feat: add claude.md and agents.md as documentation

feat: add comment-audit skill for agent-driven comment audits

feat: cut first release

feat: initial public release

A benchmark for whether inline comments steer AI coding agents on small
surgical edits. Four scenarios in the gift-card refund domain test where
the protected invariant (24h cap) lives: branch, accumulator, helper, or
comment-only. Seven comment treatments per scenario.

Deliverable is comment-policy.md. RESULTS.md describes the result patterns
the benchmark probes for.