| name | opencode-cli |
| description | Interface with the Opencode CLI to execute AI-driven coding tasks, generate code snippets, and manage agent sessions non-interactively. Use this when the user asks to "ask Opencode" or "use Opencode" to perform a task. |
Opencode CLI Skill
Goal
This skill allows the agent to utilize the opencode command-line tool to perform coding tasks, query the Opencode agent, and manage configuration. It is specifically optimized for non-interactive usage to prevent the agent from getting stuck in Terminal UI (TUI) modes.
Usage Rules
1. Non-Interactive Execution (CRITICAL)
- NEVER run the bare command
opencode without arguments. This launches an interactive TUI that will hang the agent session.
- ALWAYS use
opencode run "<prompt>" for generation, questions, or coding tasks.
- ALWAYS use the
-p or --prompt flag if run is not applicable in your specific version, but opencode run is the standard for delegation.
2. Output Handling
- The output from Opencode is standard text. Read the output to confirm the task was completed or to retrieve the generated code.
- If the user asks for JSON output and the CLI supports it, append
--output-format json.
3. Context & Directories
- If the user specifies a working directory, use the
-c or --cwd flag:
opencode run "prompt" --cwd /path/to/dir
Command Reference
| Action | Command Pattern |
|---|
| Run Prompt | opencode run "Your instruction here" |
| Check Version | opencode --version |
| List Models | opencode models |
| Auth Status | opencode auth list |
| Set Directory | opencode run "..." --cwd ./subdir |
Examples
Example 1: Generative Task
User: "Use opencode to write a Python script that calculates Fibonacci numbers."
Agent Execution:
opencode run "Write a Python script that calculates Fibonacci numbers and prints the first 10."
