Desktop TUI: /skill-name prefix injects full SKILL.md into conversation instead of using skill tool

[ericheroster]

Description

In the opencode desktop TUI, typing /skill-name (e.g., /雪球-cli) and sending the message causes the entire SKILL.md content to be injected directly into the conversation context as visible text. This is inconsistent with how skills should work — the agent should discover and load the skill via the native skill tool, not get the raw markdown dumped inline.

Steps to Reproduce

1. Open opencode desktop TUI
2. Type / to trigger the skill list popup
3. Select a skill from the list (e.g., 雪球-cli)
4. The input becomes /雪球-cli
5. Press Enter to send
6. The full content of SKILL.md is loaded into the conversation as visible output

Workaround

Delete the leading / before sending — the agent then correctly uses the skill tool to load the skill, showing only a concise tool invocation.

Expected Behavior

/skill-name should behave the same as typing the skill name without / — i.e., the agent should decide to load the skill via the skill tool on its own. The / prefix should only serve as a selection/discovery mechanism in the autocomplete popup, not trigger a premature injection of the skill content.

Environment

• opencode desktop TUI (latest version)
• Windows

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• /skill-name doesn’t invoke full skill system #24831: /skill-name doesn't invoke the full skill system — same core complaint that the slash command bypasses the skill tool
• Slash commands for skills inject raw content AND expose the skill tool, causing duplicate context #26185: Slash commands for skills inject raw SKILL.md content AND expose the skill tool, causing duplicate context — nearly identical description of the raw content injection behavior