[Bug] Multi-line skill descriptions leak into autocomplete as separate entries

Bug Description
Built-in skill multi-line description leaks into autocomplete as separate entry

The claude-api built-in skill has a multi-line description that includes a "TRIGGER when..." line. In the slash command autocomplete menu, this second line appears as a separate autocomplete entry, rather than being part of the /claude-api skill's description.

The TRIGGER when: code imports anthropic/@anthropic-ai/sdk/claude_agent_sdk... line renders as its own row in the autocomplete list. It should either be part of the /claude-api description (single row, possibly truncated) or not appear at all since it's internal routing metadata, not user-facing.

Environment Info

• Platform: darwin
• Terminal: tmux
• Version: 2.1.85
• Feedback ID: 5cab29b3-4b3a-4b37-8ad4-9e1ebda63180

Errors

[{"error":"Error: NON-FATAL: Lock acquisition failed for /Users/marwan/.local/share/claude/versions/2.1.85 (expected in multi-process scenarios)\n    at vp_ (/$bunfs/root/src/entrypoints/cli.js:2794:2168)\n    at yn6 (/$bunfs/root/src/entrypoints/cli.js:2794:1326)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-04-05T00:08:14.681Z"}]