Skip to content

CLAUDE.md

Latest commit

 

History

History
47 lines (35 loc) · 1.97 KB

CLAUDE.md

File metadata and controls

47 lines (35 loc) · 1.97 KB

DojoWatch

AI-native visual regression testing engine — Claude Code plugin + GitHub Actions CI.

Architecture

  • Plugin-first: Slash commands are the primary local interface. Claude Code IS the local AI engine.
  • Scripts in scripts/: Standalone TypeScript, runnable via npx tsx. Used by both slash commands and CI.
  • Gemini for CI: scripts/analyze-gemini.ts handles batch analysis in GitHub Actions.
  • No npm publish: This is a Claude Code plugin, not a published package.

Conventions

  • TypeScript strict mode, ESM only ("type": "module")
  • All shared types in scripts/types.ts
  • Pre-filter has a zero false-negative guarantee: only byte-identical screenshots skip analysis
  • Scripts must be runnable standalone (for CI) AND callable from slash commands via Bash
  • Use picocolors for terminal output, never chalk
  • Use execFileSync not execSync — avoid shell injection
  • Test fixtures in tests/fixtures/

File layout

Directory Purpose
commands/ Claude Code slash commands (markdown with YAML frontmatter)
agents/ Claude Code agents (markdown with YAML frontmatter)
skills/ Claude Code skills (SKILL.md + references/)
scripts/ Core TypeScript scripts (capture, prefilter, analyze, etc.)
templates/ User-facing templates (config example, GitHub Actions)
tests/ Vitest tests + PNG fixtures

Script naming

Scripts double as both importable modules and CLI entrypoints:

  • Export functions for programmatic use
  • Include if (isDirectRun) block for CLI execution
  • Accept args via process.argv when run directly

Classification schema

Visual diffs are classified as:

  • REGRESSION: Unintended visual change (bug). Severity: high/medium/low.
  • INTENTIONAL: Deliberate change (feature, design update).
  • NOISE: Insignificant rendering variance (anti-aliasing, sub-pixel).

See skills/visual-regression/references/classification-schema.md for full criteria.