Skip to content

CLAUDE.md

Latest commit

 

History

History
48 lines (42 loc) · 3.57 KB

CLAUDE.md

File metadata and controls

48 lines (42 loc) · 3.57 KB
Raw

Yearn Curation Score Guidelines

Development Commands

  • Run Scripts: uv run scripts/scan_factories.py or uv run reports/scripts/fetch_defillama_tvl.py
  • Code Formatting: uv run -m ruff format .
  • Update Lock File: uv lock (after changing dependencies in pyproject.toml)
  • Use .env for environment variables, RPC URLs, and secrets. Never commit secrets.
  • Python scripts that need .env, RPC URLs, or explorer keys should use scripts/env.py: call load_repo_env() at the entrypoint and use get_rpc_url(chain_id) / get_explorer_api_key(name) instead of duplicating load_dotenv or env-var alias logic.

Dependencies

  • Core Dependencies:
    • web3: For blockchain interaction
    • python-dotenv: For environment variable management
    • requests: For API calls
  • Dev Dependencies:
    • ruff: For code formatting and linting

Style Guidelines

  • Imports: Group standard library imports first, followed by third-party imports, then local imports
  • Code Organization: Use functions and classes with clear responsibilities
  • Type Hints: Use Python typing module for function parameters and return values
  • Error Handling: Use specific exception catching and provide informative error messages
  • Documentation: Include docstrings for functions/classes with descriptions and parameter details
  • Naming: Use snake_case for variables/functions, PascalCase for classes
  • Environment Variables: Store sensitive data (API keys, RPC URLs) in .env file (never commit)

Coding Practices

  • Cache results to minimize blockchain RPC calls
  • Use LRU cache for frequently accessed data
  • Prefer batch requests when interacting with blockchain
  • Log important information with appropriate log levels
  • Organize data output in markdown tables for readability
  • Never assume, always verify and provide reference to the source of the data
  • Do not trust protocol documentation alone. Treat docs as a map, then verify claims onchain when possible.
  • Back material findings with valid links: contract pages, transaction links, source docs, dashboards, or API outputs. TVL and allocation claims need source links too.
  • Try to use Etherscan for verification of blockchain data.
  • Use Foundry cast for direct onchain checks. Load RPC URLs from .env: use RPC_1 first for Ethereum mainnet and RPC_2 as the fallback; for other chains use the chain-specific RPC_<chain_id> variable when available.
  • For Python-based blockchain/API scripts, prefer scripts/env.py over direct os.getenv lookups for RPC and explorer keys so aliases stay consistent across scripts.

Report, Graph, and Reassessment Workflow

  • Before starting a report, graph, or reassessment task, pull latest master.
  • When the report, graph, or reassessment task is ready, open a draft pull request with a concise summary and validation notes.
  • For generating a risk assessment report or making report changes, focus only on reports unless the user asks otherwise.
  • Always read the existing report and understand the context before generating a report, graph, or reassessment.
  • If information is unavailable, mark it as TODO; do not guess.
  • Look for hidden or indirect fund-loss paths, not only documented happy paths. Check privileged minting, upgrade paths, oracle controls, emergency roles, offchain custody, redemption queues, and dependency failure modes.
  • Use skill generating-risk-reports for new or updated risk reports (reports/skill.md), generating-dependency-graphs for dependency graphs (reports/graph/SKILL.md), and risk-report-reassessment for focused refreshes (reports/reassessment/SKILL.md).