using-opencode-cli

What problem does it solve?

This Skill solves the problem of running reliable local OpenCode agent tasks from Codex without losing control over prompts, model settings, session continuity, and output monitoring.

Core Features & Use Cases

• Deterministic OpenCode invocation rules: Enforces PowerShell 7 usage, default model selection, fixed reasoning depth (variant max), and prompt handling where Codex fully composes the prompt and the script passes it through without auto-adding extra guidance.
• Three execution modes with the right monitoring strategy: Supports observe mode for interactive one-off tasks, managed mode for long-running multi-turn tasks with RunDir state tracking, and front-of-screen style foreground short tests that emit only session ID and assistant text.
• Session reuse and status summarization: Reuses sessionID for the same code path/plan, supports fork vs continue behavior, and provides status summaries via metadata.json/status.json plus optional lightweight event hooks.

Real-world use case: use this Skill to perform a multi-round repository security/code review by running OpenCode locally in managed mode, polling status by RunDir, and continuing the same session across multiple checkpoints without reopening fresh threads that dilute context.

Quick Start

Ask Codex to run a local OpenCode task in observe mode using the task markdown file docs/AI协作/本地Agent/进行中/[目录]/任务_[主题]_OpenCode.md, so you can see the session progress in a visible Windows Terminal window and capture the final conclusion.