Get productive in an unfamiliar codebase in one day
Build a working mental model of a repo you've never seen, and leave behind a CLAUDE.md that compounds.
- Outcome
- A traced end-to-end request path, a written map of conventions and hazards, and one small merged PR.
- Time
- 3–6 hours
This workflow is part of PromptKit Pro
A handful of prompts and workflows stay free for everyone — Pro unlocks the full library, including this one, plus the guided runner.
The instinct in a new repo is to read it — open the root, start at src/,
work down. This scales terribly past about 20k lines, and it optimizes for the
wrong thing: breadth of exposure rather than the ability to place a change
correctly.
What actually makes you productive is narrow and specific: how a request flows
through the layers, what conventions are enforced, and where the dangerous
parts are. This playbook gets those three things by tracing one path deeply
instead of skimming everything, then writes the result into a CLAUDE.md that
makes every future session in this repo start smarter.
The last step — a deliberately trivial PR — exists to test the model you just built. Getting a two-line change through review teaches you more about the team's actual standards than a week of reading.
When not to use this
Not for a repo you'll touch once and never return to — the CLAUDE.md investment doesn't pay back. Not for a codebase under 2k lines; just read it.
Did this work?
You can answer "where would I add X?" for three plausible features without searching, and your first PR merged without a convention-violation comment.
Tested on claude-opus-5. Evidence status is draft; it moves to battle-tested only on recorded runs, never by hand.