v0.2.0: Document Generation Quality Overhaul

What's New

6-Document Generation with Cross-Referencing

Documents are now generated sequentially in dependency order, with each document receiving all previously generated documents as context. This enables cross-referencing — PROMPTS.md references exact conventions from CLAUDE.md, README.md consolidates known gaps from SPEC.md, etc.

Generation order:

1. SPEC.md — no prior docs
2. CLAUDE.md — receives SPEC.md
3. PROMPTS.md — receives SPEC.md + CLAUDE.md
4. README.md — receives all 3 prior docs
5. START_HERE.md (new) — receives all 4 prior docs
6. CONVERSATION.md — data-driven, unchanged

New Document: START_HERE.md

A bridge document for users new to Claude Code or non-technical users. Includes:

• Prerequisites checklist (adapted to the project's tech stack)
• Step-by-step setup instructions
• First prompt to paste into Claude Code
• Troubleshooting section for common issues
• File reference table

Overhauled LLM Prompts

All 5 document generation prompts replaced with expanded versions:

• Negative/positive examples — concrete "do this / don't do this" for the LLM
• Tech stack consistency checks — no web-server patterns for desktop apps, correct build commands
• Information categorization — Decided (explicit) / Implied (inferred) / Unknown ([TBD] markers)
• Interface contracts in real language — Rust structs for Rust projects, TypeScript interfaces for TS projects, never pseudocode JSON
• Anti-patterns sections in CLAUDE.md generation
• Complexity indicators and watch-out sections in PROMPTS.md generation

New Conversation AI Personality

The planning conversation system prompt has been rewritten with:

• Readiness tracking — must-haves vs should-haves checklist before generation
• Max 2 questions per message — focused, one-topic-at-a-time dialogue
• Conversation stages — Discovery → Decisions → Convergence
• Constructive challenging — pushes back on vague answers, questions scope creep
• Typo/unclear term clarification — never propagates uncertain terms
• Advisory readiness report — summarizes gaps before generating, never blocks generation

Frontend Updates

• START_HERE.md tab appears first in document preview
• Old sessions without START_HERE.md fall back gracefully to the first available tab

What Doesn't Change

• Database schema (no migration needed)
• Tauri IPC commands
• TypeScript types
• Save-to-folder behavior (START_HERE.md auto-included)
• LLM temperature settings (0.4 primary, 0.3 retry)
• CONVERSATION.md generation (data-driven, no LLM)

Quality

• cargo build — clean
• cargo test — 41/41 passing
• npx tsc --noEmit — 0 errors

Full Changelog: v0.1.0...v0.2.0

AuraForge v0.1.0

AuraForge v0.1.0 — Initial Release

A desktop app that transforms project ideas into structured, implementation-ready planning documents through conversational AI — powered by local LLMs via Ollama.

Download

• AuraForge_0.1.0_aarch64.dmg — macOS installer (Apple Silicon)

Requirements

• macOS with Apple Silicon (Intel Macs work too, just slower inference)
• Ollama installed and running

What's Included

• Conversational planning with streaming LLM responses
• Web search integration (DuckDuckGo, Tavily, SearXNG)
• Five-document generation pipeline (SPEC, CLAUDE.md, PROMPTS, README, CONVERSATION)
• Session management with auto-naming and cascade deletion
• Setup wizard for first-time configuration
• Full keyboard shortcut support

Installation

1. Install Ollama and open it
2. Download the DMG below
3. Drag AuraForge to Applications and open it
4. Follow the setup wizard