v3.0.1: Industrial Hardening, AST Parser, Self-Healing Loop & Tiered Verification
🛡️ Vibe UI Suite v3.0.1 — Industrial Hardening & Senior Audit Remediation
This patch release addresses the comprehensive architectural audit conducted by senior engineering reviewers, replacing heuristic assumptions with deterministic AST parsers, adding a Self-Healing agent loop, and establishing transparent benchmark methodologies.
🔧 Key Engineering Hardening & Fixes
-
AST-Based HTML Parsing Engine (
vibe_core.critic&vibe_core.verifier):- Replaced fragile regular expressions with
BeautifulSoup4DOM parsing for clickable element detection (div[onclick]), inline SVG counting, and CSS style extraction. - Scoped blur budget detection strictly to
<style>tags and inline style attributes, eliminating class-name false positives. - Maintained zero-dependency graceful regex fallbacks for restricted environments.
- Replaced fragile regular expressions with
-
Self-Healing Agent Loop (
vibe_core.healer):- Implemented
SelfHealingLoopbridging independent Critic defect evaluations directly into structured, machine-consumable[VIBE-UI CORRECTION REQUEST]blocks. - Integrated into
vibe generateand added the standalonevibe heal <file>CLI command.
- Implemented
-
Tiered Verification Architecture (Fast vs. Strict):
- Fast-Path (Default, <50ms): Pure static DOM evaluation with AST and WCAG compliance checks—instant feedback during iterative agent loops.
- Strict-Path (Opt-in via
--strict, 3-8s): Headless Playwright Chromium execution asserting multi-viewport (320pxand375px) zero-overflow and physical pixel rendering.
-
Centralized Architectural Constants (
vibe_core.constants):- Single Source of Truth for
MAX_BLUR_SURFACES = 3,HARD_MIN_TOUCH_PX = 24,RECOMMENDED_TOUCH_PX = 44, and loop thresholds, eliminating cross-module policy drift.
- Single Source of Truth for
-
Dynamic Real-Time UTC Timestamps:
- Replaced static timestamps across all reports (
critic.py,verifier.py,run_benchmark.py) with dynamicdatetime.now(timezone.utc).isoformat().
- Replaced static timestamps across all reports (
-
Transparent Benchmark Methodology & Schema Contracts:
- Added
benchmark_type: "internal_deterministic_heuristic"toschemas/benchmark-result.v1.jsonandevals/benchmark/benchmark_results.json. - Added explicit methodology disclosure clarifying that benchmark metrics measure automated rule-engine compliance rather than subjective human A/B testing.
- Added
-
Full Pipeline End-to-End Integration Tests (
scripts/test_pipeline_e2e.py):- Added 4-scenario E2E test verifying complete multi-agent pipeline from user prompt to physical proof across Persian RTL, English SaaS, defect recovery, and constant integrity.
- Expanded master test suite (
scripts/run_all_tests.py) to 8 validation gates (100% PASS in <3000ms).
-
Comprehensive Repository
.gitignoreProtection:- Hardened ignores for Python virtual environments (
.venv/,venv/), test caches (.pytest_cache/,.mypy_cache/), coverage files, environment secrets (.env*), and IDE workspaces.
- Hardened ignores for Python virtual environments (
📦 Release Assets
vibe-ui-vscode-3.0.1.vsix— VS Code, Cursor & Windsurf extension binary.- Scoped NPM Package:
@omid-io/tokens@3.0.1