Conventions

Mindr detects naming and file conventions by analysing your source code with Tree-sitter ASTs. Convention profiles are stored as memories and injected into agent context so agents write code that matches your codebase style automatically.

What is detected

For each supported language, Mindr analyses:

Category	Examples
Function names	camelCase, snake_case, PascalCase
Variable names	camelCase, snake_case, SCREAMING_SNAKE_CASE
Class names	PascalCase, camelCase
File names	kebab-case, snake_case, camelCase, PascalCase
Test file patterns	.test.ts, _test.go, test_*.py, .spec.ts
Import grouping	Grouped (stdlib / external / local) vs. mixed
Error handling	Typed catch vs. generic catch(e), specific vs. bare except

Each pattern gets a consistency score from 0–100%, based on how many observed samples match it.

Supported languages

• TypeScript
• JavaScript
• Python
• Go
• Rust

Support for additional languages can be added by contributing a Tree-sitter grammar entry to packages/core/src/conventions/languages.ts.

When conventions are updated

• On mindragent init — an initial scan runs across all files in the project
• On every commit — files touched in the commit are re-scanned incrementally; the stored convention memory for that language is updated

Viewing conventions

# Via CLI (shown in mindragent status and mindragent generate agents-md)
mindragent generate agents-md

const profiles = await mindr.getConventions({ language: 'typescript' });
for (const p of profiles) {
  for (const c of p.conventions) {
    console.log(`${c.category}: ${c.pattern} (${c.score}%)`);
  }
}

Sample output:

functionNames: camelCase (97%)
classNames: PascalCase (100%)
fileNames: kebab-case (89%)
testPattern: .test.ts (100%)
importGrouping: grouped (72%)
errorHandling: typedCatch (85%)

How consistency score is computed

Mindr counts the number of sampled identifiers that match each pattern style. The score is:

score = (matching_samples / total_samples) × 100

A score of 60% for camelCase functions means 60% of observed function names use camelCase. The convention is real but not enforced — Mindr reports what it observes, not what it mandates.

In session context

Convention profiles appear in the Conventions section of mindr:get_context, showing the top 3 conventions per language by consistency score:

## Conventions (typescript)
- camelCase functions: 97%
- PascalCase classes: 100%
- kebab-case files: 89%

Agents use this to write new code that matches the existing style without needing to inspect the codebase manually.

ConventionProfile type

interface ConventionProfile {
  language: string
  analyzedFiles: number
  analyzedAt: string          // ISO timestamp
  conventions: ConventionEntry[]
}

interface ConventionEntry {
  pattern: string             // e.g. "camelCase"
  category: string            // e.g. "functionNames"
  score: number               // 0–100
  sampleCount: number         // number of observed samples
}