--- title: Linting documents description: Choosing input, selecting rules, and picking an output format. icon: file-search --- ## Input ```sh lawlint contract.docx # a file lawlint - # stdin, explicitly cat memo.txt | lawlint # stdin, implicitly ``` Format is inferred from the extension. For piped Markdown there is no extension to infer from, so pass `--markdown` — see [File formats](/docs/concepts/file-formats). ## Choosing rules **Only these** ```sh lawlint --rules no-em-dash,no-hedging draft.md ``` Runs exactly the listed rules and nothing else. Useful when you want a single question answered rather than a full review. **All but these** ```sh lawlint --disable no-legalese,no-semicolons contract.docx ``` Runs the full set minus the listed rules. This is the right shape for a house style: most of the defaults, minus the handful that fight your conventions. **Durably** ```json // .lawlint/config.json { "disable": ["no-legalese", "no-semicolons"], "severity": { "no-em-dash": "suggestion" } } ``` Committed to the repo, so everyone on the project lints the same way. See [Configuration](/docs/reference/configuration). Both flags accept full rule ids or bare aliases, and both override the corresponding config file setting. ## Output formats **pretty** ```sh lawlint draft.md ``` The default. Human-readable findings with the source line, the span underlined, and a suggestion where one exists — followed by the [human-likeness score](/docs/concepts/scoring). Add `--quiet` to suppress it entirely while keeping the exit code, which is what you usually want in CI. **json** ```sh lawlint draft.md --format json ``` The full result as JSON: every diagnostic with its rule id, severity, tier, intent, span, line and column, excerpt, suggestion, and fix — plus document stats. See the [JSON output reference](/docs/reference/json-output). **prompt** ```sh lawlint draft.md --format prompt ``` An AI revision brief: the flagged text plus instructions, formatted to hand straight to a writing assistant. This is the reverse of `--judge` — instead of asking a model to *find* problems, it asks one to *fix* the problems lawlint already found deterministically. ## Extra rule packages ```sh lawlint --rule-dir ./house-style draft.md lawlint --rule-dir ./house-style --rule-dir ./team-rules draft.md ``` Repeatable, and merged over the built-ins — a package rule with the same id as a built-in replaces it. The durable equivalent is `ruleDirs` in `.lawlint/config.json`. :::note `--rule-dir` paths resolve against the current directory. `ruleDirs` in a config file resolve against that config file's directory, so a committed config keeps working no matter where in the project you run from. ::: See [Authoring rules](/docs/guides/authoring-rules) for what goes in one. ## Limiting noise ```sh lawlint --max-warnings 10 draft.md ``` Exits `1` when warnings exceed the limit. The default is `inf` — warnings never fail on their own, only error-severity findings do. ## Next Apply what is safe to apply. Exit codes, quiet output, and gating a pull request.