Write a skill that survives compaction

The skill is short on purpose. SKILL.md holds what must survive compaction (the rules, the steps, the output format) in about 650 tokens. The eight-topic checklist, which is long and only needed while drafting, lives in reference.md. Two small tools turn the compaction behaviour from the docs into numbers: check_skill.py checks the limits, and reattach.py predicts which skills come back after /compact. The decision that matters most is size, not placement alone: rules at the top protect against truncation, but only a small skill protects against being dropped entirely when the 25,000-token budget fills up.