[FEATURE] CLI command to list resolved skills without starting a session (no API call)

Problem Statement

There is no way to list the skills that Claude Code has resolved/loaded without starting an interactive session — which requires an API call. This creates two failure modes:

1. Spend-limit lockout: When the Anthropic account hits its monthly spend limit, claude cannot start a session, so there is no way to verify which skills are loaded. Operators cannot debug skill loading topology, verify lean-set configurations, or confirm junction/symlink changes without waiting for the spend limit to reset.

1. No CI/automation path: Scripts and CI pipelines that need to verify skill loading (e.g., after topology changes, junction repointing, or lean-set swaps) have no programmatic way to check the resolved skill set without consuming API credits.

Current Behavior

• claude --help → no skills subcommand
• claude doctor → checks installation health but does not list resolved skills
• /skills → only available inside a running session (requires API call)
• The only way to see which skills are loaded is to start a session and ask the model or use /skills

Proposed Solution

Add a CLI command that resolves and lists skills without starting a session or making any API call:

# List all resolved skills (name + source path)
claude skills list

# JSON output for scripting
claude skills list --json

# Show count only
claude skills list --count

# Filter by source path
claude skills list --source .claude/skills

This would:

• Resolve skills from all configured paths (.claude/skills, .agents/skills, plugins, etc.)
• Apply the same discovery logic as session startup
• Output the result without connecting to the Anthropic API
• Work even when the account is at its spend limit

Use Cases

1. Spend-limit debugging: Verify skill topology when API is unavailable (our current situation — 675 skills expected, no way to confirm without a session)
2. CI verification: After topology changes (junction swaps, lean-set promotions/demotions), run claude skills list --count to verify the expected count
3. Fleet automation: Scripts that manage skill loading across multiple runtimes (claude-code, codex, devin) can verify each runtime's resolved set without API calls
4. Lean-set reviews: Programmatic skill count verification as part of lean-set review pipelines

Related Issues

• #79503 — Asks for in-session registered-vs-dropped diagnostic (sibling issue; this request is for pre-session CLI diagnostic)
• #76001 — Hierarchical skill discovery to reduce skill-list metadata overhead
• #81081 — Session-start skill listing silently truncates descriptions

Environment

Claude Code 2.1.220, Windows 11