Skill loading doesn't process the triggering prompt - requires double entry

Description

When a user prompt triggers a skill (e.g., typing "commit this" which triggers the commit skill), the skill loads successfully but the original prompt is not processed. The user must repeat their request a second time for the skill to actually execute.

Steps to Reproduce

1. Have a custom skill installed (e.g., ~/.claude/skills/commit/SKILL.md)
2. Type a prompt that triggers the skill, such as "commit this"
3. Observe that the skill loads but returns to prompt without acting
4. Type the same prompt again - now it works

Expected Behavior

After the skill loads, the original triggering prompt should be processed with the skill context, executing the workflow in a single interaction.

Actual Behavior

1. Skill is triggered and loads: Skill(commit) - Successfully loaded skill
2. Some garbled markdown output appears (partial rendering of skill file code blocks)
3. Returns to prompt without processing the request
4. User must repeat the prompt to get the skill to actually execute

Screenshot

_(See attached screenshot showing the two-step behavior)_

The screenshot shows:

• First "commit this" → skill loads, garbled output, returns to prompt
• Second "commit this" → actually processes the commit workflow

Environment

• Claude Code CLI
• macOS Darwin 25.2.0
• Custom skill in ~/.claude/skills/commit/SKILL.md

Additional Context

The garbled output ([runs git commit] followed by triple backticks) appears to be malformed rendering of markdown code fences from within the SKILL.md file. The skill file itself is valid markdown with proper frontmatter.

Using the explicit /commit slash command may route differently, but natural language triggers like "commit this" exhibit this double-prompt behavior.