System prompt misleadingly describes opencode as "CLI tool" in Desktop App mode

[changxumiao]

Description

System prompt inaccurately describes the running mode as "CLI tool"

Description

The system prompt starts with the line:

"You are opencode, an interactive CLI tool that helps users with software engineering tasks."

This static description does not distinguish between different running modes (CLI, TUI, Desktop App). When running in the Desktop App (or potentially TUI mode), the LLM still receives "interactive CLI tool" as its identity, which is misleading.

Impact

This caused the LLM to incorrectly believe it was running in a pure CLI environment, leading to:

• Incorrectly stating that MCP tools are unavailable (when they are actually injected and fully functional)
• Confusion about which tools can be used

The MCP tools and other desktop-specific capabilities are correctly injected at the tool level, so the system works fine functionally — the issue is purely about the misleading initial context.

Suggested Fix

Update the system prompt to use a more generic description, for example:

"You are opencode, an AI coding agent that helps users with software engineering tasks."

Or better, dynamically include the running mode in the context so the LLM can adapt its expectations accordingly.

Environment

• opencode Desktop App
• Windows

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[github-actions]

This issue might be a duplicate of an existing (closed) issue. Please check:

• System prompts force the agent to target terminal output, even in OC Desktop #15508: Same concern — system prompts contain "CLI tool" language which is misleading in the OpenCode Desktop app context

If the previous issue was closed without a fix or the fix regressed, please mention that in this issue so maintainers can track it appropriately.

[Arjith8]

You are contradicting urself a bit since at first u said

This caused the LLM to incorrectly believe it was running in a pure CLI environment, leading to:
- Incorrectly stating that MCP tools are unavailable (when they are actually injected and fully functional)
- Confusion about which tools can be used

then

The MCP tools and other desktop-specific capabilities are correctly injected at the tool level, so the system works fine functionally — the issue is purely about the misleading initial context.

So this feels purely theoretical from the explanation provided. I think it would be beneficial to the actual maintainers if you could maybe specify a instance where u faced issue, maybe a screen shot or similar.

I havent had any issues myself but if u have any general steps to reproduce I could try it out too.