lint-rule-development

name	lint-rule-development
description	Step-by-step guide for creating and implementing lint rules in Biome's analyzer. Use when implementing rules like noVar, useConst, or any custom lint/assist rule, adding code actions to fix diagnostics, implementing semantic analysis for binding references, or adding configurable options to rules.
compatibility	Designed for coding agents working on the Biome codebase (github.com/biomejs/biome).

Purpose

Use this skill when creating new lint rules or assist actions for Biome. It provides scaffolding commands, implementation patterns, testing workflows, and documentation guidelines.

Prerequisites

1. Install required tools: just install-tools
2. Ensure cargo, just, and pnpm are available
3. Read crates/biome_analyze/CONTRIBUTING.md for in-depth concepts

Common Workflows

Create a New Lint Rule

Generate scaffolding for a JavaScript lint rule:

just new-js-lintrule useMyRuleName

For other languages:

just new-css-lintrule myRuleName
just new-json-lintrule myRuleName
just new-graphql-lintrule myRuleName

This creates a file in crates/biome_<language>_analyze/src/lint/nursery/use_my_rule_name.rs

All new lint rules must be placed in the nursery group, and require a patch changeset. Use the changeset skill to learn more about writing good changesets.

Implement the Rule

Basic rule structure (generated by scaffolding):

use biome_analyze::{context::RuleContext, declare_lint_rule, Rule, RuleDiagnostic};
use biome_js_syntax::JsIdentifierBinding;
use biome_rowan::AstNode;

declare_lint_rule! {
    /// Disallows the use of prohibited identifiers.
    pub UseMyRuleName {
        version: "next",
        name: "useMyRuleName",
        language: "js",
        recommended: false,
    }
}

impl Rule for UseMyRuleName {
    type Query = Ast<JsIdentifierBinding>;
    type State = ();
    type Signals = Option<Self::State>;
    type Options = ();

    fn run(ctx: &RuleContext<Self>) -> Self::Signals {
        let binding = ctx.query();

        // Check if identifier matches your rule logic
        if binding.name_token().ok()?.text() == "prohibited_name" {
            return Some(());
        }

        None
    }

    fn diagnostic(ctx: &RuleContext<Self>, _state: &Self::State) -> Option<RuleDiagnostic> {
        let node = ctx.query();
        Some(
            RuleDiagnostic::new(
                rule_category!(),
                node.range(),
                markup! {
                    "Avoid using this identifier."
                },
            )
            .note(markup! {
                "This identifier is prohibited because..."
            }),
        )
    }
}

Note: It's critically important to follow the guidelines in the High Quality Diagnostics section below when writing diagnostics.

Using Semantic Model

For rules that need binding analysis:

use crate::services::semantic::Semantic;

impl Rule for MySemanticRule {
    type Query = Semantic<JsReferenceIdentifier>;

    fn run(ctx: &RuleContext<Self>) -> Self::Signals {
        let node = ctx.query();
        let model = ctx.model();

        // Check if binding is declared
        let binding = node.binding(model)?;

        // Get all references to this binding
        let all_refs = binding.all_references(model);

        // Get only read references
        let read_refs = binding.all_reads(model);

        // Get only write references
        let write_refs = binding.all_writes(model);

        Some(())
    }
}

Add Code Actions (Fixes)

To provide automatic fixes:

use biome_analyze::FixKind;

declare_lint_rule! {
    pub UseMyRuleName {
        version: "next",
        name: "useMyRuleName",
        language: "js",
        recommended: false,
        fix_kind: FixKind::Safe, // or FixKind::Unsafe
    }
}

impl Rule for UseMyRuleName {
    fn action(ctx: &RuleContext<Self>, _state: &Self::State) -> Option<JsRuleAction> {
        let node = ctx.query();
        let mut mutation = ctx.root().begin();

        // Example: Replace the node
        mutation.replace_node(
            node.clone(),
            make::js_identifier_binding(make::ident("replacement"))
        );

        Some(JsRuleAction::new(
            ctx.metadata().action_category(ctx.category(), ctx.group()),
            ctx.metadata().applicability(),
            markup! { "Use 'replacement' instead" }.to_owned(),
            mutation,
        ))
    }
}

Quick Testing

Use the quick test for rapid iteration:

// In crates/biome_js_analyze/tests/quick_test.rs
// Uncomment #[ignore] and modify:

const SOURCE: &str = r#"
const prohibited_name = 1;
"#;

let rule_filter = RuleFilter::Rule("nursery", "useMyRuleName");

Run the test:

cd crates/biome_js_analyze
cargo test quick_test -- --show-output

Create Snapshot Tests

Create test files in tests/specs/nursery/useMyRuleName/:

tests/specs/nursery/useMyRuleName/
├── invalid.js          # Code that triggers the rule
├── valid.js            # Code that doesn't trigger the rule
└── options.json        # Optional rule configuration

Every test file must start with a top-level comment declaring whether it expects diagnostics. The test runner enforces this — see the testing-codegen skill for full rules. The short version:

valid.js — comment is mandatory (test panics without it):

/* should not generate diagnostics */
const x = 1;
const y = 2;

invalid.js — comment is strongly recommended (also enforced when present):

/* should generate diagnostics */
const prohibited_name = 1;
const another_prohibited = 2;

Run snapshot tests:

just test-lintrule useMyRuleName

Review snapshots:

cargo insta accept # accept all snapshots
cargo insta reject # reject all snapshots

Generate Analyzer Code

During development, use the lightweight codegen commands:

just gen-rules          # Updates rule registrations in *_analyze crates
just gen-configuration  # Updates configuration schemas

These generate enough code to compile and test your rule without errors.

For full codegen (migrations, schema, bindings, formatting), run:

just gen-analyzer

Note: The CI autofix job runs gen-analyzer automatically when you open a PR, so running it locally is optional.

Format and Lint

Before committing:

just f  # Format code
just l  # Lint code

Adding Configurable Options

When a rule needs user-configurable behavior, add options via the biome_rule_options crate. For the full reference (merge strategies, design guidelines, common patterns), see references/OPTIONS.md.

Quick workflow:

Step 1. Define the options type in biome_rule_options/src/<snake_case_rule_name>.rs:

use biome_deserialize_macros::{Deserializable, Merge};
use serde::{Deserialize, Serialize};

#[derive(Debug, Default, Clone, Serialize, Deserialize, Deserializable, Merge)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "camelCase", deny_unknown_fields, default)]
pub struct UseMyRuleNameOptions {
    #[serde(skip_serializing_if = "Option::is_none")]
    pub behavior: Option<MyBehavior>,
}

Step 2. Wire it into the rule:

use biome_rule_options::use_my_rule_name::UseMyRuleNameOptions;

impl Rule for UseMyRuleName {
    type Options = UseMyRuleNameOptions;

    fn run(ctx: &RuleContext<Self>) -> Self::Signals {
        let options = ctx.options();
        let behavior = options.behavior.unwrap_or_default();
        // ...
    }
}

Step 3. Test with options.json in the test directory (see references/OPTIONS.md for examples).

Step 4. Run codegen: just gen-rules && just gen-configuration

Key rules:

• All fields must be Option<T> for config merging to work
• Use Box<[Box<str>]> instead of Vec<String> for collection fields
• Use #[derive(Merge)] for simple cases, implement Merge manually for collections
• Only add options when truly needed (conflicting community preferences, multiple valid interpretations)

Tips

• Rule naming: Use no* prefix for rules that forbid something (e.g., noVar), use* for rules that mandate something (e.g., useConst)
• Nursery group: All new rules start in the nursery group
• Semantic queries: Use Semantic<Node> query when you need binding/scope analysis
• Multiple signals: Return Vec<Self::State> or Box<[Self::State]> to emit multiple diagnostics
• Safe vs Unsafe fixes: Mark fixes as Unsafe if they could change program behavior
• Check for globals: Always verify if a variable is global before reporting it (use semantic model)
• Error recovery: When navigating CST, use .ok()? pattern to handle missing nodes gracefully
• Testing arrays: Use .jsonc files with arrays of code snippets for multiple test cases

Common Mistakes to Avoid

Generally, mistakes revolve around allocating unnecessary data during rule execution, which can lead to performance issues. Common examples include:

• Placing String or Box<str> in a Rule's State type. It's a strong indicator that you are allocating a string unnecessarily. If the string comes from a CST token, this usually can be avoided by using TokenText instead.
• Building strings or other data structures only used in the code action in run() instead of action(). run() should only decide whether to emit a diagnostic; action() should build the fix. This matters for performance because building the action can be expensive, and we should avoid doing it when no diagnostic is emitted.

Common Query Types

// Simple AST query
type Query = Ast<JsVariableDeclaration>;

// Semantic query (needs binding info)
type Query = Semantic<JsReferenceIdentifier>;

// Multiple node types (requires declare_node_union!)
declare_node_union! {
    pub AnyFunctionLike = AnyJsFunction | JsMethodObjectMember | JsMethodClassMember
}
type Query = Semantic<AnyFunctionLike>;

High Quality Diagnostics

VERY IMPORTANT: Rule diagnostics MUST convey these messages, in this order:

1. What the problem is
2. Why it's a problem (motivation to fix the issue)
3. How to fix it (actionable advice)

If the rule has an action() to fix the issue, the 3rd message should go in the action's message. If not, it should go in the diagnostic's advice.

Diagnostics must remain focused on the specific issue that the rule is flagging. Avoid including superfluous details that aren't directly relevant to the problem, as this can overwhelm users and obscure the main point. If a rule can flag multiple classes of the same category of issue, the diagnostic messages should be surgically customized to the specific issue being flagged, rather than using generic messages that apply to all cases. This ensures that users receive precise and relevant information about the problem and how to fix it.

Examples

Good:

1. "Foo is not allowed here."
2. "Foo harms readability because of X, Y, Z."
3. "Consider using Bar instead, which is more concise and easier to read."

1. "Unexpected for-in loop."
2. "For-in loops are confusing and easy to misuse."
3. "You likely want to use a regular loop, for-of loop or forEach instead."

Bad:

1. "Prefer let or const over var." // conflates the what and the how in one message,
2. "var is bad." // not meaningful motivation to fix, doesn't explain the consequences
// third message missing is bad, because it doesn't give users a clear path to fix the issue

1. "This var declaration is not at the top of its containing scope." // Good start, explains what the problem is
2. "Move standalone var declarations before other statements in the same function, script, module, or static block." // Doesn't explain why, only tells the action. The "why" must come second, after the what.
3. "At module scope, imports and leading "<Emphasis>"export var"</Emphasis>" declarations may appear before other statements." // Doesn't explain the action, just gives a superfluous detail about module scope.

Tips

• New rules are always in the nursery group. No need to move them to another category.
• Changesets are always required for new rules. New rules are patch level changes. There's a skill to help write good changesets.

References

• Full guide: crates/biome_analyze/CONTRIBUTING.md
• Rule examples: crates/biome_js_analyze/src/lint/
• Semantic model: Search for Semantic< in existing rules
• Testing guide: Main CONTRIBUTING.md testing section