A platform team’s Claude Code sessions follow the house error-handling convention, but the same repository’s nightly CI job, which runs Claude Code headlessly, ignores it. The convention was written by the team lead months ago.
A healthcare startup’s root CLAUDE.md has grown to 700 lines: general rules, 200 lines of React guidance, 150 lines on database migrations and a long release procedure. Engineers report that Claude often ignores rules near the bottom.
Which two changes will most reduce what loads into every session while keeping the guidance available? (Select 2.)
An insurer’s security team requires that Claude Code never reads files in secrets/ on any developer machine. A developer proposes adding “Never open anything in secrets/” to the managed-policy CLAUDE.md.
A payments team creates .claude/skills/release/SKILL.md, which tags a version and publishes it to the package registry. A week later, during an unrelated refactor, Claude runs the release skill on its own after tests pass.
Which change prevents this while keeping /release available to engineers?
A developer writes a skill with allowed-tools: Read Grep Glob, expecting it to be read-only. During a run, Claude edits a file after the user approves an Edit permission prompt.
A hospital software team has a /audit-phi-logging skill that scans hundreds of service files for places where patient identifiers might be logged. After it runs, developers find their session slow and cluttered.
A logistics company keeps React components in web/src/components/ and its GraphQL resolvers in server/src/resolvers/. Each has its own detailed conventions, currently all in a 450-line root CLAUDE.md.
Which approach best loads each set of conventions only when relevant?
A fintech team writes a rule with paths: ["src/payments/*.ts"]. Claude follows it when editing src/payments/refund.ts, but ignores it when working on src/payments/providers/stripe.ts.
A university’s research-software team wants conventions for database migration files, which live in db/migrations/ in six different services across the repository.
A product team needs to move session storage from in-memory to Redis. The change touches the auth service, two API gateways and the test harness, and there is debate about whether to use a cache-aside or write-through design.
How should the engineer approach this with Claude Code?
During plan mode, Claude proposes a plan to split a monolith’s reporting module into a separate service. The engineer agrees with most of it but wants the database migration moved to the end and one endpoint left untouched.
An insurance team asks Claude to “format policy numbers consistently”. Across three attempts Claude produces three different formats, each defensible. The team lead’s fourth message is “No — CONSISTENTLY, like our real policy numbers!”
A developer has corrected Claude four times in one session about how a caching layer should invalidate entries. Each fix introduces a new problem, and Claude now references approaches that were already rejected.
A fintech team wants Claude to implement a fee calculator with tiered rates, rounding rules and a cap. Previous attempts from a written description passed a quick manual check but failed in staging on edge cases.
Which two changes will most improve the next iteration? (Select 2.)
After a database column is renamed, three things break in a reporting service: a SQL query, the ORM model that maps it, and a CSV export that reads the model. A developer plans to report them to Claude one per message.
A platform team wants review findings posted as inline comments on the exact lines of a pull request, and wants the build to fail on any high-severity finding.
A CI job asks Claude Code to generate unit tests for new code, then, in the same session, asks it to review those tests for gaps. Reviewers keep finding weak assertions the review step approved.
A security-conscious fintech wants a nightly job that reads the codebase and reports dependency risks. It must never edit files or run arbitrary shell commands, and nobody will be watching.
Which configuration fits best?
You can change answers until you check. Nothing is saved or sent anywhere.