feat(skills): add wavespeed skill

[chengzeyi]

• A human has tested these changes.

---

Why

OpenHands has no skill for generating media assets. Tasks like "make a hero image for this landing page" or "animate this product photo into a short clip" currently have no guided path. WaveSpeed is an inference platform for generative media models (Seedream, Seedance, MiniMax H3, Wan, Kling, Nano Banana, Veo and a few hundred more) with an open-source CLI, which makes it a good fit for a sandboxed agent.

Summary

• Add skills/wavespeed/ (SKILL.md with triggers, README, .plugin/plugin.json, vendor symlinks): teaches the agent the find → inspect → run workflow on the wavespeed CLI — search the live catalog, read the model's input schema, quote the price, run with --json, upload local files with the @path marker. Never asks the user to paste an API key into the chat.
• Register it in marketplaces/openhands-extensions.json (category integrations) and regenerate skills/index.js with npm run build:skills.

Issue Number

None.

How to Test

1. npm run build:skills — regenerates skills/index.js with 61 skills, no warnings for wavespeed.
2. pytest tests/test_skills_catalog.py tests/test_skills_have_readme.py tests/test_plugin_manifest.py tests/test_catalog_schema.py — passes locally. test_skill_plugin_loading.py::TestIteratePluginLoading needs the openhands SDK and fails identically on a clean checkout without it.
3. In a workspace with @wavespeed/cli installed and WAVESPEED_API_KEY set: ask the agent to "generate a 16:9 image of a lighthouse at dawn"; it should run wavespeed models, then wavespeed run <model> -p ... --json, and return the output URL.

Video/Screenshots

Not applicable; Markdown and catalog changes only.

Notes

The same skill is maintained at https://github.com/WaveSpeedAI/agent-skills (MIT) alongside per-model skills and an mcp.json for the @wavespeed/mcp server. Happy to move the source of truth to this registry or keep it synced, whichever you prefer.