Notes about code should be written by whoever read the codeI maintain a tool that keeps notes about a codebase. An agent does a task, and at the end it writes down what it worked out. There is an easier way to build this. I chose not to take it. The easier deAug 3 · 6 min readcodebase-memory.hashnode.dev
From four tools to twoThis is an origin story told through deletions. Almost everything I consider a real design decision in this project was a removal, and each one taught me something I could not have reasoned my way to Aug 3 · 13 min readcodebase-memory.hashnode.dev
The tool the agent does not callYou can ship a tool, document it, and inject an instruction telling the agent to prefer it, and the agent will still reach for grep. Why availability is not adoption, with the mistake I shipped myselfAug 3 · 8 min readcodebase-memory.hashnode.dev
Why an index of your code cannot answer the same question twiceBuilding a graph of a repository is a good idea and it works. It is also the wrong shape for one specific job that people keep expecting it to do, and the reason is not a quality problem you can fix wAug 3 · 10 min readcodebase-memory.hashnode.dev
Where the tokens go in an agent sessionA long session bills for something. The instinct is to blame the tool that returned a big block of output. That instinct is almost always wrong, and the arithmetic that explains why also tells you theAug 3 · 7 min readcodebase-memory.hashnode.dev
