The Claude Code Folder Setup

# PaperCurator: RAG-Powered Research Assistant

## Quick Start
```bash
pip install -r requirements.txt
python -m uvicorn app.main:app --reload  # :8000
pytest                                    # run tests
```

## Stack
- **Runtime:** Python 3.11+
- **Framework:** FastAPI with Pydantic validation
- **Database:** PostgreSQL 16 via SQLAlchemy
- **Vector Store:** OpenSearch (BM25 + dense)
- **LLM:** Ollama (local) / OpenAI (cloud)

## Code Conventions
- 4-space indentation, type hints everywhere
- All route handlers delegate to a service
- Use Pydantic models for all request validation

## What Claude Should NOT Do
- Never modify applied Alembic migrations
- Never hardcode secrets or connection strings
- Never use `Any` type, use proper generics

{
  "mcpServers": {
    "github": {
      "type": "http",
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": {
        "Authorization": "Bearer ${GITHUB_TOKEN}"
      }
    },
    "postgres": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@bytebase/dbhub", "--transport", "stdio"],
      "env": {
        "DB_URL": "${DATABASE_URL}"
      }
    },
    "datadog": {
      "type": "http",
      "url": "https://mcp.datadoghq.com/sse",
      "headers": {
        "DD-API-KEY": "${DD_API_KEY}",
        "DD-APP-KEY": "${DD_APP_KEY}"
      }
    }
  }
}

# rules/testing.md (only loads for test files)

---
paths:
  - "tests/**/*.ts"
  - "src/**/*.test.ts"
  - "src/**/*.spec.ts"
---

# Testing Conventions

- Use pytest with fixtures and parametrize
- Group tests by endpoint in test classes
- Each test should be independent
- Always test: happy path, one edge case, one error
- Mock external services, use real test database
- Never use `Any` type in tests

# skills/investigate-errors/SKILL.md

---
name: investigate-errors
description: "Investigates error traces, stack
  traces, and runtime failures. Reads logs,
  identifies root cause, suggests targeted fixes."
when_to_use: "Activate when the user shares an
  error message, stack trace, or test failure."
allowed-tools: Bash(pytest *) Read(**)
---

# Error Investigation

When you encounter an error trace:

## Step 1: Understand the error
- Read the full stack trace
- Identify exact file and line number
- Note the error type

## Step 2: Trace the cause
- Read the file where the error occurs
- Follow the call chain
- Check recent changes with git log

## Step 3: Suggest the fix
- Smallest change that fixes root cause
- Explain WHY it happened, not just HOW