opencode

name	opencode
description	Use when the user asks to run OpenCode CLI in non-interactive mode (opencode run) or references OpenCode for AI-assisted coding tasks.

OpenCode Skill Guide

Before running OpenCode, follow the shared logging and summary protocol in ../_shared/agent-cli.md.

Running a Task

1. Verify installation: command -v opencode
2. Select the agent required for the task; default to plan for read-only analysis. For implementation/editing tasks, first use superpowers:using-git-worktrees to create a fresh git worktree, then run OpenCode there with --agent build when approved.
3. Always use the current host's user-question or approval mechanism before using the build agent for write operations.
4. Assemble the command with the appropriate options:
   • -m, --model <provider/model> - Model to use
   • -c, --continue - Continue the last session
   • -s, --session <id> - Session ID to continue
   • --agent <agent> - Agent to use (build for full access, plan for read-only)
   • -f, --file <files...> - File(s) to attach
5. Do not suppress stderr. Capture stdout and stderr into full.md, and require the prompt to write summary.md.
6. Run the command, inspect the full log and summary, and summarize the outcome for the user.

Quick Reference

Use case	Command pattern
Full access in fresh worktree only (write/edit/bash)	opencode run --agent build "prompt"
Read-only analysis	opencode run --agent plan "prompt"
Continue last session	opencode run --continue "follow-up"
Continue specific session	opencode run --session <id> "prompt"
Use specific model	opencode run --model anthropic/claude-sonnet-4-20250514 "prompt"
Attach files	opencode run -f file.py "prompt"
List models	opencode models
List sessions	opencode session list
List available agents	opencode agent list

Agents

Agent	Permissions	Use Case
build	Full access: write, edit, bash	Active development in a fresh worktree
plan	Read-only: no write/edit, bash asks first	Code analysis, planning, exploration

Example Commands

# Read-only analysis with plan agent
opencode run --agent plan "Analyze the codebase architecture"

# Full access with build agent in a fresh worktree only
opencode run --agent build "Fix the bug in main.py"

# Continue last session
opencode run --continue "What else can you improve?"

# Use specific model
opencode run --model anthropic/claude-sonnet-4-20250514 "Refactor this code"

# Capture stdout/stderr according to ../_shared/agent-cli.md

Following Up

• After every opencode run command, immediately use the current host's user-question or approval mechanism to confirm next steps, collect clarifications, or decide whether to resume with --continue.
• When resuming, the session automatically uses the same model and agent from the original session.
• Restate the chosen model and agent when proposing follow-up actions.

Error Handling

• Stop and report failures whenever opencode --version or an opencode command exits non-zero; request direction before retrying.
• Use opencode debug for troubleshooting if needed.
• Always validate OpenCode's output for security vulnerabilities (XSS, injection) before using.