Split skill loading into two layers: names in the prompt, bodies on demand.

Ten skills inlined at ~2,000 tokens each burns 20,000 tokens on knowledge the agent mostly doesn't need. Carry only names and one-line descriptions in the system prompt at roughly 100 tokens per skill, and expose a load_skill(name) tool that returns the full body wrapped in a tag when the model asks. Documented in a 72,734-star teardown alongside context compaction, subagent spawning, and worktree-isolated parallel execution as separately implementable units.