Feature request: Global skill trust setting (analogous to enableAllProjectMcpServers)

Problem

When using skills (user-level, in ~/.claude/skills/), Claude Code shows a content trust prompt every time a skill is invoked in a new project directory:

Use skill "linear-image"?
Claude may use instructions, code, or files from this Skill.

Do you want to proceed?
1. Yes
2. Yes, and don't ask again for linear-image in /Users/me/Code/project3
3. No

Option 2 stores trust per-project-path per-skill. For users with multiple git worktrees of the same repo (e.g., 8 worktrees for parallel development), this means approving the same self-authored skill 8 separate times — once per worktree path.

The PermissionRequest hook fires and can auto-approve the Skill tool call, but the content trust dialog is a separate layer that hooks cannot handle.

Proposed Solution

Add a global setting to auto-trust user-level skills, similar to how enableAllProjectMcpServers handles MCP server trust:

{
  "trustAllUserSkills": true
}

Or more granularly:

{
  "trustedSkills": ["linear-image", "co-comms", "context7"]
}

This would suppress the content trust prompt for skills in ~/.claude/skills/ across all project directories.

Why This Makes Sense

• User-level skills are self-authored. They live in ~/.claude/skills/, not in untrusted project directories. The user already controls their content.
• Precedent exists. enableAllProjectMcpServers: true auto-trusts MCP servers, which have a larger attack surface (arbitrary tool execution) than skills (instruction injection).
• Worktree workflows are common. Any developer using git worktree for parallel branches hits this friction multiplied by worktree count.
• The PermissionRequest hook gap. Hooks can gate the Skill tool call but not the content trust dialog, so there's no workaround available today.

Environment

• Claude Code (latest)
• macOS
• 8 git worktrees of the same monorepo (~/Code/magic0 through ~/Code/magic7)
• ~30 user-level skills in ~/.claude/skills/