Code Brain v0.2.0
Code Brain v0.2.0
Minor release — new retrieval/memory/eval features land since v0.1.3.
Highlights
- Graft 3 retrieval/memory/eval upgrades
self_improve↔eval_looppass-rate fitness (rule retention gates on eval pass-rate, not token regression alone)- KV-cache-stable
context_budgetprefix ordering (byte-stable cacheable prefix → higher prompt-cache hit rate) memory_tierpage-in: offline sleep-time HOT consolidation (memory_hot.py+ai memory page-in+ sleep-time job)
- Re-inject Code Brain context on /clear and compact (SessionStart hook matcher fix)
- CI/infra hardening: Windows preflight, cancellable CI, managed-merge dedup, detached-smoke focus, fake-worker decoupling, lean global kit
Verification
- 1081 tests passed / 4 skipped
- All 23 doctor checks green
- Invariants held: no network on hot paths, stdlib-only, fail-soft
Full changelog: v0.1.3...v0.2.0