Author a CLAUDE.md Hierarchy That Actually Steers a Monorepo
Take a small two-package monorepo that ships with no CLAUDE.md at all, author the root guidance file plus one package-level override, then drive Claude Code to implement a stub in each package. The two implementations must come out following diverging planted conventions (a JSDoc rule and a [shopkit] log prefix at the root, a Google-style docstring rule and an [inv] prefix inside the inventory package), and a local self-check verifies that divergence mechanically. That is your proof the hierarchy and the override actually reached the model. Submit the guidance files, the generated code, and the evidence for instant, rubric-based feedback.
1 hr
Est. time
4
Outcomes
5
Rubric criteria
65%
Pass score
What you'll learn
Skills you'll have real reps in after shipping this.
The scenario
Your team runs Claude Code across a monorepo where one set of instructions cannot speak for every package: the storefront is plain-Node JavaScript with its own logging and documentation habits, and the inventory service is standard-library Python with different ones. Pasting the right conventions into every prompt does not scale, and it transfers to exactly zero teammates. CLAUDE.md is the mechanism built for this: a memory file Claude Code loads into context at the start of a session, resolved as a hierarchy (enterprise, user, project root, subdirectory), where the file closest to the code wins for its own scope and @-path imports let one root file pull in shared docs.
This task makes you build that hierarchy and then catch it in the act. You author the root CLAUDE.md and a package-level override for packages/inventory, each stating deliberately diverging conventions, then have Claude Code implement one stub per package. If the storefront code comes out logging with [shopkit] while the inventory code logs with [inv], the override provably took effect, and check.py confirms it mechanically and writes your evidence. One accuracy point runs through the whole task: CLAUDE.md is advisory guidance the model reads, which is exactly why you test that it landed; hard enforcement is what hooks are for, and those come in a later level.
Your role
You are the engineer who owns agent guidance for a monorepo. Your deliverable is a working CLAUDE.md hierarchy (a root file with an @-path import plus one package override) and machine-verified proof that Claude Code followed each file in its own scope, packaged as a single submission.
Start the task to unlock the full brief
You'll get the step-by-step requirements, setup commands, the 5-criterion grading rubric, tips, and the ability to submit your solution for instant AI grading.
Free to start · submit when you're ready
Prove your CLAUDE.md hierarchy steers Claude Code, package by package
Every serious Claude Code setup starts with CLAUDE.md, the memory file the agent loads at the start of a session, and the skill this task builds is making that file actually work across a monorepo. You author a root CLAUDE.md with an @-path import of shared conventions plus a package-level override, then drive Claude Code to implement one stub in each package and catch the hierarchy in the act: the JavaScript comes out documented and logging one prefix while the Python comes out with Google-style docstrings and a different prefix, because the file closest to the code wins for its scope. A local self-check verifies the divergence mechanically, runs both test suites, and generates your evidence. You come away knowing how to write scoped, concise agent memory, how to verify guidance reached the model instead of assuming it did, and where advisory steering ends and hook-based enforcement begins.