No skill names, no trigger words, no stack required — your profile (or the skills' own selection rules) fills in the rest.
The prompt is matched against every skill's description; the sota router maps task + mode (BUILD or AUDIT) to the domains it touches.
Each skill's index picks the matching rules files (each < 500 lines) — never the whole library, so context stays lean.
BUILD applies the rules while writing and self-checks the diff against each audit checklist; AUDIT emits file:line | rule | severity | effort | fix.