Add CLAUDE.md with project instructions

CLAUDE.md

@@ -0,0 +1,55 @@
+# Claude Code Instructions
+
+## Git Workflow
+
+**IMPORTANT: Never commit directly to `main` branch.**
+
+- Always create a feature branch for any changes
+- Branch naming: `feature/<description>`, `fix/<description>`, `chore/<description>`
+- Create a Pull Request for review before merging to main
+- Run tests before pushing: `pytest tests/e2e/test_knowledge_creation_live.py`
+
+Example workflow:
+```bash
+git checkout -b feature/my-new-feature
+# make changes
+git add -A
+git commit -m "Description of changes"
+git push -u origin feature/my-new-feature
+# Then create PR on GitHub
+```
+
+## Project Structure
+
+- `src/knowledge_base/` - Main application code
+- `src/knowledge_base/slack/` - Slack bot handlers
+- `tests/e2e/` - End-to-end tests
+- `deploy/terraform/` - GCP infrastructure
+- `docs/adr/` - Architecture Decision Records
+
+## Deployment
+
+The Slack bot is deployed on GCP Cloud Run:
+- Project: `ai-knowledge-base-42`
+- Region: `us-central1`
+- Service: `slack-bot`
+
+To deploy after merging to main:
+```bash
+cd /home/coder/Devel/keboola/headless-knowledge-base
+gcloud builds submit --project=ai-knowledge-base-42 --config=cloudbuild.yaml .
+```
+
+## Testing
+
+Run E2E tests with:
+```bash
+set -a && source .env.e2e && set +a
+.venv/bin/python -m pytest tests/e2e/test_knowledge_creation_live.py -v
+```
+
+## Key Configuration
+
+- LLM Provider: Gemini (`gemini-2.0-flash`)
+- Vector Store: ChromaDB on Cloud Run
+- Database: Ephemeral DuckDB (analytics only)