[BUG] TUI: Non built-in skills do not pass prompt context when invoked as slash commands [Claude Code, Opencode, Github Copilot] [Dev, Production]

[flora131]

Description

When non built-in skills are invoked via slash commands, the additional prompt context provided after the command is not passed to the skill as input. However, the same skills work correctly when invoked via natural language prompting, where the full context is passed through.

Affected Agents

• Claude Code
• Opencode
• Github Copilot

Environment

• Dev (bun run / local development)
• Production (binary build)

Reproduction Steps

1. Launch Atomic with any coding agent
2. Invoke a non built-in skill via slash command with additional context, e.g.:
   • Type: /technical-blog-optimizer write a blog post for me about games written in Rust
3. Observe that the skill is triggered but the context ("write a blog post for me about games written in Rust") is NOT passed to the skill
4. Now invoke the same skill via natural language:
   • Type: use the technical blog optimizer skill to write me a blog about games in Rust
5. Observe that the full context IS passed to the skill and it processes the request correctly

Expected Behavior

When a non built-in skill is invoked via slash command (e.g., /technical-blog-optimizer write a blog post about X), the text following the slash command should be passed to the skill as input/context, just as it would be when invoked via natural language.

Actual Behavior

• Slash command invocation: The skill is triggered but the additional prompt context after the command name is dropped/ignored. The skill executes without knowing what the user wants it to do.
• Natural language invocation: The full prompt context is correctly passed to the skill, and it executes as expected.

Additional Context

• This is specifically for non built-in skills (e.g., custom/project-level skills)
• Built-in skills may or may not exhibit the same behavior (separate issue)
• Affects all coding agents (Claude Code, Opencode, Github Copilot)

Investigation Notes

IMPORTANT: Do NOT assume or speculate on the root cause in the initial filing. The purpose of this section is to capture raw observations only. Root cause analysis happens during investigation.

• The skill itself is correctly triggered via slash command — the issue is specifically that the trailing text/context is not forwarded
• Natural language invocation works correctly with the same skill, suggesting the context parsing/forwarding is handled differently between the two invocation paths