A skill is a directory under .claude/skills/ holding a SKILL.md — frontmatter (name, description) plus an instructional body.
There is no central routing table: Claude Code auto-loads a skill when its description: matches what the current conversation is doing.
That makes the description the entire activation mechanism — a skill with a vague or narrow description simply never fires, silently, with no error to debug.
Dotboost ships nineteen skills.
Eighteen are dotkernel-*, each covering one slice of Dotkernel convention; the nineteenth, dependency-policy, is generic to any package decision and detailed enough to warrant its own page.
Most dotkernel-* skills apply to every application variant; dotkernel-api-endpoint applies only in an API application, where it is the end-to-end checklist for a new or changed endpoint and points to the skill that owns each topic.
Each skill's description is written to include the words a developer would actually type — "where does this go", "403", "InputFilter" — not just the framework term for the concept. If a skill isn't triggering when you expect it to, the fix is almost always to add the phrase you used to its description, not to rewrite its instructions. See Maintaining This.
Reference groups all nineteen by theme with a one-line trigger description each.