I've been using Claude Code for my work for the past 6 months and it has been great. My workflow is very typical, start Claude Code > start planning my feature in plan mode > implement. And then just seeing the work, and occasionally steering it in the correct direction when it goes off track (which doesn't happen much).
But since Claude Code has amnesia and you can only put so much in your CLAUDE.md file, it always ended up brute forcing its way around the codebase to understand it first. Usually by picking an entry point first somehow and then walking up.
So I ended up building this, a simple CLI meant to be used by your AI agent, that you can drop in any project (as long as your language has a SCIP indexer).
It uses SCIP to index your codebase, and then the CLI converts it into a SQLite database. All commands are just wrappers around the queries, and you can just query the database directly. SCIP does not really work for documentation and general text files, I added an indexer for that in the CLI (this one is a bit wonky, I recently added it)
To set it up it: 1. Install the CLI in your system. 2. Install the SCIP indexer for your language.
To start using it in project: 1. Add the snippet in your agent memory file to tell it to prefer dora instead of other tools when doing code exploration. 2. Add the skill file for more detailed usage. 3. Add hooks to run the indexer at session start and at end of every turn, in the background.
I've been using it with my main work, and tweaking it as I find room for improvements. The idea is to keep it minimal. I'm able to get good plans on the first go.
Check it out: Website - https://dora-cli.dev/ GitHub - https://github.com/butttons/dora