You are a senior developer working on the 'pexfi' project.

Any files generated by project scripts must be configured to live under /.cache/
When implementing new features reading blockchain lean towards direct contract reads, in exceptional cases if you believe data should be indexed in subgraph explain solid reasoning and ask approval.
When editing always read files before thinking to update your context, they might have changed since last read.
There are agent skills located in .agents/skills/. This is a canonical path for skills and new skills must be stored there.

Code Style

Follow formatting in .editorconfig and .prettierrc
Always use configured @ aliases when importing modules
Never comment WHAT code does, only WHY if really necessary
Never comment changes reasoning or your thinking process
Never comment your thinking process in the code
Never add 'as any' in TypeScript code, let typescript infer types or fix them properly
Always prefer explicit types over implicit ones in TypeScript code
Always prefer type/interface definitions over type assertions
You are only allowed to use viem/wagmi and never ethers from blockchain interactions.

Docs writing

Tests are stored in tests/ mirroring src/ folders.
vitest and react-testing-library is used
respect "globals: true" and do not import vitest vars when editing tests
After any of your changes tests must pass
For end to end tests we inject a wallet into app's global window. This wallet signs any tx. It is required to pass code between server and cli running tests.

All issues, tasks, comments, requests, commits are in English only regardless of input prompt.

For tasks, issues and contexts agents are only allowed to use YouTrack. Stop processing if not available. State the error.
For all chats related to tasks agents must use tools from YouTrack MCP server
Source of Truth for all tasks, tickets and statuses.
Agents are only allowed to interact with project: pexfi (id: PEX)

New functionality is covered by tests
All tests pass with no errors
Documentation and comments added to code are always in english regardless of prompt input. You may answer in any language.