Slim agent suite stops premium models from doing recon

What it does oh-my-opencode-slim is an orchestration plugin for OpenCode that breaks down software tasks and delegates them to a built-in roster of specialist agents—scouting codebases, reviewing architecture, handling UI, and fixing bugs—rather than feeding everything to a single monolithic model. It ships with pre-configured agent roles and model assignments designed to balance output quality against token spend.

The interesting bit The V2 beta turns the orchestrator from a simple executor into a background scheduler: it queues specialist agents as asynchronous tasks, polls their status, and reconciles results before continuing, while a /deepwork command translates large goals into file-backed execution plans. The plugin also explicitly pairs cheap, fast models with low-stakes reconnaissance work and reserves heavy hitters like openai/gpt-5.5 or anthropic/claude-opus-4.6 for architectural decisions.

Key highlights

• Role-based “Pantheon” including Orchestrator, Explorer, Oracle, Council, Librarian, Designer, and Fixer, each with dedicated prompt logic.
• Model-agnostic presets for OpenAI and OpenCode Go, plus a documented $30 preset aimed at cheap setups.
• Automatic delegation rules in the orchestrator prompt, with manual override via @agentName <task>.
• V2 background orchestration for parallel specialist execution and /deepwork planning for complex, multi-file goals.

Caveats

• V2 background orchestration requires OpenCode’s experimental background subagent support and is explicitly labeled beta.
• The README does not quantify the claimed token savings versus the upstream fork.

Verdict Worth a look if you already live in OpenCode and want to stop paying premium model rates for codebase reconnaissance. Skip it if you prefer a single-model, single-turn workflow or aren’t prepared to tune a multi-agent configuration.