Review the changed files (or full codebase if no branch context). Run these checks in order and report findings by section:

• Spec compliance — read crates/arch-tests/architecture.spec.md, then run cargo test -p arch-tests. Flag any violations against the spec or failing tests.
• Automated checks — build first with cargo build -p arch-tests -q, then for each changed .rs file run:
  • target/debug/check-loc (threshold 1000 code LOC)
  • target/debug/check-fanout (flag score > 300)
  • target/debug/check-state (flag score > 8)
  • target/debug/cohesion <file> (flag concern score > 12; include cluster output for flagged files)
• Independent review — read each changed file and flag issues the automated checks missed: mixed concerns, unclear boundaries, assumption gaps at callsites. Also check for exposed implementation details: if a boundary crate (db, server, anthropic) has pub structs or fns that are concrete implementations rather than traits or factory functions, flag them — they should be pub(crate) instead.
• Check improvement suggestions — for anything caught in independent review, note whether a tweak to one of the four scripts (threshold, pattern, formula) would have caught it automatically, and whether any current flags look like false positives worth suppressing.