Track agent memory in git for coding work
Store project context in git so a coding agent can reuse decisions and notes across sessions.
- What they did
- They created a small Go program and used it to initialize a project with persistent memory. The bootstrap step wrote out a knowledge/ tree of Markdown pages with YAML headers, along with an agent skill file and an AGENTS.md guide. The agent can consult that memory through a CLI or an MCP server over stdio, and the setup is meant to fit Claude Code, Cursor, and Codex. Its rules tell the agent to check existing material before adding anything new and to open only the index pages required for the task.
- How it went
- According to the README, searches come back in under 300µs, validating 50+ concepts takes around 4ms, and token use falls by 80%. Those results are based on the project’s own benchmark numbers and comparison table, not outside testing. The repository itself is very early, with just one commit and 3 stars.
- Tips
- If you try this pattern, keep the memory structure simple and make the “read first, write later” rule explicit. Treat the performance claims as self-reported unless you rerun the benchmark yourself with make benchmark on a local LLM.
Try it yourself with Claude Code
Create a git-backed memory system for [project name] inside [folder name]. Use Markdown files with YAML frontmatter to record decisions, conventions, and open tasks, and add the instructions needed so future coding-agent sessions read the right index pages before they write anything new. After you draft the files, stop and show me the exact changes you plan to make; do not commit, delete, or submit anything until I approve the file list and contents. Then carry out the setup for [agent/tool] and keep the memory updated at the end of each session.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Create a fast phone agent with live voice streaming
They assembled a call-answering agent with live transcription, generation, and speech output, then moved it closer to the services to get about 400ms latency.
Trace C# symbol usage with a Roslyn-backed index
Graphify C# helps agents map real C# usages, including overrides and implementations, from compiler-resolved project data.
Build a terminal calling tool with human backup
They wired a CLI agent to place calls and loop in a person only when it needed an answer.
Assemble a R coding agent from three basic tools
They wired a Claude agent in R to file and shell helpers, then hardened it with safer path checks and editing functions.
Profile and tune a CUDA kernel with an agent
They used a LangGraph agent to test CUDA kernel tweaks, verify results, and profile candidates with Nsight.
Let an AI agent inspect a Windows crash dump
Use MCP to give an AI agent structured crash details from a Windows dump so it can track down the fault and propose a fix.