Repository: github.com/maya-chen-demo/distributed-cache-toolkit
AI-collaboration discipline
CLAUDE.md (124 lines) carries explicit specs, invariants, and failure modes for the cache eviction policy. Commit history shows refactor PRs annotate Copilot suggestions before merge — accept/reject ratio ~3:1, never blind-applied. .cursorrules limits AI to inline edits within tested modules; integration tests are written by hand.
Code craft
Distributed-cache eviction policy has a state machine documented inline; tests cover 14 scenarios including partition-recovery races. Naming is consistent (verbNoun for actions, isX for predicates). No dead code; every public symbol has a callsite within the package or in tests.
Spec discipline
spec/eviction.md predates the implementation by 9 commits — coordination signal that design happens before code. Open issues are labelled by spec section; PR descriptions reference the spec section being modified. Healthy bidirectional spec↔code traceability.
Synthesis
Maya pairs senior distributed-systems instincts with rigorous AI-collaboration discipline. Strong recommend for technical interview.