opencode

name	opencode
description	Delegate coding tasks to the OpenCode CLI for feature work, refactoring, PR review, and autonomous-style runs. Requires opencode installed where the agent can execute shell commands.
version	1.0.0
license	MIT
metadata	{"clawcode":{"tags":["Coding-Agent","OpenCode","CLI","Refactoring"],"related_skills":["coding-standards","python-patterns","codex"]}}

OpenCode CLI (clawcode)

Use OpenCode as an external coding agent when the user wants OpenCode specifically.

Do not confuse these entry points (they are not interchangeable):

Entry	Role
/opencode-cli (TUI slash)	One-shot probe or command via opencode_cli_bridge: merged stdout/stderr, not a PTY session and not multi-turn TUI control. Default args often --version.
bash	One-shot shell in agent turns, e.g. opencode run '…'.
terminal + process (Agent tools)	Multi-turn driving of a background shell/OpenCode TUI: terminal with background=true (and pty=true on POSIX with clawcode[terminal-pty]), then process for poll / submit / kill. Requires CLAWCODE_TERMINAL_ENV=local for full PTY on Linux/macOS; Windows host is limited—use WSL or Linux for interactive TUI automation.

Optional: pip install 'clawcode[terminal-pty]' for ptyprocess on POSIX.

When to use

• User asks to use OpenCode or opencode.
• You need a bounded task: prefer opencode run '...' via bash in the project workspace.
• You only need to verify the CLI is installed: /opencode-cli (default opencode --version) after /claw; same terminal stack as /claude-cli (CLAWCODE_TERMINAL_ENV / TERMINAL_ENV).

Prerequisites

• Install OpenCode (e.g. npm i -g opencode-ai@latest or vendor instructions).
• Auth: opencode auth login or provider env vars (see OpenCode docs).
• Check: opencode auth list should list at least one provider when run in the same environment as the agent.

clawcode-specific capabilities

Goal	What to use
Quick probe (version, install check)	/opencode-cli only—single run, not multi-turn. Same Claw gate as /claude-cli.
Run a command in agent turns	bash: e.g. opencode run 'your task' with working_directory set to the repo root.
Docker/SSH terminal backend	Same as bash: settings.shell.use_environments_backend and CLAWCODE_TERMINAL_*; ensure the image or remote has opencode on PATH.
Interactive OpenCode TUI (agent-driven)	terminal (background=true, pty=true where supported) + process (poll / submit / …). Non-local backends: log-polled background only (no stdin). Optional check_interval on terminal schedules completion notifications in the TUI (see CLAW_SUPPORT_MAP.md).

One-shot tasks (recommended)

Use opencode run for bounded work:

opencode run 'Add retry logic to API calls and update tests'

Attach files with -f when needed. Use --thinking, --model, etc. per OpenCode CLI help.

Binary resolution

If which opencode differs between environments, use an absolute path in the bash command:

"$HOME/.opencode/bin/opencode" run '...'

Pitfalls

• On Windows, prefer WSL/Linux for full PTY TUI automation; pipe mode or non-local sandboxes do not match a full local PTY.
• Interactive opencode still benefits from a real user terminal for some UX; bash alone does not replace PTY driving.
• opencode exit codes and stderr: see bash tool output; /opencode-cli shows stdout/stderr blocks in the chat UI.
• Avoid /exit in OpenCode TUI (invalid per OpenCode UX); prefer Ctrl+C in a real terminal session.

Verification

opencode --version
opencode run 'Respond with exactly: OPENCODE_SMOKE_OK'

Rules

1. Prefer opencode run for automation.
2. Scope work to a single repo root per task when possible.
3. Use /opencode-cli only for quick checks or demos; heavy work goes through bash with explicit commands.