Here are a few things I hadn't used before:
🔧 Custom agents — just drop a .md file into .claude/, and you get a separate agent with a specified name and set of permissions. Boris has a code-simplifier that simplifies code after work, and verify-app that tests the result.
🔧 Hooks — scripts that trigger on Claude events. Boris uses code formatting after each change to avoid errors in CI. You can also set up a notification in Telegram when Claude requests permission.
🔧 Sandbox — an isolated environment where Claude cannot see files outside the project. This reduces the number of permission requests since there’s nothing to break inside the sandbox.
🔧 Plugin directory — LSP, MCP servers, agents, hooks. There’s an official directory from Anthropic, and you can create your own within the company.
🔧 Default effort level is set to high. Boris doesn’t lower it even for small tasks and doesn’t mind the tokens. Switching between levels happens right in the session using the arrows ← →.
