catalog / cross-platform

"Don't assume which tool to use" is unenforceable without an in-skill picker

verified

Telling the model 'don't assume which tool to use' in SKILL.md doesn't reliably prevent assumption — only an explicit AskUserQuestion/ask_user_input_v0 picker does.

  • Category: cross-platform
  • Applies to: Claude.ai ❓ | Claude Code ✅ | Claude Desktop ❓
  • Verified on: Claude Code, 2026-05-17
  • Verification tier: verified
  • Severity: Silent UX divergence

Symptom

A SKILL.md that instructs "do not assume which tool to use; let the user pick" runs on Code with multiple connected MCP servers, and Claude silently picks one — overriding the SKILL.md instruction.

Cause

There is no platform-level picker UI on Code. With multiple candidates available, the model infers from inventory and proceeds. Verified 2026-05-17: with several MCP servers connected, Claude picked Atlassian without surfacing the choice to the user despite the SKILL.md saying not to assume.

Fix

If user choice matters to the skill, the skill body itself must invoke an elicitation tool (AskUserQuestion / ask_user_input_v0) with the candidate options. Don't delegate the choice to "the platform."

Notes

Related: mcp-discovery-layer-absent-on-code.md.