Add OpenCode winui-dev orchestrator agent

[Zhanyuanium (Zhanyuanium)]

Description

Exposes the winui-dev orchestrator to OpenCode so OpenCode users get the same end-to-end WinUI 3 build flow as Copilot CLI (opencode run --agent winui-dev).

OpenCode loads custom agents from <config-dir>/agent/*.md markdown files (frontmatter metadata + body as the system prompt), so this adds a small adapter that mirrors the existing Copilot plugins/winui/agents/winui-dev.agent.md:

• plugins/winui/opencode/agent/winui-dev.md — a mode: all agent (selectable as primary, spawnable as subagent) with question and plan_enter permissioned allow to match OpenCode's native build agent, so it can drive interactive builds.
• Prompt adaptations: the self-hop guard uses subagent_type instead of agent_type, the general-purpose helper maps to OpenCode's general subagent, and skills load via /winui-dev-workflow and /winui-design slash commands.
• README documents the one-time agent + skills junction setup.

Related Issue

Part of #149

Type of Change

• ✨ New skill or feature

Affected area

• Agent (plugins/winui/agents/) (new OpenCode agent: plugins/winui/opencode/agent/winui-dev.md)
• Repo-level docs / governance

Checklist

• Tested locally on Windows (build + agent invocation if applicable)
• If a CLI command or agent invocation changed: README.md updated
• If user-facing: added a bullet to ## [Unreleased] in CHANGELOG.md
• Did not edit any version field in plugins/winui/plugin.json, .github/plugin/marketplace.json, or .claude-plugin/marketplace.json (versions bump only on the staging → main promotion PR — see RELEASING.md)

Screenshots / Demo

Additional Notes

AI Description

This section is auto-generated by AI when the PR is opened or updated. To opt out, delete this entire section including the marker comments.

[Nikola Metulev (nmetulev)]

This is great, but i wonder if there is a way to not duplicate the agent prompt - or a way to enforce it in sync?

[Zhanyuanium (Zhanyuanium)]

Good point. The two agent files are near-identical — only the frontmatter schema, the self-hop guard, and the skill-invocation syntax really differ (and those are runtime-specific). Easiest fix: move the shared Process + Best Practices into a skill, and leave each agent as a thin shell. That kills the duplication instead of policing it with a sync check. Sound good?