opencode-driven-development

Multi-file refactoring and test writing can be automated by delegating the heavy engineering work to OpenCode's multi-agent harness. This offloads complex parallel subtasks, file edits, and tool calls, satisfying requirements through structured JSON event streams without manual coordination.

Iterative improvements during feature implementation are managed using multi-turn sessions. By maintaining an OpenCode session, you can apply Hermes feedback across multiple interactions, allowing the multi-agent harness to execute ongoing parallel subtasks and edits continuously.

Multi-agent orchestration handles bug fixing by dispatching the work through OpenCode's CLI, which parses JSON event streams into structured results. This enables parallel execution of many tool calls and multi-file edits required to resolve the bugs efficiently.

You should dispatch coding tasks to OpenCode when they require many tool calls, multi-file edits, and parallel execution. A decision framework guides the right choice between doing it yourself, using a standard subagent, or offloading heavy engineering to OpenCode.

Yes, you can run tests in a specific target directory by using the run action with a specific prompt. You describe the feature, list affected files, and specify that tests must be run in the target directory before OpenCode executes the task.

Limitations to consider include the need to validate inputs and parse JSON event streams from the CLI. If a task does not require many tool calls, multi-file edits, or parallel execution, using a multi-agent harness may be unnecessary compared to simpler subagent approaches.