cawcut app list --schema              # grouped by category + source
cawcut app list --schema --category image
cawcut app list --schema --category video
cawcut app list --schema --json       # includes categories[] per app

Ready to run Photo to Sticker. Choose how to fill inputs:
1. Use defaults only (if any exist) and run — list which fields use defaults
2. Scene photo — pick one:
   A. Use a photo you already shared in this chat (I'll use that file/URL)
   B. Paste an HTTPS image URL
   C. Give a local file path (e.g. ~/Pictures/photo.jpg)
3. Text fields (only when required and no default) — pick a starter or write custom:
   A. Use suggested prompt: "..."
   B. Custom (you describe; one short message)
4. Run now with choices above

cawcut app list --schema

Output: Apps grouped by **output category** (Image Apps / Video Apps), then by **source** (Official / My / Shared), with **name**, description, credits, and input schema lines when available (`cawcut app list` does **not** print app IDs in human output). Identify likely matches by app name, description, `categories`, source, credits, and required input kinds. Use `cawcut app list --json` internally when you need the app ID for automation.

Example human output shape:

Do not treat Official Apps, My Apps, and Shared Apps as interchangeable. When the same app is visible from multiple sources, prefer the source displayed by the CLI.
Translate app names/descriptions when `reply_language` is not English; keep input keys and source group names exact. **Never show app IDs in user-facing App lists.**

If the user provided a natural-language request, rank likely matches:

1. Best official app/template match, when present.
2. Best owned/shared app match.
3. `cawcut-generate` fallback, only when no app matches or the user chooses raw generation.

**Single clear match:** treat as selected — run `describe`, then present metadata + inputs tables + input menu (UX rule 3). Do not reply with only "this app can…".

**Multiple matches:** show a numbered list with name, description, source, credits, and input summary for each; after the user picks, run `describe` and disclose full details before collecting inputs.

If the user did not provide a request, still present useful choices immediately. Do not ask an empty "what would you like to run?" question.

Present choices in `reply_language`. Default example (bare skill invoke / English):

If the user later writes in another language, use the same structure in `reply_language`.

If the user chooses an App (or a single App was auto-matched), immediately inspect it:

```bash
cawcut app describe "Product video generator"

Found a matching App: **First/Last Frame Extractor (qa test)**

| Field | Value |
|-------|-------|
| Name | First/Last Frame Extractor (qa test) |
| Source | Official Apps |
| Cost | … credits |
| Output | First frame + last frame images |

Extracts the first and last frame from a video and outputs them as images.

| Name | Type | Required | Notes |
|------|------|----------|-------|
| … | video | yes | The video to process (MP4, WebM, etc.) |

Provide the video:
A. Use a video already shared in this chat
B. Paste an HTTPS video URL
C. Provide a local file path

cawcut app describe "<app name>"

App: Product video generator
Description: Generate a product ad video

Inputs — use the KEY column with --input KEY=value:

  KEY                   KIND    REQ   DEFAULT           DESCRIPTION
  prompt                text    yes   no                Ad script or product description
    requirements: Required text input.
  product_image         image   yes   no                Product photo
    requirements: Required image input. Provide an HTTPS URL, local file path, or {asset_id,url} object.

Run: cawcut app run "Product video generator" --input 'prompt=...' --input 'product_image=@/path/to/file'

cawcut app describe "<app name>" --json

cawcut app run "3D Model Generator" --pick 2 --input ... --wait --download --json

Scene photo for Photo to Sticker:
A. Use a photo already shared in this chat
B. Paste an HTTPS image URL
C. Provide a local file path

cawcut app run "Photo to Sticker" \
  --input "Input - Scene Photo=@/path/to/photo.jpg" \
  --wait --download --json

cawcut app run "Product video generator" \
  --input prompt="summer sale, bright colors" \
  --input product_image="https://cdn.example.com/shoe.jpg" \
  --wait --download --json

cawcut app run "Product video generator" \
  --input prompt="summer sale" \
  --input product_image="@./shoe.jpg" \
  --wait --download --json

cawcut app run "<app name>" \
  --input key1=value1 \
  --input key2=value2 \
  --wait --download --json

cawcut app run "<app name>" --input-json ./inputs.json --wait --download --json

{ "prompt": "...", "product_image": "https://cdn.example.com/img.jpg" }

{
  "result_url": "https://cdn.cawcut.com/...?Expires=...&Signature=...&Key-Pair-Id=...",
  "result_urls": [
    "https://cdn.cawcut.com/...?Expires=...&Signature=...&Key-Pair-Id=...",
    "https://cdn.cawcut.com/...?Expires=...&Signature=...&Key-Pair-Id=..."
  ],
  "task_id": "...",
  "local_path": "/Users/.../file.png",
  "local_paths": ["/Users/.../file-1.png", "/Users/.../file-2.png"],
  "status": "done",
  "credits_used": 12.0,
  "credits_balance": 88.0
}