**UTILITY SKILL** — Decide where a change belongs in the NVDA AI Assistant. USE FOR: determining which layer owns a feature; scoping cross-boundary protocol work; identifying owning abstraction before editing. DO NOT USE FOR: implementing features; writing code; general coding questions. INVOKES: file reads (docs/architecture.md), grep. FOR SINGLE OPERATIONS: skip this skill and edit directly.
**UTILITY SKILL** — Choose the right validation checks for NVDA AI Assistant changes across Python add-on, Rust host, Web UI, or cross-boundary protocol updates. USE FOR: validating Python add-on code; checking Rust host compilation; building Web UI after changes; validating cross-boundary protocol updates; choosing the smallest reliable validation. DO NOT USE FOR: implementing features; writing code; debugging runtime errors; running tests unrelated to validation. INVOKES: terminal (ruff, cargo check, npm build), file reads. FOR SINGLE OPERATIONS: For a quick single-layer lint or build, run the command directly — no skill needed.
**UTILITY SKILL** — Add or change translator-facing WebView UI labels, debug missing POT entries, or move label ownership from JS/Rust to Python. USE FOR: translatable WebView labels; missing msgid in POT; moving label ownership to Python; translator comments; gettext extraction. DO NOT USE FOR: implementing features; non-WebView i18n; Rust/JS logic. INVOKES: file reads (session_state.py, buildVars.py, gettexttool), terminal (xgettext, npm build, ruff). FOR SINGLE OPERATIONS: add the string directly.