norahe0304-art/30x-video / cli/skill
30x-web-to-video
FlaggedUse when a user gives a brand URL and wants a premium launch, demo, or marketing video. Zero-manual workflow — Claude runs the V2 orchestrator itself (scrapes brand evidence, auto-installs missing tools, generates the first-cut Remotion project). User only sees the URL and the Studio preview.
benchmarks/README.md
View on GitHub# URL-to-Video V2 Benchmarks
This directory defines the canonical offline benchmark suite for URL-to-video V2.
It is framework-agnostic and intentionally thin: the main orchestrator should read the manifest, run each URL, and write results in its own output format.
The concrete runner now lives at `../scripts/benchmark-suite.ts`.
Offline expectation:
- Use `--reuse-brand-root <dir>` when you maintain one evidence pack per benchmark.
- Use `--reuse-brand-dir <dir>` when you want to pressure-test one benchmark against one known evidence pack.
- If you run live instead of offline, the runner may fetch directly from the URL and skip local fixtures.
Example:
```bash
node --experimental-strip-types scripts/benchmark-suite.ts --match stripe-fintech --offline --reuse-brand-dir ../stripe-launch-video/public/brand --install --verify --render
```
## What This Suite Measures
Each benchmark run should produce the same six required outputs:
- `evidence_score`: overall score plus sub-scores and a short rationale.
- `chosen_mode`: one of `product-evidence`, `editorial`, or `blocked`.
- `archetype`: primary archetype, with an optional secondary archetype.
- `first_cut`: the initial Remotion-ready artifact or project path.
- `self_review`: evidence, taste, and edit gate notes.
- `human_evaluation`: reviewer judgment on whether the first cut is worth polishing.
## Required Output Shape
The orchestrator can store results however it wants, but each benchmark run should preserve these fields:
```json
{
"benchmark_id": "stripe-fintech",
"url": "https://stripe.com",
"evidence_score": {
"overall": 0,
"subscores": {}
},
"chosen_mode": "product-evidence",
"archetype": {
"primary": "",
"secondary": null
},
"first_cut": {
"status": "not-run",
"artifact": ""
},
"self_review": {
"evidence_gate": "",
"taste_gate": "",
"edit_gate": ""
},
"human_evaluation": {
"score": 0,
"decision": "",
"notes": ""
}
}
```
## Benchmark Entry Fields
The manifest entries are designed to stay stable over time:
- `id`: stable slug for the benchmark.
- `brand`: display label.
- `category`: one of the target groups requested by the user.
- `url`: canonical source URL.
- `mode_hint`: suggested production line for evaluation only.
- `evidence_profile`: shorthand for expected evidence shape.
- `notes`: human-readable guidance for the benchmark.
## Run Expectations
For each URL, the orchestrator should:
1. Collect brand evidence from the URL.
2. Compute evidence strength and mode selection.
3. Infer an archetype.
4. Produce a first cut or a blocked report.
5. Run self-review.
6. Record a human evaluation slot for later scoring.
The benchmark suite does not define rendering code. It only defines the contract that the renderer and reviewer must satisfy.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| planning-and-task-breakdownaddyosmani/agent-skillsBreaks work into ordered tasks. Use when you have a spec or clear requirements and need to break work into implementable tasks. Use when a task feels too large to start, when you need to estimate scope, or when parallel work is possible. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| screenpipethedotmack/claude-memSearch screen and audio history recorded by Screenpipe and optionally save selected context into claude-mem. Use when the user asks to recall desktop activity, a meeting, or something seen in another app through Screenpipe. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |
| contact-widgetnexu-io/open-designSelf-contained floating chat widget with welcome screen, social links, meeting button, and message input. Single HTML file, zero dependencies. | Stars: 108k | Updated | Claude CodeCodex | Verified | Install |
| meeting-notesnexu-io/open-designMeeting notes page — title bar with attendees, agenda checklist, decisions block, action items table with owners + dates, and a "next meeting" footer. Use when the brief mentions "meeting notes", "minutes", "1:1 notes", "all-hands recap", or "会议纪要". | Stars: 108k | Updated | Claude CodeCodex | Verified | Install |