Unify OpenCode and Codex starter kits

- Add .shared/prompts/ with tool-neutral prompts
- Add .opencode/config/project.md, rules/, skills/ from opencode-kit
- Add .codex/config/project.md from codex-kit
- Copy .shared/prompts/ to both .opencode/prompts/ and .codex/prompts/
- Add features/INDEX.md and features/README.md
- Update README with unified structure

The two original starter kits are now redundant.

.opencode/rules/general.md

@@ -1,20 +1,35 @@
 # General Rules
 
-## Projekt Start
+General development rules for the AI Coding Starter Kit.
 
-Bei einem neuen Projekt:
-1. README.md lesen / erstellen
-2. AGENTS.md lesen
-3. docs/planung/ öffnen und Projektstruktur verstehen
+## Feature Tracking
 
-## Naming
+- All features MUST be tracked in `features/INDEX.md`
+- Feature IDs are sequential: PROJ-1, PROJ-2, etc.
+- Never skip IDs or use non-sequential numbering
 
-- Kleinbuchstaben mit Bindestrichen: `mein-projekt`
-- Subdomains als Verzeichnisse unter `www/`
-- Document-Root immer `public/`
+## Git Workflow
 
-## Git Flow
+- Commits: `feat(PROJ-X): description`, `fix(PROJ-X): description`
+- Never commit secrets, API keys, or credentials
+- Always read file before modifying (never assume contents)
 
-1. Branch für Feature: `feature/<name>`
-2. Commit oft mit aussagekräftigen Messages
-3. Merge via PR oder direkt (bei kleinen Fixes)
+## File Structure
+
+- Feature specs: `features/PROJ-X-feature-name.md`
+- Tests: co-located next to source (`*.test.ts` next to `*.ts`)
+- shadcn/ui components: `src/components/ui/` (NEVER recreate)
+
+## State Management
+
+- State lives in files, not in memory
+- Re-read files when context is compacted
+- All progress in markdown files (INDEX.md, feature specs)
+
+## Context Recovery
+
+If context is compacted:
+1. Re-read relevant feature spec
+2. Check `git diff` for uncommitted changes
+3. Continue from where you left off
+4. Never restart from scratch without checking progress