language-idioms-refiner

Language Idioms Refiner

What This Produces

• Output: .lattice/standards/language-idioms.md (or custom path from .lattice/config.yaml -> paths.language_idioms)
• Mode: Always standalone -- no embedded language defaults exist in atoms to overlay on. For revisions, load existing document and update sections in place.
• Config keys:
  • language (top-level) -- language identifier (e.g., go, rust, python, java, typescript)
  • paths.language_idioms -- path to the produced document
• Template: Read ./assets/template.md for the full document structure, pre-populated examples, and interview guidance comments
• Consumed by: Multiple atoms reference specific sections by heading name:

Section	Consumed by
Error Handling	clean-code atom (§8), secure-coding atom (§1 trust boundary messages)
Type System & Object Model	clean-code atom (§1 SRP/cohesion), domain-driven-design atom (entities, VOs, aggregates)
Naming Conventions	clean-code atom (§4)
Testing Patterns	test-quality atom (§5 naming, §4 isolation, §6 builders)
Parameter & Function Design	clean-code atom (§2 function size, §5 parameters)
Dependency Management	clean-code atom (§9 testability/DI), architecture atom (dependency direction)

These six section headings are the stable contract. Atoms reference them by name. Additional sections can be added by consumers but these six must be present.

Scope Clarification

This document captures how the project's language expresses engineering patterns -- the language-level idioms that atoms need to adapt their pseudocode defaults. Clear boundaries:

Concern	Where It Belongs	Not Here
Project identity, tech stack, directory layout	knowledge-priming atom	No project structure or framework docs
Code craftsmanship rules (thresholds, heuristics)	clean-code atom / overlay	No function size limits or DRY rules
Architecture layers, dependency direction	architecture atom / overlay	No layer definitions
Domain modeling guardrails	domain-driven-design atom / overlay	No aggregate rules
Team-specific preferences within language	Atom-specific overlays	No team decisions (see below)

Key distinction from atom overlays: This document describes how the language works. Atom overlays describe how the team works within the language.

• Language idioms doc: "Go uses explicit error returns (if err != nil), not exceptions"
• Clean-code overlay: "We use fmt.Errorf('context: %w', err) for wrapping, custom error types for domain errors"

Language idioms are facts about the language. Atom overlays are team choices.

Before You Begin

Check for existing documents

1. Read .lattice/config.yaml -- does paths.language_idioms point to a file?
2. If yes, read that file. Ask the user:
   • "You already have a language idioms document for [language]. Would you like to revise it (update specific sections), start fresh (new interview), or switch language?"
   • Revise: Load the existing document, walk through only the sections the user wants to change.
   • Start fresh: Proceed with the full interview flow below.
   • Switch language: Proceed with full interview for the new language; existing document will be replaced.
3. If no config or no existing document, proceed with the full interview flow.

Detect the language

Determine the project language before starting the interview:

1. From config: Check .lattice/config.yaml for language key.
2. From project files (if no config key):
   • package.json → TypeScript / JavaScript
   • tsconfig.json → TypeScript (confirm over JavaScript)
   • go.mod → Go
   • pom.xml or build.gradle or build.gradle.kts → Java or Kotlin
   • Cargo.toml → Rust
   • requirements.txt or pyproject.toml or setup.py → Python
   • Gemfile → Ruby
   • *.csproj or *.sln → C# / .NET
   • Package.swift → Swift
3. Multiple languages detected: Ask the user which is the primary language. One language-idioms document per project (covers the primary language).
4. No detection: Ask the user directly.

Present the detected language: "I detected this is a Go project (found go.mod). I'll propose Go-idiomatic patterns for each section. You can confirm or adjust."

Interview Flow

This refiner works differently from other refiners. Instead of showing defaults and asking "change or keep?", it proposes language-specific content and asks "does this match your team's usage?"

Step 1: Confirm language and version

"I detected [Language] [version]. Is this correct?"

Record language and version. These go in the document frontmatter.

Step 2: Walk through sections with proposals

For each of the 6 sections:

1. Propose pre-populated content based on the detected language (see Language-Specific Proposals below).
2. Present the proposal: "Here's what I'd recommend for [Language]. Does this match how your team uses [Language]?"
3. User confirms → record as-is.
4. User adjusts → discuss specifics, record their version.

Step 3: Additional sections

After the 6 core sections, ask: "Any language-specific patterns I should add? For example: concurrency patterns, memory management, async/await idioms, or framework-specific conventions."

Record any additional sections the user wants.

Step 4: Produce document

Assemble and write the document.

Section-by-Section Interview Guide

Read ./assets/template.md and follow the <!-- INTERVIEW GUIDANCE: --> comments for each section.

The 6 core sections

#	Section	What It Captures
1	Error Handling	Language error philosophy (exceptions, error returns, Result types), error propagation patterns, error creation idioms
2	Type System & Object Model	Classes vs structs, interfaces (nominal vs structural), inheritance vs composition, generics, type safety idioms
3	Naming Conventions	Case conventions, visibility modifiers, acronym style, package/module naming, idiomatic patterns
4	Testing Patterns	Test framework idioms, test organization, assertion patterns, mocking approach, test naming style
5	Parameter & Function Design	Argument passing idioms, options/config patterns, multiple returns, named parameters, function signatures
6	Dependency Management	DI approach (container vs manual), interface placement, wiring patterns, import/module conventions

Cross-section awareness

Decision in	Affects	How
§1 Error Handling	§4 Testing	Error patterns determine how error paths are tested
§2 Type System	§5 Parameters, §6 Dependencies	Object model shapes function signatures and DI approach
§3 Naming	§4 Testing	Naming conventions apply to test names too

Language-Specific Proposals

For well-known languages, pre-populate each section with idiomatic defaults. The interview confirms or adjusts these. For unrecognized languages, ask open-ended questions.

Go

Section	Proposal summary
Error Handling	Explicit error returns (value, err := ...), if err != nil, error wrapping with fmt.Errorf("context: %w", err), sentinel errors for expected cases, no exceptions
Type System	Structs with methods (receiver functions), implicit interfaces (structural typing), composition via embedding, no inheritance, no classes
Naming	Exported = capitalized, unexported = lowercase, short names in small scopes, acronyms fully uppercase (HTTP, ID), package name is part of identifier (http.Client not http.HTTPClient)
Testing	Table-driven tests, t.Run for subtests, testing.T parameter, test files _test.go co-located, no assertion library required (stdlib comparisons)
Parameters	Accept interfaces return structs, functional options pattern for config (WithTimeout(5*time.Second)), multiple return values, no method overloading
Dependencies	Pass interface parameters (not constructor DI), define interfaces at consumer not provider, no DI container, explicit wiring in main() or cmd/

Rust

Section	Proposal summary
Error Handling	Result<T, E> for recoverable, panic! for unrecoverable, ? operator for propagation, thiserror for library errors, anyhow for application errors
Type System	Structs + impl blocks, traits (explicit implementation), enums with data (algebraic types), ownership/borrowing, no inheritance, no null (Option<T> instead)
Naming	snake_case functions/variables, PascalCase types/traits, SCREAMING_SNAKE constants, lifetime names short ('a, 'b)
Testing	#[test] attribute, #[cfg(test)] mod tests in same file, integration tests in tests/ directory, assert_eq!/assert! macros
Parameters	Ownership: borrow (&T) vs move, generic bounds (impl Trait), builder pattern for complex config, no default parameters
Dependencies	Trait objects (dyn Trait) or generics (impl Trait) for abstraction, no DI container, explicit construction

Python

Section	Proposal summary
Error Handling	EAFP over LBYL (try/except, not if-checks), context managers (with) for cleanup, custom exceptions inheriting from base classes, raise/except
Type System	Classes, dataclasses (@dataclass), protocols for structural typing (PEP 544), duck typing, type hints encouraged but optional at runtime
Naming	snake_case functions/variables, PascalCase classes, SCREAMING_SNAKE constants, _private convention (single underscore), __dunder__ for magic methods
Testing	pytest preferred, fixtures for setup/teardown, @pytest.mark.parametrize for data-driven tests, plain assert (pytest rewrites), test files test_*.py
Parameters	**kwargs for options, named/keyword arguments, default values, dataclass or TypedDict for config objects
Dependencies	Constructor injection with protocols/ABCs, or function parameters, no heavyweight DI container (or dependency-injector if needed)

Java / Kotlin

Section	Proposal summary
Error Handling	Java: unchecked exceptions preferred over checked (modern style), custom exceptions extend RuntimeException. Kotlin: sealed class Result pattern, runCatching, no checked exceptions
Type System	Java: classes, interfaces, records (16+), sealed classes (17+). Kotlin: data classes, sealed hierarchies, null safety (?), extension functions
Naming	camelCase variables/methods, PascalCase classes/interfaces, SCREAMING_SNAKE constants, packages lowercase.dotted
Testing	JUnit 5, @Test, @ParameterizedTest, @Nested for grouping, Mockito (Java) / MockK (Kotlin), AssertJ for fluent assertions
Parameters	Java: builder pattern for >3 params, method overloading. Kotlin: named arguments, default values, data class config
Dependencies	Constructor injection (Spring, Guice, or manual), DI containers are idiomatic, program to interfaces

TypeScript

Section	Proposal summary
Error Handling	try/catch with custom Error subclasses, typed error handling optional (Result pattern via libraries), no checked exceptions
Type System	Interfaces, type aliases, union/intersection types, generics, unknown over any, discriminated unions for state
Naming	camelCase variables/functions, PascalCase types/classes/enums, SCREAMING_SNAKE constants, no I prefix for interfaces
Testing	Jest or Vitest, describe/it blocks, mock functions (jest.fn()), expect().toBe() assertions, .test.ts or .spec.ts co-located
Parameters	Options objects with destructuring, default values, rest parameters, overloaded signatures for type narrowing
Dependencies	Constructor injection, DI optional (tsyringe, inversify), or module-level factory functions

C# / .NET

Section	Proposal summary
Error Handling	Exceptions for exceptional cases, custom exceptions from Exception base, try/catch/finally, no error codes for business logic, Result pattern growing in popularity
Type System	Classes, interfaces (explicit), records (C# 9+), structs (value types), nullable reference types (C# 8+), generics
Naming	PascalCase methods/properties/classes, camelCase local variables/parameters, _camelCase private fields, I prefix for interfaces (IService)
Testing	xUnit or NUnit, [Fact]/[Theory] (xUnit), [Test]/[TestCase] (NUnit), FluentAssertions, Moq for mocking
Parameters	Named parameters, optional parameters with defaults, builder pattern or options pattern (IOptions<T>) for config
Dependencies	Constructor injection via built-in DI (IServiceCollection), DI containers idiomatic, interface-first

Other Languages

For languages not listed above, use open-ended questions for each section:

1. "How does [Language] handle errors? Exceptions, error returns, algebraic types, or something else?"
2. "What's the object model? Classes, structs, traits, protocols? Inheritance or composition?"
3. "What are the naming conventions? Case style, visibility markers, module naming?"
4. "What's the testing ecosystem? Framework, organization, assertion style?"
5. "How do you pass configuration and options to functions? Named params, objects, builders?"
6. "How is dependency injection handled? Containers, manual wiring, interface parameters?"

Output Assembly

1. YAML frontmatter: language and version
2. Title line: # Language Idioms: {Language}
3. All 6 core sections with confirmed/adjusted content
4. Any additional sections the user added
5. Strip all <!-- INTERVIEW GUIDANCE: --> comments

Target size: 40-60 lines of focused content. Each section should be 4-8 lines: a brief philosophy statement plus the key idiomatic patterns as a concise list. No code examples -- atoms have their own examples in pseudocode that they adapt using this document's guidance.

Determine output path:

1. If .lattice/config.yaml exists and has paths.language_idioms, use that path.
2. Otherwise, default to .lattice/standards/language-idioms.md.

Update config:

1. Set language: {language} at top level (create or update).
2. Set paths.language_idioms pointing to the output file.
3. If .lattice/config.yaml does not exist, create it. Preserve all existing content.

Confirm to user: "Your language idioms document has been written to [PATH] for [Language] [version]. The following atoms will now adapt their patterns: clean-code, test-quality, secure-coding, domain-driven-design, and architecture."

Document Quality Checks

Before writing the final document, verify:

• All 6 core section headings are present and match exactly: Error Handling, Type System & Object Model, Naming Conventions, Testing Patterns, Parameter & Function Design, Dependency Management
• Content is specific to the language, not generic advice ("use error returns" not "handle errors properly")
• No code craftsmanship rules (thresholds, complexity limits) -- those belong in atom overlays
• No project identity information (tech stack, directory layout) -- that belongs in knowledge-priming
• Content is concise -- each section 4-8 lines, total document under 60 lines
• Frontmatter has correct language and version values
• Config file is correctly updated with both language key and paths.language_idioms
• No <!-- INTERVIEW GUIDANCE: --> comments remain in output