Claude ignores skill triggers even when user request clearly matches skill description

Description

Claude Code ignores configured skills even when the user's request exactly matches the skill's trigger description.

Steps to Reproduce

1. Configure a skill with a clear trigger description, e.g.:

• Skill: coderabbit-workflow
• Description: "Use when the user asks to address, fix, or handle CodeRabbit review feedback on GitHub Pull Requests"

1. Ask Claude: "address coderabbit's comments on the pull request"

1. Expected: Claude invokes the coderabbit-workflow skill
2. Actual: Claude manually performs what it assumes the task means (fetching comments, editing code, committing) - completely bypassing the skill

The Problem

Claude assumes it "knows how to do it" and proceeds with its own interpretation instead of using the defined skill. This is problematic because:

1. Skills encode user-specific workflows - The user may want specific actions (e.g., posting reply comments to CodeRabbit, specific commit formats, etc.) that Claude cannot know without the skill
2. Claude's assumptions are wrong - In this case, Claude only fixed code but didn't post reply comments, which may have been part of the intended workflow
3. The trigger matched exactly - "address coderabbit's comments" clearly matches "address... CodeRabbit review feedback"

Expected Behavior

When a user's request matches a skill's trigger description, Claude should invoke the skill rather than attempting to handle the task manually with its own assumptions.

Environment

• Claude Code CLI
• Model: claude-opus-4-5-20251101