[ajianaz]

What

Item 8 of the round-2 deepening epic (#1523). The server, the CLI and MCP no longer reach past Uteke: Uteke::store() and Uteke::graph_store() are pub(crate); transports use count_active, list_deprecated (new, next to the existing count_deprecated) and graph() (the knowledge graph without the raw SQLite connection). The transport test suites seed rows through store_for_tests(), which exists only behind the test-support feature. The private module holding the Uteke room operations is renamed room_ops so it stops sharing a name with memory::rooms. The one-line room forwarders stay: they are the facade that keeps Store private (the issue text was corrected on this).

Why

Handlers that call uteke.store().count_active or build a GraphStore from the raw connection depend on the storage layer, so the seam leaks. The compiler now enforces it. Breaking for crates that depend on uteke-core and called the two accessors (noted in the CHANGELOG; 0.x minor). No wire or CLI behaviour change. Refs #1521.

Testing

• clippy -D warnings, fmt, cargo test --workspace pass; cargo run -p docgen changes nothing
• 2 new tests (lifecycle counts and the deprecated list through Uteke; the graph reachable through graph()); grep shows no store() / graph_store() call left in the server, CLI or MCP, and the compiler would reject one
• reviewed locally with the Cora MCP agent reviewer before pushing: no blocking finding (the deterministic notes are unwrap hints in test fixtures)

[github-actions]

🔍 Cora AI Code Review

✅ No issues found. Code looks good!

---

Review powered by cora-code · BYOK · MIT