ripwire-write-tests

# Writing tests with ripwire > Routing: > • Vetting a diff you already wrote (does it have test cover) → **ripwire-change-check**. > • Judging whether your OWN code got better/worse → **ripwire-quality-bar**. > • Don't know which symbol is even the bug → **ripwire-find-bug**. > • Not sure which skill? → **ripwire-router**. `<dir>` = repo root. This skill answers "what should I write a test for, and what does it need to cover" — not "is my diff safe" (that's change-check). ## Find the gap 1. **Untested integration seams** — `ripwire <dir> --seams` → `<seams>` lists cross-module call edges no test reaches. These are the highest-value targets: a bug here crosses a boundary silently, and one test per seam buys the most coverage per line written. Scope with `<dir>` = a subsystem to focus the seam list. 2. **The `tested=1` coverage lens** — `ripwire <dir> --metrics --top-k=50` (or `--for="<area>"`, which carries the same lens inline) → `tested="1"` means an indexed test-path symbol transitively reaches this production symbol. On these explicit metric surfaces the attribute is omitted when no indexed test reaches it; the binary never prints a fabricated zero. Cross an omitted `tested` with `in=` (fan-in): a widely-relied-on function with no indexed safety net is a sharper priority than an uncovered leaf. Shell/subprocess test coverage remains outside the call graph, so absence is a structural finding, not proof that no external test exists. 3. **Pick the ...