Health

H in the list, or:

opi --health

opi detects which tools the project depends on, runs them concurrently, and prints each result as it finishes, so a slow test run does not look like a hang. Run on opi’s own repository:

opi  health

✓ Format (rust)       0.1s  cargo
✓ Lint (rust)         0.1s  cargo
✓ Docs (rust)         0.1s  cargo
✓ Lockfile (rust)     0.0s  cargo
✓ Tests (rust)        0.3s  cargo

5 checks passed.

What is detected

Check	Tools, first match wins
TypeScript	astro check (with @astrojs/check), tsc --noEmit
Lint & format	Biome
Lint	ESLint
Format	Prettier
Tests	Vitest, Jest
Dead code	fallow, Knip
Secrets	nosecrets, secretlint
Lockfile	the package manager’s own frozen install, changing nothing — pnpm, npm and bun
Rust	cargo fmt --check, cargo clippy and cargo rustdoc with warnings denied, cargo test --all-features, and cargo metadata --locked where Cargo.lock is committed

A tool applies when the package declares it as a dependency; its binary is taken from a node_modules/.bin at or above the package. A yarn project using Plug’n’Play has no node_modules, so there the tool is started through yarn run instead. Where several tools answer for the same concern, only the first detected one runs, and its name is shown next to the result.

Lockfile asks whether the lockfile still matches package.json — the state in which every CI that installs with a frozen lockfile stops. Each manager is asked with a command measured to change nothing: pnpm install --frozen-lockfile --lockfile-only --offline, npm ci --dry-run, bun install --frozen-lockfile --dry-run. yarn gets no lockfile check, because its only strict form is a full install. It is the one project prerequisite opi checks: a declared packageManager version, engines.node and rust-toolchain.toml were measured never to disagree with what ran, since pnpm and rustup switch versions themselves.

In a workspace the checks run per member, in the member’s directory, because that is where the tools and their config live.

Reading the result

A failing check’s output is relayed whole and capped at twenty lines, with the command to see the rest. A tool that will not start is reported apart from one that found something, since a broken install needs a different remedy.

There is no health score. A composite number stops meaning anything within weeks; what a failing tool actually said does not.

opi --health exits non-zero when a check fails.