Files
PromptArch/CHANGELOG.md

259 lines
14 KiB
Markdown

# Changelog
All notable changes to this project will be documented in this file.
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning]
## [2.0.0] - 2026-03-18 21:33 UTC
### Added
- **SEO Audit Export** — Export SEO/GEO audit reports as HTML or PDF with comprehensive fix instructions
- **SEO Report Generator** — Standalone `lib/seo-report.ts` utility with color-coded scores, issue tables with fix instructions, and print-friendly CSS
- **Default Vibe Architect** — Welcome screen now opens to Vibe Architect by default (was Prompt Enhancer)
- **Vibe Architect Dedicated Route** — Full-screen immersive mode at `/vibe` with `vibeMode` prop
### Changed
- **General Agent Plain Chat** — General mode no longer triggers plan/code flow, now works as plain chat
- **SEO Follow-up Fix** — Non-visual agents (SEO, content, SMM) preserve existing canvas on follow-up messages
### Technical Details
- Files modified: 3 (AIAssist.tsx, page.tsx)
- Files added: 2 (seo-report.ts, vibe/page.tsx)
(https://semver.org/spec/v2.0.0.html).
## [1.9.0] - 2026-03-18 21:05 UTC
### Changed
- **Global Rebrand: AI Assist -> Vibe Architect** — All instances renamed across UI, sidebar, translations (EN/RU/HE)
- **Sidebar Highlight** — Vibe Architect button now has a gradient blue-purple border and bold blue text when not active, making it stand out in the navigation menu
### Technical Details
- Files modified: 3 (Sidebar.tsx, AIAssist.tsx, translations.ts)
## [1.8.0] - 2026-03-18 21:02 UTC
### Added
- **Vibe Architect Dedicated Mode** — Full-screen immersive AI coding experience
- New route: `/vibe` ([rommark.dev/tools/promptarch/vibe/](https://rommark.dev/tools/promptarch/vibe/))
- No sidebar, no navigation — just AI Assist in full screen
- Rebranded as "Vibe Architect" with dedicated messaging
- `vibeMode` prop on AIAssist component for label overrides
### Fixed
- **SEO follow-up preview** — Canvas no longer breaks when asking follow-up questions after SEO audit
- `isModifying` overlay only triggers for visual agents (code, web, app, design) — not SEO/content/SMM
- Non-visual agent follow-ups preserve existing canvas if no new preview is generated
### Technical Details
- Files added: 1 (`app/vibe/page.tsx`)
- Files modified: 1 (AIAssist.tsx: +8/-3 lines)
- New prop: `vibeMode` on AIAssist component
## [1.7.0] - 2026-03-18 20:44 UTC
### Added
- **Comprehensive SEO Audit Engine** — Industry-grade URL analysis modeled after Screaming Frog, Ahrefs, Sitebulb, and Semrush
- Full meta tag analysis: title (length/status), description (length/status), keywords, viewport, charset, robots directives
- Open Graph & Twitter Card validation
- Heading structure audit: H1 count check, hierarchy analysis (H1-H6), multiple H1 detection
- Link analysis: internal/external/nofollow counts, external link sampling
- Image audit: total count, alt text coverage %, lazy loading detection, missing alt samples
- Content analysis: word count, sentence count, paragraph count, avg words/sentence
- Structured data detection: JSON-LD, Microdata, 12+ schema types (Article, FAQ, Product, LocalBusiness, etc.)
- Performance signals: server info, response time, HTML size, external scripts/stylesheets, inline styles, preconnect/preload/dns-prefetch, async/defer detection
- Accessibility checks: lang attribute, ARIA labels, first image alt text
- Hreflang tag detection for international SEO
- Canonical tag validation with mismatch detection
- **Automated scoring**: Overall, Technical, Content, Performance, Social scores (0-100)
- **Issue detection**: Critical/Warning/Info severity levels across all categories
- **Plan-First Flow Fix** — Non-code agents (SEO, content, SMM, design, web, app) now bypass the plan-first workflow
- No more "Start Coding" or "Modify Plan" buttons for non-code agent responses
- SEO agent goes straight to preview mode after generating audit reports
- Plan card only shown for code/general agents
### Changed
- `/api/fetch-url` route completely rewritten: 300+ lines of comprehensive HTML parsing and scoring
- SEO agent data injection now passes structured audit data with scores, issues, and category breakdowns
### Technical Details
- Files modified: 5 (AIAssist.tsx, qwen-oauth.ts, ollama-cloud.ts, zai-plan.ts, openrouter.ts)
- Files rewritten: 1 (`app/api/fetch-url/route.ts` — complete rewrite)
- New component-level guard: `isCodeAgent` condition for plan-first workflow
## [1.6.0] - 2026-03-18 20:34 UTC
### Added
- **SEO Web Audit** — SEO agent can now fetch and analyze live websites
- `/api/fetch-url` route extracts: title, meta tags, headings (h1-h6), links (internal/external), images (alt text coverage), canonical URL, OG tags, text content length
- Auto-detects URLs in user input when SEO agent is active
- Pre-fetches page data before sending to AI for comprehensive audit reports
- Supports up to 2 URLs per request
- **SEO Auto Web Search** — When SEO agent is active and no URL is provided, web search is automatically triggered
- Uses same SearXNG infrastructure as manual web search toggle
- No manual toggle needed in SEO mode
- **Updated SEO Agent Prompt** — All 4 AI services now instruct SEO agent about web audit and search capabilities
- Added `[WEB_AUDIT:url]` and `[WEB_SEARCH:query]` tool markers
- Added "WEB TOOLS" section to system prompts
### Technical Details
- Files modified: 5 (AIAssist.tsx, qwen-oauth.ts, ollama-cloud.ts, zai-plan.ts, openrouter.ts)
- Files added: 1 (`app/api/fetch-url/route.ts`)
- New API endpoint: `GET /api/fetch-url?url=https://example.com`
## [1.5.0] - 2026-03-18 20:29 UTC
### Added
- **Modification Progress Overlay** — Visually appealing spinner when modifying existing generated code
- Full-screen canvas overlay with spinning ring + Wrench icon
- "Modification in Progress" label with animated bouncing dots
- Canvas hidden during modification, revealed only when AI finishes
- `isModifying` state tracked via `assistStep === "preview"` detection
### Fixed
- **Preview blink during modification** — Old preview no longer flashes while AI rewrites code
### Technical Details
- Files modified: 1 (AIAssist.tsx: +21/-2 lines)
- New state:
- New import: from lucide-react
## [1.4.0] - 2026-03-18 19:57 UTC
### Added
- **Review Code Button** — Post-coding action to send generated code back to AI for review
- `reviewCode()` function sends code with review-focused prompt
- Emerald-green "Review" button alongside Preview and Modify
- 3-column post-coding action grid: Preview / Review / Modify
- **Web Search Grounding** — Enrich AI prompts with live web search results
- `lib/services/search-api.ts` — SearXNG public API wrapper with 4 instance fallback
- `/api/search` route — server-side search proxy endpoint
- Toggle button in toolbar to enable/disable (amber highlight when active)
- Shows "Searching the web..." status while fetching
- Appends top 5 results as `[WEB SEARCH CONTEXT]` to user prompt
- **Responsive Preview** — Device size selector in canvas panel
- Full / Desktop (1280px) / Tablet (768px) / Mobile (375px) buttons
- Centered device frame with border, shadow, and scroll on overflow
- Only visible in preview mode, below Live Render / Inspect Code tabs
### Fixed
- **Model selector text color** — Option text now white on dark theme (`bg-[#0b1414] text-white`)
- **Button text overflow** — Shortened labels (Preview/Review/Modify), added `min-w-0` for proper truncation
- **Duplicate activateArtifact button** — Original button now hides when post-coding action row is shown
### Technical Details
- Files modified: 1 (AIAssist.tsx: +85/-11 lines)
- Files added: 2 (`lib/services/search-api.ts`, `app/api/search/route.ts`)
- New state: `deviceSize`, `webSearchEnabled`
- New function: `reviewCode()`
## [1.3.0] - 2026-03-18 18:51 UTC
### Added
- **OpenRouter Integration** — 4th AI provider with API key auth, 20+ model support
- New `lib/services/openrouter.ts` streaming service
- Provider selector in AI Assist: Qwen, Ollama, Z.AI, OpenRouter
- Default model: `google/gemini-2.0-flash-exp:free`
- Custom model selector with popular free/paid model presets
- Settings panel: API key input with validation and model picker
- **Plan-First Workflow** — AI now generates a structured plan before code
- PLAN MODE instructions injected into all 4 service system prompts
- Plan card UI with architecture, tech stack, files, and steps
- `parsePlanFromResponse()` extracts plans from AI markdown output
- `[PLAN]` tags hidden from displayed chat messages
- Three action buttons: **Modify Plan** / **Start Coding** / **Skip to Code**
- **Post-Coding UX** — Preview + Request Modifications after code generation
- After "Start Coding" approval, AI generates code with `[PREVIEW]` tags
- Canvas opens automatically with renderable previews
- Two post-coding buttons: **Preview** (re-opens canvas) and **Request Modifications**
- `isApproval` flag prevents stale React closure bugs in approval flow
- **Enhanced Prompt Engine** — New modular prompt enhancement system
- `lib/enhance-engine.ts` with 9 enhancement strategies
- Strategies: clarify, add-context, add-constraints, structure, add-examples, set-tone, expand, simplify, chain-of-thought
- Context-aware enhancement based on detected intent type
- 11+ intent detection patterns (coding, creative, analysis, etc.)
- Smart strategy selection per intent for optimal prompt refinement
- **Streaming Plan Mode** — Real-time plan parsing during AI response
- `wasIdle` flag captures initial request phase before state updates
- Canvas display suppressed during plan generation, enabled after approval
- Post-stream routing: plan card for initial requests, preview for approvals
- Tab `showCanvas` state gated by plan phase
### Changed
- **AIAssist.tsx** — Major refactor for plan-first flow
- `handleSendMessage` now accepts `isApproval` parameter to prevent stale closures
- `approveAndGenerate()` passes `isApproval=true` to bypass idle detection
- `assistStep` state machine: `idle -> plan -> generating -> preview`
- `parseStreamingContent()` filters `[PLAN]` tags from displayed output
- **PromptEnhancer.tsx** — Rebuilt with modular enhance engine
- Moved enhancement logic to `lib/enhance-engine.ts`
- Added expand, simplify, and chain-of-thought strategies
- Improved intent detection and strategy mapping
- **SettingsPanel.tsx** — Added OpenRouter provider configuration
- API key input with validation
- Model selector with preset dropdown
- Provider-specific endpoint display
- **model-adapter.ts** — Extended with OpenRouter provider support
- New adapter mapping for OpenRouter service
- Unified interface across all 4 providers
- **translations.ts** — Added i18n keys for plan mode, OpenRouter, post-coding actions
- Keys: `modifyPlan`, `startCoding`, `skipToCode`, `requestModifications`
- OpenRouter provider labels and descriptions
- English, Russian, Hebrew translations updated
- **store.ts** — Added `selectedProvider` state for multi-provider selection
- **types/index.ts** — Added `PreviewData` interface for typed canvas rendering
- **adapter-instance.ts** — Registered OpenRouter in provider registry
### Fixed
- **Stale React closure** in `approveAndGenerate``setAssistStep("generating")` followed by `handleSendMessage()` read stale `assistStep` value. Fixed with explicit `isApproval` boolean parameter.
- **Plan card reappearing after code generation** — Post-stream logic now correctly routes to `preview` mode after approval coding, not back to `plan` mode.
- **Canvas auto-opening during plan phase** — `setShowCanvas(true)` in `onChunk` now gated by `!wasIdle` flag.
- **i18n missing keys** — Added `syncComplete` for Hebrew, fixed double commas in multiple translation strings.
### Technical Details
- Files modified: 11 (960 insertions, 194 deletions)
- Files added: 2 (`lib/enhance-engine.ts`, `lib/services/openrouter.ts`)
- Total project lines: ~10,179 across core files
- System prompt PLAN MODE block added to: `qwen-oauth.ts`, `ollama-cloud.ts`, `zai-plan.ts`, `openrouter.ts`
## [1.2.0] - 2026-01-19 19:16 UTC
### Added
- **SEO Agent Behavior Fixes**
- SEO agent now stays locked and answers queries through an SEO lens
- Smart agent suggestions via `[SUGGEST_AGENT:xxx]` marker for clearly non-SEO tasks
- Visual suggestion banner with Switch/Dismiss buttons
- Prevented unwanted agent auto-switching mid-response
- **z.ai API Validation**
- Real-time API key validation with 500ms debounce
- Inline status indicators:
- Checkmark with "Validated Xm ago" for valid keys
- Red X with error message for invalid keys
- Loading spinner during validation
- "Test Connection" button for manual re-validation
- Persistent validation cache (5 minutes) in localStorage
### Changed
- Updated Settings panel with improved UX for API key management
- Enhanced agent selection behavior to prevent unintended switches
## [1.1.0] - 2025-12-29 17:55 UTC
### Added
- GitHub integration for pushing AI-generated artifacts
- XLSX export functionality for Google Ads campaigns
- High-fidelity HTML report generation
- OAuth token management for Qwen API
## [1.0.0] - 2025-12-29 13:51 UTC
### Added
- Initial release of PromptArch
- Multi-provider AI support (Qwen, Ollama, Z.AI)
- Prompt Enhancer with 11+ intent patterns
- PRD Generator with structured output
- Action Plan generator with framework recommendations
- Visual canvas for live code rendering
- Multi-language support (English, Russian, Hebrew)