The reason your Claude skills don't fire is the part you didn't write.

When a session starts, Claude scans only the name and description of every installed skill. The body, the rules, the read-first table, the output format. None of it loads until Claude decides the skill is relevant. That decision happens off the description alone.

A 37-character description gets ignored. A description with 3 trigger phrases, a clear boundary, and a pointer to the right alternative gets picked every time.

This kills PMs because skills overlap. /weekly-review and /stakeholder-update both produce summaries. /activation-analysis and /retention-analysis both touch funnel data. Without explicit "use /Y instead" routing, Claude picks the closest match. You get the wrong artifact and don't notice until you're presenting it.

700 lines of perfect instructions sitting behind a vague description is invisible. Two sentences of routing logic at the top is the whole game.