@prestyj/cli 5.28.0 → 5.28.1

package/assets/motion/skills/mixkit-split-text-617/references/VERIFICATION.md

@@ -1,107 +0,0 @@
-# Evidence, uncertainty and acceptance
-
-## What is verified already
-
-The user's AEP was parsed using `forticheprod/py-aep` at commit
-`e12a451c35bacd3f34a080090265f9370e66162b`. All 194 fetched parser source files
-were checked against that commit's Git blob hashes. The exporter reported
-zero field/serialization errors. Repeating extraction returned the same
-content except elapsed time. No expressions were evaluated.
-
-The skill's JSON composition files preserve the full exported records,
-including numeric values and flags marking parser-synthesized defaults.
-They are not a handwritten approximation or a newly designed timeline.
-`data/manifest.json` records their SHA-256 hashes and source AEP fingerprint.
-
-Source counts:
-- 22 compositions, 99 layers.
-- 1,875 leaf properties backed by non-synthetic binary chunks.
-- 17,795 parser-synthesized leaf properties. Do not call those authored values.
-- Exported animation-key counts can duplicate separated Position data; do not
-  describe every exported key entry as a distinct authored keyframe.
-
-### Opening-reveal spot-check
-
-The first scene's matte-height Bezier was independently evaluated from its
-exported times, values, speeds and influences, without the parser's interpolator.
-At ten timestamps the height was compared with a brightness-based edge
-measurement of the real 640×360 Mixkit preview. Maximum absolute difference:
-**1.728 pixels**. Actual preview frame timestamps were checked with ffprobe.
-All samples are preserved in the manifest.
-
-This is a **limited numeric spot-check**, not a full render comparison.
-It is affected by compression and thresholded edge measurement. It does not
-validate text animators, all scenes, all effects or the complete choreography.
-
-## What is not verified
-
-- No whole-video reconstruction has been generated or compared.
-- No export/render inside After Effects has independently checked the full
-  parser output. The parser is a reverse-engineered implementation, not AE.
-- Per-letter selector weighting, randomized order, font layout and combination
-  of text animators have not been reproduced and checked in another renderer.
-- Fast Blur, stroke/outline rendering and other compositing semantics may differ
-  from apparently similar browser/Canvas/SVG effects.
-- Original footage is absent and intentionally replaceable. Font substitution
-  is allowed. Neither asset substitution validates the surrounding motion.
-- No reference audio was supplied. No sound timing was extracted.
-
-A future agent must not turn "all data files passed integrity checks" into
-"the video is visually verified."
-
-## Acceptance procedure after a reconstruction exists
-
-1. **Validate the data package.**
-   Run `python3 "<skill-root>/tools/inspect_motion.py" verify`. This only validates
-   bundled hashes, counts and links; it does not render anything.
-2. **Record the implementation mapping.**
-   For every used source composition/layer, record the target object and where
-   its keyed properties, selectors, effects, matte, stack and clock are handled.
-   Mark every unsupported or approximate semantic explicitly. Do not invent a
-   coverage percentage from raw property counts (most exposed defaults are unused).
-3. **Check actual technical output.**
-   Confirm 1920×1080, 30 fps, 361 frames, complete layer/media/font loading,
-   deterministic seeking, no render errors and no substituted silent fallbacks.
-   In EZ Motion, retain the existing HyperFrames validation and completion review.
-4. **Check source timing, not only screenshots.**
-   Sample exact keys and intermediate times, including steep Bezier sections,
-   HOLD boundaries, negative-start echoes, overlapping scenes and stretched
-   decorative instances. The source video's ~59.94 fps and the project's 30 fps
-   are different: compare by seconds, not matching frame indices.
-5. **Compare the entire sequence in motion.**
-   Cover at least 0–2.2 s, 2.1–4.8 s, 4.7–7.2 s, 7.0–9.7 s and 9.4 s through
-   the end. These overlapping review windows are for inspection, not new cuts.
-   Compare reveal dimensions, zoom timing, directions, silhouettes, overlaps,
-   outline echo spacing and final settling/hold. A contact sheet alone is not
-   playback evidence. Report if the available tooling only inspected frames.
-6. **Separate content differences from motion differences.**
-   A different person/image or changed glyph silhouette is expected. A different
-   reveal axis, easing shape, stagger rule, outline-copy count, scene overlap or
-   duration is a reconstruction defect unless the user explicitly requested it.
-   A raw pixel-diff score with different content is not a useful fidelity verdict.
-7. **Resolve or disclose failures.**
-   Fix concrete mismatches from the source data and recheck the affected interval.
-   When semantics cannot be reproduced/validated, deliver an honestly labelled
-   draft with the exact unresolved layer/interval. Do not replace it with a
-   stylistically similar effect and claim fidelity.
-
-Reference page:
-https://mixkit.co/free-after-effects-templates/split-text-intro-617/
-
-Known public preview URL (availability may change):
-https://assets.mixkit.co/video-templates/617/mixkit-617-360.mp4
-
-No reference video is bundled. Use the user's supplied/cached reference or an
-otherwise authorized source through the host's normal media workflow. Access
-failure does not authorize bypassing a site's restrictions.
-
-## Reporting language
-
-- **Source-grounded:** implemented from the recorded values and mechanisms.
-- **Integrity checked:** data/hash/link checks passed, not visual verification.
-- **Partially verified:** identify the actual tested intervals/mechanisms.
-- **Visually compared:** state what was watched or which frames were inspected;
-  list remaining deviations. Do not imply human approval or pixel identity.
-
-This skill currently has source data and limited extraction evidence. It is
-not a certificate of rendering correctness and should not claim otherwise.

package/assets/motion/skills/mixkit-split-text-617/tools/inspect_motion.py

@@ -1,130 +0,0 @@
-"""Read the bundled animation evidence without loading it all into an agent's context.
-Python standard library only. No writes, network, dependency loading, or rendering.
-"""
-import argparse
-import hashlib
-import json
-from pathlib import Path
-import sys
-
-
-def load_manifest(root: Path) -> dict:
-    path = root / 'data' / 'manifest.json'
-    if path.stat().st_size > 65536:
-        raise ValueError('Manifest exceeds 64 KiB')
-    data = json.loads(path.read_text(encoding='utf-8'))
-    if data.get('schema_version') != 1 or not isinstance(data.get('compositions'), list):
-        raise ValueError('Unsupported manifest')
-    return data
-
-
-def load_comp(root: Path, manifest: dict, comp_id: int) -> dict:
-    entry = next((c for c in manifest['compositions'] if c['id'] == comp_id), None)
-    if entry is None:
-        raise ValueError(f'Unknown composition ID: {comp_id}')
-    path = (root / 'data' / 'compositions' / f'comp-{comp_id}.json').resolve()
-    if not path.is_relative_to(root) or path.stat().st_size > 8 * 1024 * 1024:
-        raise ValueError('Composition path or size is not allowed')
-    raw = path.read_bytes()
-    if hashlib.sha256(raw).hexdigest() != entry['sha256']:
-        raise ValueError(f'Composition {comp_id} failed its integrity check')
-    comp = json.loads(raw)
-    if comp['id'] != comp_id or len(comp['layers']) != entry['layer_count']:
-        raise ValueError('Composition identity/count mismatch')
-    return comp
-
-
-def walk(nodes: list, ancestors: tuple = (), depth: int = 0):
-    if depth > 32:
-        raise ValueError('Property nesting exceeds the inspection limit')
-    for ordinal, node in enumerate(nodes):
-        path = ancestors + (f"{ordinal}:{node['match_name']} ({node['name']})",)
-        yield path, node
-        yield from walk(node.get('children', []), path, depth + 1)
-
-
-def layer_summary(layer: dict) -> dict:
-    return {k: v for k, v in layer.items() if k != 'properties'}
-
-
-def verify(root: Path, manifest: dict) -> dict:
-    comps = [load_comp(root, manifest, c['id']) for c in manifest['compositions']]
-    ids = {c['id'] for c in comps}
-    if len(ids) != len(comps):
-        raise ValueError('Duplicate composition IDs')
-    totals = {'compositions': len(comps), 'layers': 0, 'stored_properties': 0, 'synthetic_properties': 0}
-    for comp in comps:
-        layer_ids = {l['id'] for l in comp['layers']}
-        if len(layer_ids) != len(comp['layers']):
-            raise ValueError('Duplicate layer IDs')
-        totals['layers'] += len(comp['layers'])
-        for layer in comp['layers']:
-            source = layer.get('source')
-            if source and source['type'] == 'CompItem' and source['id'] not in ids:
-                raise ValueError('Missing source composition')
-            for link in ('parent', 'track_matte_layer'):
-                ref = layer.get(link)
-                if ref and ref['id'] not in layer_ids:
-                    raise ValueError(f'Missing linked layer: {link}')
-            for _, prop in walk(layer['properties']):
-                if 'binary_chunk_synthetic' in prop:
-                    key = 'synthetic_properties' if prop['binary_chunk_synthetic'] else 'stored_properties'
-                    totals[key] += 1
-    for key, actual in totals.items():
-        if actual != manifest['expected_counts'][key]:
-            raise ValueError(f'Evidence count mismatch: {key}')
-    return {'integrity': 'PASS', 'counts': totals, 'full_video_reconstruction_verified': False}
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(description=__doc__)
-    sub = parser.add_subparsers(dest='command', required=True)
-    sub.add_parser('list', help='List compositions, dynamic slots and evidence scope')
-    sub.add_parser('verify', help='Check data integrity and links; does not render or validate motion')
-    for name in ('layers', 'keys'):
-        p = sub.add_parser(name)
-        p.add_argument('composition', type=int)
-    prop_parser = sub.add_parser('property')
-    prop_parser.add_argument('composition', type=int)
-    prop_parser.add_argument('layer', type=int)
-    prop_parser.add_argument('match_name')
-    prop_parser.add_argument('--include-defaults', action='store_true', help='Also show explicitly labelled parser-synthesized defaults')
-    args = parser.parse_args()
-    root = Path(__file__).resolve().parents[1]
-    try:
-        manifest = load_manifest(root)
-        if args.command == 'list':
-            result = manifest
-        elif args.command == 'verify':
-            result = verify(root, manifest)
-        else:
-            comp = load_comp(root, manifest, args.composition)
-            if args.command == 'layers':
-                result = {k: v for k, v in comp.items() if k != 'layers'}
-                result['layers'] = [layer_summary(l) for l in comp['layers']]
-            else:
-                matches = []
-                if args.command == 'property' and not any(l['id'] == args.layer for l in comp['layers']):
-                    raise ValueError(f'Unknown layer ID: {args.layer}')
-                for layer in comp['layers']:
-                    if args.command == 'property' and layer['id'] != args.layer:
-                        continue
-                    for path, prop in walk(layer['properties']):
-                        if args.command == 'property':
-                            selected = prop['match_name'] == args.match_name and (args.include_defaults or not prop.get('binary_chunk_synthetic', False))
-                        else:
-                            selected = bool(prop.get('keyframes')) and prop['match_name'] != 'ADBE Marker'
-                        if selected:
-                            matches.append({'layer_id': layer['id'], 'layer_name': layer['name'], 'indexed_path': list(path), 'property': prop})
-                if args.command == 'property' and not matches:
-                    raise ValueError('No stored property matched that exact name; use --include-defaults to inspect labelled parser defaults')
-                result = matches
-        print(json.dumps(result, indent=2, ensure_ascii=False, allow_nan=False))
-        return 0
-    except (OSError, ValueError, KeyError, TypeError) as exc:
-        print(f'Inspection failed: {exc}', file=sys.stderr)
-        return 1
-
-
-if __name__ == '__main__':
-    raise SystemExit(main())