Biome Developer

Purpose

This skill provides general development best practices, common gotchas, and Biome-specific patterns that apply across different areas of the codebase. Use this as a reference when you encounter unfamiliar APIs or need to avoid common mistakes.

Prerequisites

• Basic familiarity with Rust
• Understanding of Biome's architecture (parser, analyzer, formatter)
• Development environment set up (see CONTRIBUTING.md)

Common Gotchas and Best Practices

Working with AST and Syntax Nodes

DO:

• Use parser crate's quick_test to inspect AST structure before implementing
• Understand the node hierarchy and parent-child relationships
• Check both general cases AND specific types (e.g., Vue has both VueDirective and VueV*ShorthandDirective)
• Verify your solution works for all relevant variant types, not just the first one you find

DON'T:

• Do NOT build the full Biome binary just to inspect syntax (expensive) - use parser crate's quick_test instead
• Do NOT assume syntax patterns without inspecting the AST first

Example - Inspecting AST:

// In crates/biome_html_parser/tests/quick_test.rs
// Modify the quick_test function:
#[test]
pub fn quick_test() {
    let code = r#"<button on:click={handleClick}>Click</button>"#;
    let source_type = HtmlFileSource::svelte();
    let options = HtmlParserOptions::from(&source_type);
    let root = parse_html(code, options);
    dbg!(&root.syntax());  // Shows full AST structure
}

Run: just qt biome_html_parser

String Extraction and Text Handling

DO:

• Use inner_string_text() when extracting content from quoted strings (removes quotes)
• Use text_trimmed() when you need the full token text without leading/trailing whitespace
• Use token_text_trimmed() on nodes like HtmlAttributeName to get the text content
• Verify whether values use HtmlString (quotes) or HtmlTextExpression (curly braces)

DON'T:

• Do NOT use text_trimmed() when you need inner_string_text() for extracting quoted string contents

Example - String Extraction:

// WRONG: text_trimmed() includes quotes
let html_string = value.as_html_string()?;
let content = html_string.value_token()?.text_trimmed(); // Returns: "\"handler\""

// CORRECT: inner_string_text() removes quotes
let html_string = value.as_html_string()?;
let inner_text = html_string.inner_string_text().ok()?;
let content = inner_text.text(); // Returns: "handler"

Working with Embedded Languages

DO:

• Verify changes work for different value formats (quoted strings vs text expressions) when handling multiple frameworks
• Use appropriate EmbeddingKind for context (Vue, Svelte, Astro, etc.)
• Check if embedded content needs is_source: true (script tags) vs is_source: false (template expressions)
• Calculate offsets correctly: token start + 1 for opening quote, or use text_range().start() for text expressions

DON'T:

• Do NOT assume all frameworks use the same syntax (Vue uses quotes, Svelte uses curly braces)
• Do NOT implement features for "widely used" patterns without evidence - ask the user first

Example - Different Value Formats:

// Vue directives use quoted strings: @click="handler"
let html_string = value.as_html_string()?;
let inner_text = html_string.inner_string_text().ok()?;

// Svelte directives use text expressions: on:click={handler}
let text_expression = value.as_html_attribute_single_text_expression()?;
let expression = text_expression.expression().ok()?;

Borrow Checker and Temporary Values

DO:

• Use intermediate let bindings to avoid temporary value borrows that get dropped
• Store method results that return owned values before calling methods on them

DON'T:

• Do NOT create temporary value borrows that get dropped before use

Example - Avoiding Borrow Issues:

// WRONG: Temporary borrow gets dropped
let html_string = value.value().ok()?.as_html_string()?;
let token = html_string.value_token().ok()?; // ERROR: html_string dropped

// CORRECT: Store intermediate result
let value_node = value.value().ok()?;
let html_string = value_node.as_html_string()?;
let token = html_string.value_token().ok()?; // OK

Clippy and Code Style

DO:

• Use let chains to collapse nested if let statements (cleaner and follows Rust idioms)
• Run just l before committing to catch clippy warnings
• Fix clippy suggestions unless there's a good reason not to

DON'T:

• Do NOT ignore clippy warnings - they often catch real issues or suggest better patterns

Example - Collapsible If:

// WRONG: Nested if let (clippy::collapsible_if warning)
if let Some(directive) = VueDirective::cast_ref(&element) {
    if let Some(initializer) = directive.initializer() {
        // ... do something
    }
}

// CORRECT: Use let chains
if let Some(directive) = VueDirective::cast_ref(&element)
    && let Some(initializer) = directive.initializer()
{
    // ... do something
}

Code Comments

Comments exist for the next developer who reads this code, not for the developer currently writing it.

DO:

• Explain code that is hard to read, or document exceptions and edge cases
• Provide context when names alone are not descriptive enough
• Describe the business logic a function implements
• Clarify contextual words like "normalize" — e.g., "normalize a file path" and "normalize a URL" mean different things; spell out what normalization means here

DON'T:

• Do NOT embed the context of the current work into comments. A comment like // As per issue #1234, we skip this case ties the code to a transient artifact. Instead, explain why the case is skipped in terms any future reader would understand.
• Do NOT scope comments to the specific trigger that prompted the change. For example, if a bug was reported for Astro but the fix applies broadly, do NOT write // Fix for Astro embedding. Write a comment that describes the general condition being handled.

Think big picture, not current task. Before writing a comment, ask: "If someone reads this a year from now with no knowledge of the issue or PR, does this comment give them the context they need?"

Example:

// WRONG: Carries issue/task context
// Fix for #5678: Astro files need special handling here
if is_embedded_script(node) {
    return normalize_offset(node);
}

// WRONG: Describes what the code does (the code already says that)
// Check if the node is an embedded script and normalize the offset
if is_embedded_script(node) {
    return normalize_offset(node);
}

// CORRECT: Explains why and clarifies "normalize"
// Embedded script blocks (e.g. <script> inside .vue/.svelte/.astro files)
// report offsets relative to the embedding document, not the script itself.
// Normalize here means: subtract the script block's start position so the
// offset is relative to the script content.
if is_embedded_script(node) {
    return normalize_offset(node);
}

Cargo Dependencies: workspace = true vs path = "..."

Internal biome_* crates listed under [dev-dependencies] MUST use path = "../<crate_name>", not workspace = true. Using workspace = true for dev-dependencies can cause Cargo to resolve the crate from the registry instead of the local workspace, which is incorrect.

Regular [dependencies] still use workspace = true as normal — this rule only applies to [dev-dependencies].

DO:

• Use path = "../biome_foo" for all biome_* dev-dependencies
• Preserve any extra attributes like features when converting

DON'T:

• Do NOT use workspace = true for biome_* crates in [dev-dependencies]

Example:

# WRONG: may resolve from registry
[dev-dependencies]
biome_js_parser = { workspace = true }
biome_formatter = { workspace = true, features = ["countme"] }

# CORRECT: always resolves locally
[dev-dependencies]
biome_js_parser = { path = "../biome_js_parser" }
biome_formatter = { path = "../biome_formatter", features = ["countme"] }

All crates live as siblings under crates/, so the relative path is always ../biome_<name>.

Legacy and Deprecated Syntax

DO:

• Ask users before implementing deprecated/legacy syntax support
• Wait for user demand before spending time on legacy features
• Document when features are intentionally not supported due to being legacy

DON'T:

• Do NOT implement legacy/deprecated syntax without checking with the user first
• Do NOT claim patterns are "widely used" or "common" without evidence

Example: Svelte's on:click event handler syntax is legacy (Svelte 3/4). Modern Svelte 5 runes mode uses regular attributes. Unless users specifically request it, don't implement legacy syntax support.

Testing and Development

For testing commands, snapshot workflows, and code generation, see the testing-codegen skill. Key reminders specific to Biome development patterns:

• Test with multiple variants when working with enums (e.g., all VueV*ShorthandDirective types)
• Use CLI tests for testing embedded languages (Vue/Svelte directives, etc.)
• Do NOT try to test embedded languages in analyzer packages (they don't have embedding capabilities)

Pattern Matching Tips

Working with Node Variants

When working with enum variants (like AnySvelteDirective), check if there are also non-enum types that need handling:

// Check AnySvelteDirective enum (bind:, class:, style:, etc.)
if let Some(directive) = AnySvelteDirective::cast_ref(&element) {
    // Handle special Svelte directives
}

// But also check regular HTML attributes with specific prefixes
if let Some(attribute) = HtmlAttribute::cast_ref(&element) {
    if let Ok(name) = attribute.name() {
        // Some directives might be parsed as regular attributes
    }
}

Checking Multiple Variant Types

For frameworks with multiple directive syntaxes, handle each type:

// Vue has multiple shorthand types
if let Some(directive) = VueVOnShorthandDirective::cast_ref(&element) {
    // Handle @click
}
if let Some(directive) = VueVBindShorthandDirective::cast_ref(&element) {
    // Handle :prop
}
if let Some(directive) = VueVSlotShorthandDirective::cast_ref(&element) {
    // Handle #slot
}
if let Some(directive) = VueDirective::cast_ref(&element) {
    // Handle v-if, v-show, etc.
}

Common API Confusion

String/Text Methods

Value Extraction Methods

References

• Main contributing guide: ../../CONTRIBUTING.md
• Testing workflows: ../testing-codegen/SKILL.md
• Parser development: ../parser-development/SKILL.md
• Biome internals docs: https://biomejs.dev/internals

Documentation and Markdown Formatting

DO:

• Use spaces around table separators: | --- | --- | --- | (not |---|---|---|)
• Ensure all Markdown tables follow "compact" style with proper spacing
• Test documentation changes with markdown linters before committing

DON'T:

• Do NOT use compact table separators without spaces (causes CI linting failures)

Example - Table Formatting:

<!-- WRONG: No spaces around separators -->
| Method | Use When | Returns |
|--------|----------|---------|

<!-- CORRECT: Spaces around separators -->
| Method | Use When | Returns |
| --- | --- | --- |

The CI uses markdownlint-cli2 which enforces the "compact" style requiring spaces.

When to Use This Skill

Load this skill when:

• Working with unfamiliar Biome APIs
• Getting borrow checker errors with temporary values
• Extracting strings or text from syntax nodes
• Implementing support for embedded languages (Vue, Svelte, etc.)
• Wondering why your AST inspection doesn't match expectations
• Making decisions about legacy/deprecated syntax support
• Writing or updating markdown documentation