feat(skills): add production media generation

SKILLS/README.md

@@ -12,9 +12,10 @@
 
 ## Current active skills
 
-As of 2026-07-13, the active skill directories are:
+As of 2026-07-25, the active skill directories are:
 
 - `SKILLS/df-meta-mcp/`
+- `SKILLS/media-generation/`
 - `SKILLS/new-api-manage/`
 - `SKILLS/qwen3-asr-tts-hf2api/`
 - `SKILLS/vits-tts-hf2api/`

SKILLS/media-generation/.gitignore

@@ -0,0 +1,4 @@
+node_modules/
+*.log
+.env
+.DS_Store

SKILLS/media-generation/SKILL.md

@@ -0,0 +1,69 @@
+---
+name: media-generation
+description: Generate or edit images and generate videos through an OpenAI-compatible new-api endpoint. Use for gpt-image-2, gpt-5-image, nanobanana variants, Grok image, or Grok video; also use doctor to verify endpoint credentials and model routes.
+homepage: https://github.com/QuantumNous/new-api
+metadata: {"clawdbot":{"requires":{"bins":["node"]}}}
+---
+
+# Media Generation
+
+Use the bundled CLI from this skill directory. It needs no runtime package installation.
+
+```bash
+./media-generation <doctor|models|image|video> [options]
+```
+
+Set the exact environment variables before network commands:
+
+```bash
+export OPENAI_COMPATIBLE_BASE_URL="https://new-api.example.com/v1"
+export OPENAI_COMPATIBLE_API_KEY="..."
+```
+
+## Choose A Command
+
+- `doctor`: authenticate `GET /v1/models` and compare provider routes with the registry.
+- `models`: inspect aliases and capabilities locally; no credentials or network required.
+- `image`: generate an image or edit one or more local images.
+- `video`: submit a new-api video task, poll it, and stream the result to disk.
+
+## Common Calls
+
+```bash
+./media-generation doctor
+./media-generation models
+
+./media-generation image \
+  --prompt "A quiet harbor at dawn" \
+  --model gpt-image-2 \
+  --aspectRatio 3:2 \
+  --outputDir ./media
+
+./media-generation image \
+  --prompt "Replace the sky with morning light" \
+  --model gpt-5-image \
+  --inputImages ./base.png,./reference.png \
+  --maskImage ./mask.png \
+  --outputDir ./media
+
+./media-generation video \
+  --prompt "A slow cinematic camera move" \
+  --model grok-imagine-video \
+  --inputImage ./frame.png \
+  --seconds 5 \
+  --size 1280x720 \
+  --path ./media/clip.mp4
+```
+
+Options accept both `--field value` and `--field=value`. Repeat `--inputImages` or use a
+comma-separated value. There is no provider fallback and image retries are disabled.
+
+Success is one JSON object on stdout. Failure is one JSON object on stderr with exit code 1.
+Success payloads contain output metadata, never prompts or credentials.
+
+## Detailed References
+
+- [CLI_REFERENCE.md](references/CLI_REFERENCE.md): every argument, JSON envelopes, file behavior,
+  timeout and download limits.
+- [MODEL_REGISTRY.md](references/MODEL_REGISTRY.md): alias routes and per-model capability rules.
+- [PROTOCOLS.md](references/PROTOCOLS.md): doctor, AI SDK image, and new-api video HTTP flows.