RivaanRanawat/ai-coding-agent ? reverse-engineered prompt

Build me a Claude Code style AI coding agent in Python that can take a task, use tools, and keep working until it finishes. I want it to work in both interactive chat mode and a single run mode, with streaming replies, multi turn conversations, and a nice terminal interface that shows what the agent is doing in real time.

It should be able to read, write, and edit files, list folders, search text, run shell commands, browse the web, and keep simple memory and todo lists. Please add session saving and resuming, checkpoints, token usage tracking, and context management so it can trim old tool output when things get too long.

I also want safety controls for risky commands, approval prompts for edits, and path based checks. Support MCP servers, custom hooks before and after runs or tool calls, and a couple of helpful subagents for investigating code and reviewing changes. If useful, look up current docs online as you build it.