[mosobande]

Adds a required quality lens for structure, defaults for apps with child users, and two Rust entries for a client core shared across platforms.

What changes

• practices/architecture.md: Quality lens, required. Judge backend structure by AX and DX. Judge frontend structure by AX, DX and UX, and a structural change must not make the UI or UX worse. The definitions link the AI Coding Dictionary's AX and DX entries; human review is part of DX.

• practices/child-users.md (new), with four entries:

  • a separate child app when the adult app carries adult-only features;
  • no third-party ads, analytics or crash SDKs (rejected);
  • content-free push;
  • birth month and year as the minimum age data.

  Each one quotes the store rule it rests on (Google Play Families policy; Apple App Review 1.3, 4.5.4 and 5.1.4; Chrome's userVisibleOnly requirement) and says where our default is stricter than the store. Age thresholds stay per-jurisdiction configuration with their sources.

• stacks/rust.md, two entries:

  • a sans-I/O client core for clients that share protocol and state rules across platforms;
  • UniFFI's own Swift and Kotlin generators over a Kotlin Multiplatform bridge, with an explicit cancel call because UniFFI does not cancel async calls itself.

• SKILL.md: a routing row for apps with child users.

• A minor changeset.

Why

These came out of an architecture review of a family messaging app with separate adult and child experiences. It has a Rust messaging core used from the web through WASM, with native apps planned. The child-app and push rules hold for any app with child users. The client-core entries hold for any product that shares protocol rules across web, iOS and Android.

Proof

• python3 scripts/lint.py: 0 errors. The 5 pending entries it reports are outside this change.

• Every source link was opened on 2026-10-23.

• Independent review by a different model family, over two rounds. Round 1 blocked on six points, all accepted:

  • a wrong reason against the Kotlin Multiplatform bridge (its generators are past 1.0);
  • store rules stated more broadly than the stores state them;
  • a diff rule that contradicted contracts.md;
  • age-threshold guidance that could promote a child early;
  • storage mechanisms promoted into a general default;
  • entries outside the standard shape.

  Round 2 approved once the quality lens became a standard entry, which the last commit does.