# 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](https://semver.org/spec/v2.0.0.html).

## [1.3.0] - 2025-03-18

### 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] - 2025-01-19

### 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] - 2025-01-15

### 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] - 2025-01-01

### 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)