CCAR-F scenarios › Code Generation with Claude Code · Question 13 of 60
Scenario 2 · Code Generation with Claude Code
Your engineering team relies on Claude Code to generate, refactor, debug, and document code. You're responsible for fitting it into the team's workflow: custom slash commands and skills, a sensible CLAUDE.md setup, and knowing when a task calls for plan mode and when direct execution is enough.
Claude Code Configuration & Workflows
The root CLAUDE.md has a long section on database migration conventions. Migration files live in many places (services/*/db/migrations/ and libs/*/migrations/), and the team notices that the section takes up context in every session, including pure frontend work. What is the best way to make these conventions load only when they are relevant?
Show answer and explanation
Answer: D — Move the conventions into .claude/rules/migrations.md with YAML frontmatter paths set to a glob such as "**/migrations/**/*.sql", so they load only when matching files are edited.
Path-scoped rules in .claude/rules/ load only when Claude is working on files matching their glob, which cuts irrelevant context, and they work across directories in a way directory-level CLAUDE.md files can't (task 3.3). Per-directory CLAUDE.md files (a) duplicate content and need manual upkeep for every new service. A skill (b) depends on someone remembering to invoke it. A rules file without a paths field (c) is better organized but still loads in every session, so the token cost the team noticed stays.
More from this scenario
- Your team's /map-dependencies skill reads dozens of files and produces several thousand…
- Two hours into exploring an unfamiliar payments codebase, Claude Code's answers start to…
- Your team is deciding which upcoming tasks should start in plan mode and which can go…
- You ask Claude Code to convert legacy log lines into a new structured format. Despite…
- Before a large refactor, Claude Code must survey around 300 files to catalog every place…