style parity resolver read flat.styles[op.property] with the
kebab-case PatchOperation key ("background-color"), but ElementSnapshot
inlineStyles are camelCase ("backgroundColor"), so the read-back was always
null → a false value_mismatch on every hyphenated CSS property. Single-word
props (color, opacity) coincide, so unit tests missed it.
Found live: a color edit on a box emitted op:property mismatchCount:1 with
{property:"background-color", expected:"rgb(255,79,88)", actual:null}. Convert
kebab→camel for the read-back (fall back to the raw key).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(studio): emit op:delete shadow for timeline-clip deletes
The Delete/Backspace hotkey routes to handleTimelineElementDelete whenever a
timeline element is selected (useAppHotkeys: `if (selectedElementId) {
handleTimelineElementDelete(el); return; }`), returning before the
shadow-wired handleDomEditElementDelete. Every clip is a timeline element, so
clip deletes — the common case — emitted no op:delete; the delete shadow only
fired for a non-timed DOM selection.
Add runShadowDelete(sdkSession, element.hfId) to handleTimelineElementDelete's
success path, mirroring the move/resize timing taps.
Verified live (browser-use): deleting a clip now emits
sdk_shadow_dispatch op:delete dispatched:true mismatchCount:0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(studio): match GSAP fidelity tweens by resolved element, not raw selector
gsap_fidelity keyed tweens by id (targetSelector-method-position). On tween
ADD, the SDK writer emits [data-hf-id="X"] selectors while the server emits
class selectors (.x) for the same element — different ids → false
present/absent mismatch (mc:2) on every add. Update/remove were clean (the
tween already existed with one consistent selector).
Key by resolved element (selector → data-hf-id via the pre-op DOM) + method +
position, so equivalent tweens match and only real value drift registers.
Falls back to raw selector when resolution isn't possible.
Found live (browser-use): adding a tween emitted gsap_fidelity mc:2 with
{[data-hf-id="hf-b"]-to-0 present-only} + {.b-to-0 present-only}.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(studio): don't shadow studio-internal data-hf-* marker attributes
Property-path parity false-mismatched on canvas-drag (path-offset) edits, which
emit attribute ops like {property:"data-hf-studio-path-offset"}:
1. The name was built as `data-${op.property}` → double-prefix
"data-data-hf-studio-path-offset".
2. The SDK model excludes all data-hf-* attributes, so even the right name
reads back null → false value_mismatch.
attrName() prefixes only when needed; isShadowableOp() drops data-hf-* attribute
ops (studio-internal markers the SDK can't represent), filtered in
sdkShadowDispatch before dispatch + parity.
Code-confirmed via handleDomPathOffsetCommit → commitPositionPatchToHtml →
persistDomEditOperations → onDomEditPersisted; live repro blocked because the
test comp's elements were GSAP-animated (drags route to the GSAP path).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(studio): document tweenKey + selector-resolver ceilings (PR review)
Per review (Rames, non-blocking): name the two silent fail-modes in the GSAP
fidelity diff rather than build speculative disambiguators (not observed in
studio-emitted templates).
- tweenKey: coincident tweens (same element+method+position) collapse, last
wins. Props can't join the key — a matched pair must share a key for the
field-diff to run. Upgrade path: property-name hash.
- makeSelectorResolver: first-match heuristic; ambiguous shared-class selectors
may misunify. Upgrade path: querySelectorAll + uniqueness.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Studio-triggered renders emit render_complete / render_error from the CLI
preview-server process, which stamps every event with the install's
anonymousId (client.ts drainQueueToPayload). The browser, meanwhile, fires
studio_session_start / studio_render_start under its own getAnonymousId(). So
the render outcome and the render start never share a person_id — verified in
data: of 15,125 users who started a studio render in 30d, ZERO have any
render_complete under any source, and the 898 studio-tagged completers are
disjoint server UUIDs. The studio render funnel — the product's core value
moment and strongest retention signal — is therefore unmeasurable.
Thread the browser's telemetry id through to the render-outcome events:
- client.ts: trackEvent takes an optional distinctId; drainQueueToPayload uses
`event.distinctId ?? config.anonymousId`. CLI renders unchanged.
- events.ts: trackRenderComplete/trackRenderError forward an optional distinctId.
- studioRenderTelemetry.ts: emitStudioRender* pass opts.distinctId through.
- core studio-api (types.ts + routes/render.ts): the render route reads
`telemetryDistinctId` from the request body (validated string) and passes it
to the adapter's startRender, which already forwards opts to the emitters.
- studio (useRenderQueue.ts): include getAnonymousId() as telemetryDistinctId
in the render POST — the same id studio_* events already use.
Result: studio render_complete/error now carry the browser user's id and join
studio_session_start / studio_render_start. Older clients that don't send the
field fall back to anonymousId (no regression). No new tracking surface — it's
the existing anonymous studio id.
Tests: per-event override forwarding (events), studio render distinctId
threading + older-client fallback (studioRenderTelemetry), and route body →
adapter forwarding incl. non-string rejection (core render route).