@yandy0725/mediagen-cli 0.0.0-stage → 0.1.0

package/package.json

@@ -1,6 +1,64 @@
 {
   "name": "@yandy0725/mediagen-cli",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
-}
+  "version": "0.1.0",
+  "description": "Non-interactive image generation and editing, plus short video generation, for agents: qwen-image, doubao/seedream, doubao/seedance, MiniMax H3, files on disk, JSON output.",
+  "type": "module",
+  "license": "MIT",
+  "bin": {
+    "mediagen": "./dist/cli.js"
+  },
+  "files": [
+    "dist",
+    "skills"
+  ],
+  "keywords": [
+    "cli",
+    "image",
+    "image-generation",
+    "image-editing",
+    "video",
+    "video-generation",
+    "text-to-video",
+    "seedance",
+    "minimax",
+    "text-to-image",
+    "qwen",
+    "dashscope",
+    "seedream",
+    "doubao",
+    "ark",
+    "agent"
+  ],
+  "engines": {
+    "node": ">=18"
+  },
+  "scripts": {
+    "build": "tsup src/cli.ts --format esm --target node18 --clean",
+    "test": "vitest run",
+    "dev": "tsx src/cli.ts",
+    "prepack": "npm run build"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/yandy/mediagen-cli.git"
+  },
+  "bugs": {
+    "url": "https://github.com/yandy/mediagen-cli/issues"
+  },
+  "homepage": "https://github.com/yandy/mediagen-cli#readme",
+  "publishConfig": {
+    "access": "public",
+    "registry": "https://registry.npmjs.org/"
+  },
+  "dependencies": {
+    "commander": "^12.1.0",
+    "smol-toml": "^1.3.1"
+  },
+  "devDependencies": {
+    "@types/node": "^22.10.2",
+    "tsup": "^8.3.5",
+    "tsx": "^4.19.2",
+    "typescript": "^5.7.2",
+    "vitest": "^2.1.8"
+  }
+}

package/skills/mediagen/SKILL.md

@@ -0,0 +1,88 @@
+---
+name: mediagen
+description: How to drive `mediagen`, a non-interactive CLI that generates and edits images and generates short videos, writing every result to a local file and printing its path. Covers what it can and cannot do, one-time setup, four commands worth copying, and the rules that keep a paid request from being wasted.
+---
+
+# mediagen
+
+Images and video from a command line: every result is written to a local file and the path
+is printed. Each successful call is a paid vendor request — and for video the money is
+spent when the task is created, minutes before the clip exists.
+
+Flags, models, sizes, exit codes and more examples live in the CLI itself:
+
+```bash
+mediagen --help
+mediagen image generate --help
+mediagen video generate --help
+```
+
+## What it does
+
+- `image generate "<prompt>"` — a new image from a description.
+- `image edit "<prompt>" --image ./photo.jpg` — a change to an existing image: restyle,
+  recolour, add or remove something, render text (CJK or Latin) into it, upscale (the same
+  `--image` with a larger `--size`), or combine several references (repeat `--image`, and
+  say which image is which in the prompt).
+- `video generate "<prompt>"` — a short clip from text, or from one first frame.
+- `video status <task-id>` — collect a clip whose run was interrupted. Reads only; it
+  cannot spend money.
+
+## What it cannot do
+
+Audio, documents, long-form video, editing or joining existing clips, describing or
+analysing an image (it writes images, it cannot see them), and local pixel work — crop,
+resize, convert, compress. Those belong to other tools.
+
+## If the project already generates media
+
+Another CLI, a hosted service, a script in the repo, a provider the user already pays for:
+use that one. `mediagen` is one option among several, so reach for it when it is what the
+project has, or when the user asks for it by name.
+
+## Setup, once
+
+```bash
+mediagen config show                          # any command writes and validates config.toml on first use
+mediagen login --api-key sk-...               # the key for the default provider; --provider <name> for another
+mediagen auth status                          # which key each provider would use, masked
+```
+
+Providers, their protocols and their models live in `config.toml`; `mediagen config add`
+creates an entry without hand-editing, and `mediagen config show` prints what resolves.
+The README covers the file's shape.
+
+## Four commands worth copying
+
+```bash
+# a new image; files land in --out-dir (default: config output_dir, else the cwd)
+mediagen image generate "a red fox in fresh snow, soft morning light" --out-dir ./out
+
+# change a local image (local paths are read and sent as data URLs)
+mediagen image edit "make the sky sunset orange, keep the composition" --image ./photo.jpg
+
+# a video task without waiting for the render (the id comes back immediately)
+mediagen video generate "a fox running through deep snow, golden hour" --no-wait --json
+
+# collect a task you already paid for; nothing is created and nothing is billed
+mediagen video status cgt-20261028-abcd --json
+```
+
+These go to whichever entry `config.toml` marks as its default. `--provider <name>` picks a
+different entry from that file; the names in the CLI's own examples are only what the first
+run seeds, and `mediagen config show` prints the ones this machine has.
+
+Add `--json` whenever you intend to parse the result: without it stdout is one absolute
+path per line and warnings go to stderr. Keep `--n` at 1 while iterating on a prompt: on
+some dialects each image is a separate paid request (`image generate --help` says which).
+
+## Rules
+
+1. Report the saved path(s) exactly as printed. That is the deliverable; a generation the
+   user cannot find was wasted money.
+2. Never re-run a request that already produced files, and never re-submit a video task
+   whose id you have: `video status <id>` collects it without paying again. Retry only a
+   clean failure that produced no task at all.
+3. Never print an API key. The CLI only ever prints it masked.
+4. When your own command timeout is short, prefer `video generate --no-wait --json`: a
+   blocking generate that you kill mid-poll leaves a paid task behind.