Claude Code¶

An agentic coding assistant which not only address coding assistant, but also discovery and design.

The architecture of the assistant looks like:

• CLAUDE.md file represents memory across sessions. It can include style guidelines and common commands. This file is automatically loaded into the context.
• See overview.

Capabilities¶

• Conversation history is stored locally, which is loaded into context.
• Can launch sub-agents
• Manage to dos
• Refer back to previous information
• Agentic search

Integrating with existing git repo.¶

Here are 3-step playbook for getting Claude Code productive and safe on existing repos.

• Create a first CLAUDE.md file. Don't use /init as-is — it produces bloated output. As good practices:

• Adopt TDD
• Plan First, then Scale with skills and plugins.
  • Planning is most valuable when you're uncertain about the approach or the change spans multiple files. Ctrl-G to update the plan.
  • Prefer CLIs over MCP
  • Leverage plugin like Superpowers
  • Manage context aggressively — Use /clear between unrelated tasks, /compact to summarize mid-session, and subagents for investigation

Useful prompts¶

Sources¶

• Deeplearning training
• Using Claude for a lot of things
• Use LiteLLM as proxy for Claude code - tutorial