Following Keep a Changelog and semantic versioning. Every notable change, newest first.
Unreleased
IN PROGRESS
on main
0.2.3 (RC): Fail-closed pipeline & production hardening
Release candidate 0.2.3 hardens the comparison pipeline to fail closed, enforces read-only baseline checks, keeps AI classifications advisory, and closes route-discovery, plugin, and Cloud API schema gaps.
·Deterministic AI pass/fail contract: AI classifications and fix suggestions are advisory annotations and never silently downgrade failing pixel diffs.
·Visual Cortex MCP tools & Cloud CAS deduplication: adds inspect_diff_cluster_dom, verify_css_patch, and promote_baseline to @frontguard/mcp with local report fallback, plus SHA-256 Git CAS manifest tracking and Cloud CAS endpoints.
·CLI agent guide, anti-flake controls & multi-provider AI: adds frontguard guide [topic] (--json), waitForStability multi-frame convergence, redactSensitive DOM/HTML input masking, freezeTime ISO / seedRandom / blockUrls anti-flake controls, pre-capture route actions with colorScheme emulation, and Google Gemini + custom baseUrl AI provider support.
SECURITY
·HTTP security headers & supply-chain updates: strict CSP/HSTS/frame-ancestors headers on frontguard.dev, monitor SSRF validation, and updated transitive dependencies.
0.2.2
LATEST RELEASE
Config loader & package metadata alignment
Preserves optional dependencies in the published CLI config loader and aligns README and release metadata across all five published npm packages.
CHANGED
·CLI config loader dependencies: keeps optional TypeScript/ESM config loader dependencies intact in the published @frontguard/cli tarball.
·Workspace package READMEs & version sync: ships workspace-local READMEs and closes release metadata drift across @frontguard/cli, @frontguard/playwright, @frontguard/mcp, create-frontguard-plugin, and @frontguard/netlify-plugin.
0.2.1
PATCH RELEASE
Production-close remediation
Republishes the five public npm packages with config loader, MCP stdio, Storybook 8.6, supply-chain, SSRF, and documentation fixes.
·MCP server & Storybook 8.6 readiness: realpath-safe MCP entry guard with clean stdio JSON-RPC output, and Storybook 8.6 finished-phase parameter discovery.
·Netlify plugin & canonical install docs: inspects real cloud-api status values and standardizes scoped @frontguard/cli commands across docs.
SECURITY
·Supply-chain & Slack SSRF guards: migrated to @daytona/sdk, cleared high/critical audit findings, and added SSRF validation on Slack slash-command URLs.
0.2.0
MINOR RELEASE
The "earn trust" release
The core engine is joined by an AI auto-fix moat, a cloud platform, production monitoring, and a full integration surface.
ADDED
·Storybook integration: a storybook config block enumerates stories from a running Storybook (8.x / 7.x) and renders each, with play()-aware capture and per-story overrides. frontguard init auto-detects .storybook.
·OpenTelemetry export: run completions emit OTLP/HTTP metrics (runs, comparisons, regressions, duration) to a configurable endpoint; runs on Cloudflare Workers.
·Native Slack app: a Hono handler with signing-secret verification, the /frontguard slash command, OAuth v2 install, and result posting.
·Run-over-run perf regressions: the perf-budgets plugin can persist each run’s metrics and flag any metric that degraded since the last run.
·Accessibility-aware AI: axe-core violations are fused into the AI prompt so the model can correlate a visual change with a known a11y issue.
·frontguard doctor: environment diagnostics for sources of non-determinism (Node, Playwright/Chromium, browsers, config, git state).
·frontguard monitor: live production URL monitoring with --once, --watch/--interval daemon polling, --history, and webhook alerts.
·AI fix generation + sandbox verification: generateFixes produces minimal CSS patches; verifyFixes applies them in a sandbox, re-renders, and re-compares against baseline.
·Fix-pattern database (reuse patterns accepted ≥3 times): a SQLite store with accept-fix / reject-fix / export-patterns; the pipeline reuses patterns accepted ≥3 times before calling the AI.
·Accessibility & performance plugins: axe-core WCAG audits and LCP/CLS/TTFB budgets, both correlated inline with the visual diff.
·Cloud platform (Hono, D1+R2+GitHub OAuth): a Hono service with run submission, baseline approval and usage metering; Cloudflare D1 + R2 + GitHub OAuth.
·Teams & billing (Stripe): multi-tenant teams with roles, invitations, approvals, an activity feed, and Stripe billing.
·Integrations (Vercel OAuth, Netlify Build Plugin, GitHub App): real Vercel OAuth app, Netlify Build Plugin, and a GitHub App with Check Runs.
CHANGED
·docs migrated VitePress → Fumadocs
·reporters render a11y/perf/3rd-party sections
0.1.0
INITIAL RELEASE
The core engine
CLI, route discovery, multi-browser capture, pixel diffing, AI analysis, Git baselines and the plugin architecture: the foundation.
ADDED
·CLI: frontguard init and run with proper flow control and exit codes.
·route discovery: auto-crawl plus filesystem detection for Next.js, Nuxt, SvelteKit, Astro and Remix.
·multi-browser capture: Chromium, Firefox and WebKit via Playwright.
·pixelmatch diffing (0–100%): pixel-level diffing via pixelmatch with a standardized 0–100 diff percentage.
·AI analysis (BYOK): BYOK support for OpenAI and Anthropic vision models on a unified confidence scale.
·Git baselines (orphan branch): orphan-branch storage for baseline screenshots via worktrees.
·plugin architecture (6 hooks + Figma/perf/monitor): 6 lifecycle hooks with error isolation, plus Figma, perf-budgets and monitoring plugins.
SECURITY
·hardened by default: shell-injection prevention, path-traversal guards, API-key redaction, and XSS-escaped HTML reports.
TESTING
·395 tests across 26 test files covering 27 source files: vision, crawler, Playwright, plugins, reporters, storage and E2E.