Skip to content

armgit / Orkas

Public Imported from GitHub
0

Orkas is an open-source (MIT), local-first AI team for your desktop: a Commander turns your goal into to-dos and coordinates 16 built-in specialist agents plus Claude Code, Codex, OpenCode, OpenClaw and Hermes, with projects, scheduled automations and connectors. Free app for macOS & Windows; built-in models or your own keys.

Preserve natural English casing in VideoStudio

wyt56fb841bd04c
diff --git a/resources/builtin/_manifest.json b/resources/builtin/_manifest.json
index f2e3d4c8..511a3f44 100644
--- a/resources/builtin/_manifest.json
+++ b/resources/builtin/_manifest.json
@@ -1,6 +1,6 @@
{
"schema": 1,
- "content_tree_sha256": "5f9d72aaf7c24f8b533b35872019035f42dc79ec500a7b30a3921e343f54f087",
+ "content_tree_sha256": "b0aa8dc347fb09ecd29a1c43f348aebcc7579b8d697f81796e89f739ec3a0d82",
"files": [
{
"path": "marketplace/agents/78900d8758bc/_meta.json",
@@ -14,8 +14,8 @@
},
{
"path": "marketplace/agents/79df9cc89f5f/agent.json",
- "bytes": 31743,
- "sha256": "799f6712df800fcaf401d4c45e9560405718662e2233a75a7259283bad0e9bb3"
+ "bytes": 32307,
+ "sha256": "2b5a98f6cf67d9fe16e7fb9228c312a217de4755d30a73c0580e04efbceb65c7"
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/_shared/scripts/src/bundled-runtime.ts",
@@ -64,8 +64,8 @@
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/composition-design-review/SKILL.md",
- "bytes": 6272,
- "sha256": "f49fb4498bec6e626e6dbb00c61ece2070e9412e1c498d0893c6c5dae2b9b6ee"
+ "bytes": 6638,
+ "sha256": "e01f3e58c5dd0e89fcfc4c7720a741dccf0dd896fa95b099c931722fd5182c33"
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/design-system-importer/SKILL.md",
@@ -79,8 +79,8 @@
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/frontend-design/references/visual-primitives.md",
- "bytes": 7573,
- "sha256": "77343403effb6fabfbeb7313ea1367536f91b3b64bbe3c6f5df21ac46490ef77"
+ "bytes": 7769,
+ "sha256": "de41211f415eccf5f71b73f0dc038ee8e365a294ea77dd2d8ad48d28a3252963"
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/frontend-design/references/worked-compositions.md",
@@ -89,8 +89,8 @@
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/frontend-design/SKILL.md",
- "bytes": 18320,
- "sha256": "6f1800d3f8e99bef91b65c48a065d62657e7c898683b021f89a6007cba19006f"
+ "bytes": 18960,
+ "sha256": "d45a918c2ffdfc742c65d2526c94f58027f3eb1aa2a3218ed8a3d44df9bb84a3"
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/stage-assemble/SKILL.md",
@@ -104,8 +104,8 @@
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/stage-compose/SKILL.md",
- "bytes": 46030,
- "sha256": "8b0c987d598455ae0537c538d39ba8590a8f101f05c0a29205c5611bda56e5ec"
+ "bytes": 46367,
+ "sha256": "165c208266e025e9afccbef2cc3fbff9081012b4ddc341475a13bc3ae6155d7c"
},
{
"path": "marketplace/agents/79df9cc89f5f/skills/stage-consistency/SKILL.md",
@@ -677,7 +677,7 @@
{
"id": "79df9cc89f5f",
"name": "VideoStudio",
- "version": "1.0.44",
+ "version": "1.0.45",
"skill_list": [
"composition-design-review",
"design-system-importer",
diff --git a/resources/builtin/marketplace/agents/79df9cc89f5f/agent.json b/resources/builtin/marketplace/agents/79df9cc89f5f/agent.json
index 314d6896..cf1efc9c 100644
--- a/resources/builtin/marketplace/agents/79df9cc89f5f/agent.json
+++ b/resources/builtin/marketplace/agents/79df9cc89f5f/agent.json
@@ -1,13 +1,13 @@
{
"agent_id": "79df9cc89f5f",
- "version": "1.0.44",
+ "version": "1.0.45",
"min_app_version": "1.6.0",
"name": "VideoStudio",
"description_zh": "做视频、也剪视频:解说/动画、AI 口播数字人、给已有视频加字幕/配音旁白、剪辑切片、高光、本地化、混音降噪——凡是“做一段视频”或“对已有视频做处理”的需求都路由到它,而不是 commander 自己拿命令行/ffmpeg 硬拼。三条产线:①解说/动画(脚本→分镜→HTML 合成渲染 mp4);②AI 生成实拍(口播/数字人);③剪辑你上传的真实视频——加旁白/配音、加/烧字幕、剪辑切片、做高光、本地化配音、混音降噪。适合“做个 60 秒讲解 X 的动画”“竖屏短视频介绍产品”“把文案做成带字幕解说”“给这段视频加中文旁白/配音”“给我的视频加字幕/剪一下/做高光/做本地化”。触发词:做视频、解说视频、动画、短视频、视频制作、配音、旁白、加字幕、剪辑、剪视频、给视频加…、给这段视频…、高光切片、本地化配音、竖屏视频",
"description_en": "Makes and edits videos. Routes video requests into three production lines: explainers/animations, AI-generated footage or talking-head/avatar clips, and edits to uploaded real footage. Use it for short videos, captioned explainers, vertical product clips, narration or voiceover, subtitles, trimming, highlights, localization, dubbing, mixing, denoise, or any request to change an existing video. Triggers: make video, explainer, animation, short video, edit video, voiceover, narration, add captions, trim, highlights, localize, dubbing, vertical video",
"workflow": "You are Video Studio, an in-app video production agent for short videos. Route each request into one locked production line, produce reviewable artifacts, and stop only at named user gates. Load only the line-specific skills and files needed for the current stage.\n\n### Operating model\n- Route first, then lock COMPOSE, GENERATE, EDIT, or AUTO. Do not silently switch lines after a gate.\n- User-facing gates are GATE A concept/brief, GATE B script/storyboard or EDL, optional GATE C billable/sensitive generation, the conditional COMPOSE HTML Preview Gate, and GATE D draft review.\n- Internal states do not ask the user: normalization, narration fit, canonical manifest preparation, scaffold generation, visual HTML authoring, fail-closed preflight, lint/inspect, render, media QA, and reports.\n- Treat built-in tools and skill scripts as stable interfaces. Do not inspect platform runtime/source code to reverse-engineer them.\n- Recover state from disk, not memory. After any mid-turn context checkpoint/compaction, re-read your durable state before acting — project/plan.json (AUTO/EDIT) or the canonical composition-manifest.json plus VideoProductionStateV1 and the latest draft/QA report (COMPOSE) — and never re-produce a segment, re-synthesize narration, or re-run a render those files already mark done.\n\n### Gate protocol\nAt every named gate:\n1. Show the artifact in chat using the current User UI language from system context: concise markdown, image/contact sheet, or [video](chat-media://local/<absolute path>).\n2. State the next action and any cost or QA note.\n3. Emit one <agent-input-form> with approval/revision choices and a free-text adjustments field, then <plan-interaction status=\"open\" /> and stop. Do not call another tool after opening a gate. Decision field ids are a stable host contract: Gate B uses `gate_b_decision`, Gate C uses `gate_c_decision`, HTML Preview uses `preview_decision`, and Gate D uses `gate_d_decision`; every decision select uses value `approve` for approval and `revise` for revision. Never rename these ids in confirmation, recovery, or resumed turns.\n4. Approval must be explicit. A new turn, question, or unrelated message is not approval. Revisions invalidate downstream approvals.\n5. Artifact expectations are fixed: Gate A shows concepts/brief, Gate B shows script plus shotlist/EDL, HTML Preview shows the contact sheet, Gate D shows the draft video, and final delivery shows the exported video. If the artifact cannot be shown, report that as a blocker instead of asking for approval.\n\n### Route and lock\nRead video-router before choosing:\n- COMPOSE: designed explainers, product promos, typography, diagrams, motion graphics, captions/lower-thirds, deterministic HTML/SVG.\n- GENERATE: AI-generated footage, talking heads, avatars, characters, or cinematic scenes. Billable generation requires Gate C.\n- EDIT: supplied footage to cut, subtitle, dub, localize, mix, denoise, or highlight.\n- AUTO: a true cross-line deliverable combining supplied, generated, composed, narration, or edit segments in one EDL.\n\n### GATE A - Proposal\nShow the inferred line, aspect ratio, duration, video language, and 1-3 concepts. Video language controls the deliverable copy/narration/captions; chat prose still follows the current User UI language from system context. Each concept includes hook, visual approach, rough length, and any billable cost note. Inspect supplied assets before proposing concepts and use them in at least one opening hook. Options: pick concept, revise brief/mix, new direction. Stop.\n\n### Shared craft\nApply video-craft to the locked line before authoring: a clear hook, one idea per beat, readable safe-zone text, restrained motion, muted-friendly captions, and audio fit.\n\n### Shared narration capability\nFor every off-screen narration path, call video_studio speech.capabilities before Gate B. Choose only a returned route_ref + voice_ref, copy its display_name and a natural speed into the canonical manifest or EDL, show that narrator at Gate B, and never invent or override a provider voice id during execution.\n\n### COMPOSE line\nAfter Gate A locks COMPOSE, read stage-compose and frontend-design; add other skills only when the approved concept needs them.\n1. Write project/script.md, project/shotlist.json, and the candidate project/composition/composition-manifest.json v2 together before Gate B. When narration is active, audio.narration_intent must contain the selected route_ref, voice_ref, display_name, and speed. The manifest is the single structural source of truth for canvas, scenes, target timing, video language, approved copy, source mapping, semantic roles, signed narration intent, audio ownership, and art direction; the user must approve the same payload that production signs.\n2. GATE B: show the script, compact shotlist, and selected narrator display name/speed. Options: approve, revise, change direction. Stop.\n3. In the same explicit Gate B approval turn, call video_studio composition.approve_plan, then composition.doctor, then composition.prepare. Prepare validates or one-time migrates legacy design-contract/scene-map input, creates the protected root/clip/paused-GSAP scaffold, and records scaffold_ready in VideoProductionStateV1. For standalone narration, immediately call composition.materialize_narration: it reads the signed manifest text, performs free duration preflight, synthesizes once, records text/audio hashes and measured duration, allocates scene windows within the immutable Gate B target duration, and rebuilds the untouched scaffold with declarative audio. A material measured-duration mismatch fails closed with recoverable audio instead of shortening the video or billing again. Never call generic generate_speech for standalone COMPOSE narration. Then model-author only visual HTML/CSS/SVG and deterministic timeline tweens inside that scaffold. Never use imperative audio/video playback, remote runtime resources, tl.call() scene switching, or spec.json.\n4. Run composition.inspect. When duration is at least 20 seconds, scene count is at least 3, visuals are dense/complex, or a prior draft failed, run composition.snapshot. Snapshot runs the same fail-closed preflight and semantic multi-frame QA. The system publishes the returned contact_sheet as the visible HTML Preview artifact; show it, open the HTML Preview Gate, and stop. Only after a later explicit user approval call composition.approve_preview; composition.draft remains blocked without that approval.\n5. Run composition.draft. The command reuses the same manifest/preflight, seeks the deterministic timeline, renders, probes media/audio, captures semantic evidence frames, checks expected scene/title visibility and frozen runs, and writes one normalized report. Structural errors stop before full render. Repair at most two localized manifest/HTML/content issues; do not loop on advisory craft findings.\n6. GATE D opens only when draft returns ok:true and draft_ready:true and any required signature-bound composition.submit_design_review verdict has passed. If draft says design_review_required, run the bounded composition-design-review against its contact sheet/frame evidence and submit the structured verdict before opening Gate D. The system publishes draft.mp4 as the visible Gate D artifact even while the approval form is open; show the video, QA headline/report, and any advisory design-review findings. Stop. After a later explicit approval call composition.approve_draft, then composition.export. The system publishes the exported mp4 as the final visible deliverable. Raw composition.render is unavailable; export reruns high-quality QA. Any changed composition input invalidates approval.\n\n### AUTO line\nRead stage-plan then stage-assemble. Probe/transcribe/OCR supplied media before planning. Every compose segment carries the complete child `spec.composition_plan`; every generate segment declares `spec.media_kind` and exact provider settings. Validate project/plan.json, then Gate B on the EDL. In the explicit approval turn call video_studio production.approve_plan. Gate C appears only for billable generation, uses `gate_c_decision`, and its approval turn calls production.approve_generation before any generate_image/generate_video call. Child compositions write local script.md/shotlist.json/manifest files and call composition.approve_plan with the parent plan_path and segment_id to inherit Gate B; never open another child Gate B. Assemble idempotently, never regenerate completed billable segments, and never retry a pending/failed billable transaction under the same Gate C. production.status reports local state only and cannot query the provider; a user-requested retry needs a fresh Gate C, and an uncertain pending attempt needs a new output path. Mix narration once and keep assembler-owned compose segments silent. Before Gate D run video_plan --op promise_check --probe-produced; any failure blocks delivery.\n\n### GENERATE line\nRead stage-plan and stage-generate; add stage-consistency for recurring characters or continuity. Represent every billable image/video or consistency portrait as a generate segment in project/plan.json. Gate B approval calls production.approve_plan; Gate C uses `gate_c_decision` and calls production.approve_generation. Every generate_image/generate_video call passes production_plan_path and production_segment_id. A pending/failed transaction is not retried under its current Gate C; production.status cannot query the provider, and a user-requested paid retry needs a fresh Gate C. Preserve built-in lip-synced audio and do not synthesize over a speaking mouth.\n\n### EDIT line\nRead stage-plan and stage-edit and inspect actual media: probe every clip, transcribe speech, and OCR silent/screen/slideshow media. Persist the edit EDL in project/plan.json; Gate B approval calls production.approve_plan before executing trims, concat, mix, subtitles, loudness, and localization. Gate D shows the draft and QA headline.\n\n### Follow-up edits\nWhen a draft/final exists and the user asks for one local change, edit only the affected segment, narration line, caption, trim, volume, speed, or shot, then reproduce/reassemble only what changed. Re-plan only for a real timeline restructure.\n\n### Deliver\nEvery final response includes the produced media link or a clear blocker. Keep gate and delivery messages short; the artifact is the message.",
"created_at": "2026-06-23T19:35:00",
- "updated_at": "2026-07-16T09:08:45.000Z",
+ "updated_at": "2026-07-16T16:01:24.000Z",
"knowhow": [
"Routes video requests into COMPOSE, GENERATE, EDIT, or AUTO production lines and loads only the skills needed by the locked line.",
"Separates user-visible gates from internal production states so technical work proceeds without extra approval loops.",
@@ -17,6 +17,7 @@
"Blocks COMPOSE preview/draft on structural QA and semantic visual defects such as unreadable text, overflow, occlusion, unsafe placement, low contrast, or primary content outside canvas; palette and decorative-variety findings remain advisory.",
"Uses SVG as the preferred visual layer for diagrams, lines, nodes, charts, progress, and background geometry while GSAP orchestrates deterministic motion.",
"Applies a frontend-design aesthetic thesis before composition so subject matter drives typography, palette, layout signature, and motion choices.",
+ "Preserves natural English casing for readable COMPOSE copy and treats all caps as a bounded accent for one short metadata label, acronym, or code instead of a default typography style.",
"Uses frontend-design generation references and adapted CSS/SVG visual primitives during initial HTML authorship, with a private pre-code art-direction pass that adds no user gate.",
"Applies HyperFrames-style composition discipline in COMPOSE: visual identity before HTML, resolved hero frames before motion, and GSAP entrances/reveals into those stable layouts.",
"Front-loads COMPOSE visual direction with design tradition, lazy-default rejection, video scale, depth layers, typography register, motion verbs, and rhythm before HTML authoring.",
@@ -59,6 +60,7 @@
"COMPOSE initial HTML generation must apply the frontend-design pre-code art-direction pass and generation references internally; it must not add a separate blueprint approval or user gate.",
"COMPOSE visual authoring must confirm manifest art_direction and VisualDirectionV1 first, build each scene's readable resolved/hero frame as static HTML/CSS/SVG, then add GSAP entrances, reveals, and transitions into that layout.",
"COMPOSE VisualDirectionV1 must reject generic lazy defaults and define a real design tradition, video-scale floor, three-layer scene depth, typography register, motion verbs, and rhythm pattern before model-authored HTML.",
+ "COMPOSE typography must preserve approved English casing by default: titles use sentence or natural title case, body/captions/subtitles/CTAs use sentence case, and all caps is limited to one short metadata label, acronym, or code when explicitly justified by the user, brand source, or art direction. Never apply text-transform: uppercase through a broad selector.",
"COMPOSE preview and draft preflight treat missing preview-required art direction as blocking: aesthetic thesis, VisualDirectionV1, motion budget, scene variation budget, per-scene depth layers, and per-scene motion verbs must be complete before HTML preview or mp4 rendering.",
"COMPOSE art direction must include style_source when an external DESIGN.md, brand guide, screenshot, reference site, Figma note, existing app UI, or named style influenced the design.",
"COMPOSE Gate D may open only after draft QA has zero blocking errors and any required composition.submit_design_review verdict is passed for the current draft signature.",
diff --git a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/composition-design-review/SKILL.md b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/composition-design-review/SKILL.md
index 01f397ce..c8eb1d38 100644
--- a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/composition-design-review/SKILL.md
+++ b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/composition-design-review/SKILL.md
@@ -58,6 +58,7 @@ Fix:
- Repeated layout, transition, or card pattern three or more times in a row.
- Palette uses extra chromatic colors beyond the contract.
- Type hierarchy is flat or labels feel like UI residue instead of video graphics.
+- English titles, body copy, captions, subtitles, or CTAs are forced to all caps, or two or more English text roles in one scene use all caps without an explicit user, brand, or art-direction reason. Restore the approved natural casing and use scale, weight, width, color, or spacing for hierarchy; keep all caps only for one short metadata label, acronym, or code.
- Scene density is too high for phone viewing.
- Style-source adaptation is vague: it borrows mood words but no concrete tokens.
diff --git a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/SKILL.md b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/SKILL.md
index 412cb3a0..8c466ebb 100644
--- a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/SKILL.md
+++ b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/SKILL.md
@@ -175,7 +175,8 @@ When writing `index.html`:
- Derive the main CSS variables from `manifest.art_direction`; keep extra chromatic colors intentional and named enough to audit. Do not flatten the design or recolor the whole video only to reduce a static palette count.
- Let one element carry personality: a custom progress line, typographic reveal, diagram grammar, texture, data mark, or transition family. Keep surrounding elements quiet.
- Use type as design material: contrast display/body roles, make title/body hierarchy unmistakable, and keep labels large enough for the video-craft floor.
-- Do not default every composition to Arial/Helvetica with only size changes. Use available local or deterministic system font stacks deliberately, create contrast through family/width/weight/case/tracking, and vendor any required font inside `assets/`; never fetch a remote font at render time.
+- Preserve the authored casing of approved English copy by default. Use sentence case or natural title case for titles and sentence case for body copy, captions, subtitles, and CTAs. All caps is an accent only when the user, brand source, or art direction calls for it, and then only for one short metadata label, eyebrow, acronym, or code. Never apply `text-transform: uppercase` through a broad selector or use all caps for multi-line copy; if two or more English text roles in one scene read as all caps, treat that as overuse and revise the hierarchy with family, width, weight, scale, color, or spacing instead.
+- Do not default every composition to Arial/Helvetica with only size changes. Use available local or deterministic system font stacks deliberately, create contrast through family/width/weight/tracking and restrained case changes, and vendor any required font inside `assets/`; never fetch a remote font at render time.
- Use structural marks only when they encode meaning: steps for sequences, coordinates for maps, ticks for time, nodes for relationships, brackets for comparison.
- Prefer SVG for diagrams, paths, meters, masks, charts, and signature geometry; keep prose and captions as real HTML text.
- Default to SVG/static held states first. Use GSAP only when a timed reveal, transition, or emphasis genuinely improves comprehension; animate SVG groups or a few containers instead of many HTML cards.
diff --git a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/references/visual-primitives.md b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/references/visual-primitives.md
index 3f19b2c1..c688087b 100644
--- a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/references/visual-primitives.md
+++ b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/frontend-design/references/visual-primitives.md
@@ -49,12 +49,12 @@ body {
font-size: 38px;
line-height: 1;
font-weight: 650;
- letter-spacing: .12em;
- text-transform: uppercase;
+ letter-spacing: .06em;
}
```
Derive tokens from the design contract. Do not paste this palette into unrelated work.
+Preserve the authored case by default. All caps is an optional accent for one short metadata label, acronym, or code—not the default treatment for labels and never a broad selector for titles, body copy, captions, or CTAs.
## 2. Topic-derived background field
diff --git a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/stage-compose/SKILL.md b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/stage-compose/SKILL.md
index bf214618..2370d0b9 100644
--- a/resources/builtin/marketplace/agents/79df9cc89f5f/skills/stage-compose/SKILL.md
+++ b/resources/builtin/marketplace/agents/79df9cc89f5f/skills/stage-compose/SKILL.md
@@ -209,7 +209,7 @@ The contract must declare these budgets compactly:
- `style_source`: from `design-system-importer` when a DESIGN.md, brand guide, screenshot, reference site, Figma notes, existing app UI, or explicit named style was used. Omit when there is no external style source.
- `scenes`: visual focus and layout type only. Structural start/duration, approved copy, source shots, and narration refs stay exclusively in the manifest. Designed scenes may include `scene_world`, `hero_visual`, `composition`, `depth_layers`, `motion_verbs`, `opening_state`, `resolved_state`, `continuity_in`, `continuity_out`, and selected `primitive_refs`.
- `layout_boxes`: safe text box, visual box, caption box, and maximum label count per scene.
-- `typography_tokens`: title/body/caption/label floors plus type roles and register. Default floors for 1920x1080: title >=72px, body/supporting text 28-42px, label/caption 18-26px, safe margin >=96px, no more than two text blocks and about 12-16 English words per scene. Preserve the same readability intent for 9:16 and 1:1. Avoid default two-sans pairings unless the style source explicitly requires them; use scale, weight, width, case, mono/data roles, or serif/sans contrast to make hierarchy visible.
+- `typography_tokens`: title/body/caption/label floors plus type roles and register. Default floors for 1920x1080: title >=72px, body/supporting text 28-42px, label/caption 18-26px, safe margin >=96px, no more than two text blocks and about 12-16 English words per scene. Preserve the same readability intent for 9:16 and 1:1. Preserve approved English casing by default: sentence/natural title case for titles and sentence case for body, captions, subtitles, and CTAs. Reserve all caps for one short metadata label, acronym, or code when the user, brand source, or art direction requires it; never use a broad `text-transform: uppercase` rule. Avoid default two-sans pairings unless the style source explicitly requires them; use scale, weight, width, restrained case changes, mono/data roles, or serif/sans contrast to make hierarchy visible.
- `color_tokens`: named baseline values with rationale: background, surface, text, muted, primary accent, and any purposeful supporting accents the approved visual idea needs.
- `motion_budget`: max animated groups per scene, allowed transitions, easing, rhythm pattern, which SVG/HTML groups move, what each motion communicates, and the concrete motion verbs assigned to primary elements.
- `scene_variation`: how the sequence avoids three near-identical layouts, transitions, or card/title scenes in a row.
diff --git a/test/main/util/video-agent-resources.test.ts b/test/main/util/video-agent-resources.test.ts
new file mode 100644
index 00000000..737c2e0d
--- /dev/null
+++ b/test/main/util/video-agent-resources.test.ts
@@ -0,0 +1,34 @@
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+
+import { describe, expect, it } from 'vitest';
+
+const VIDEO_STUDIO_ROOT = path.join(
+ process.cwd(),
+ 'resources',
+ 'builtin',
+ 'marketplace',
+ 'agents',
+ '79df9cc89f5f',
+);
+
+const read = (...segments: string[]) => fs.readFileSync(path.join(VIDEO_STUDIO_ROOT, ...segments), 'utf8');
+
+describe('open-source VideoStudio resources', () => {
+ it('preserves natural English casing and bounds all-caps accents', () => {
+ const agent = JSON.parse(read('agent.json')) as { standards?: string[] };
+ const frontendDesign = read('skills', 'frontend-design', 'SKILL.md');
+ const stageCompose = read('skills', 'stage-compose', 'SKILL.md');
+ const compositionDesignReview = read('skills', 'composition-design-review', 'SKILL.md');
+ const visualPrimitives = read('skills', 'frontend-design', 'references', 'visual-primitives.md');
+ const standards = (agent.standards ?? []).join('\n');
+
+ expect(visualPrimitives).not.toMatch(/text-transform\s*:\s*uppercase/i);
+ expect(visualPrimitives).toMatch(/Preserve the authored case by default/i);
+ expect(frontendDesign).toMatch(/Preserve the authored casing of approved English copy by default/i);
+ expect(frontendDesign).toMatch(/two or more English text roles.*all caps/i);
+ expect(stageCompose).toMatch(/Preserve approved English casing by default/i);
+ expect(compositionDesignReview).toMatch(/two or more English text roles.*all caps/i);
+ expect(standards).toMatch(/Preserve approved English casing/i);
+ });
+});