Database over .md?
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
A developer working on a year-long project managed with Claude Code is experiencing increasing errors. These issues stem from agents imperfectly grepping a growing number of larger and longer .md files for context and identifiers. The developer questions whether using a database instead of .md files would be a better approach, wondering if others are already implementing such a solution.
In the midst of a probable year-long development project largely managed through Claude Code, recently (3 months in or so) I've been encountering an increasing number of errors that at their root have to do with agents imperfectly grep'ing larger and large numbers of longer and longer .md files for context, information, identifiers, etc.
Bringing this problem in front of my main controller agent, it conceded the point, and has collated our conversation into the following report. I'm enclosing it, but further am curious about all of your experiences.
One Fact, One Home
Truly, beyond that there's a real argument that these subreddits could build from their users' experiences a compendium of better methods that remove some of the weaknesses of these otherwise impressive tools.
PS Or I'm an idiot, and you are all doing this already and I hadn't realised!