Building a Coding Agent

• ✓Build a coding agent that reads, edits, and runs code in a tool-use loop
• ✓Implement the agentic loop from scratch (perceive, decide, act, observe)
• ✓Define a tool-use contract and parse tool calls from a local LLM
• ✓Sandbox all file access and gate risky tools behind an approval policy
• ✓Ground the agent in a real codebase and let it self-correct against tests
• ✓Evaluate the agent on an auto-graded task suite (does the code pass?)

Build the brain behind tools like Claude Code, Cursor, and Aider — an LLM that actually reads and changes your codebase in a loop, not a chat window you copy-paste from.