Opencode-controller

**Judgement:** A focused, high-structure controller skill for Opencode that is very suitable if you already work inside Opencode and want a disciplined, auditable Plan → Build coding workflow. It is not a general-purpose coding helper outside that ecosystem. **What it does well:** - Orchestrates Opencode via explicit slash commands for sessions, agents, and models. - Enforces a strong separation of planning and coding: the assistant never writes code, only Opencode does. - Guides a rigorous Plan → Build loop with plan review and revision before implementation. - Handles provider selection and authentication transparently by surfacing login links verbatim and waiting for user confirmation. **Key risks / limitations:** - Requires users to have and use Opencode; on its own it won’t help users who expect direct code output in the chat. - The strict Plan-first and Plan ↔ Build switching rules add interaction overhead and may feel slow for small or trivial tasks. - Misuse (e.g., user insisting on code directly in chat) will conflict with the skill’s core rule that Clawdbot does not write code. **Recommended scenarios:** - You are already invested in Opencode and want the assistant to manage sessions, models, and agents rather than write code itself. - You care about process discipline (clear plans, review before implementation) for medium-to-large coding tasks. - You want all significant actions (session changes, agent switches, model choices, auth steps) to be explicitly visible and reproducible via slash commands.