Imported from PRANAV05092003/Research-copilot (AGENTS.md). Install upstream with npx skills add PRANAV05092003/Research-copilot. Copyright stays with the author.

Agent Guidelines for Research Copilot

This document outlines how future coding agents should interact with this repository.

1. Strict Type Checking: The backend enforces mypy with strict checking. All functions must have type hints for arguments and return types.
2. Testing Constraints: Every unit test MUST execute fully offline. The default LLM_PROVIDER and EMBEDDING_PROVIDER must be set to mock. Any tests that hit an external provider must be explicitly marked and skipped in CI.
3. Database Usage: Do not use Base.metadata.create_all() in production code. Use Alembic for all schema migrations.
4. Agent Workflow: The multi-agent capabilities are built using LangGraph. Adding a new agent requires modifying the graph definition in backend/app/ai/agents/graph.py and creating a new node in the nodes/ directory.
5. Linting: The backend uses ruff and the frontend uses eslint and prettier. Run make lint before pushing any code.