v0.18.4

What's Changed

• fix(mcp): inject server instructions into system prompt by @bug-ops in #2639
• fix(tools): utility gate explicit-request bypass and ML classifier false-positive by @bug-ops in #2638
• fix(tools): explicit_request check falls back to m.content when parts is empty by @bug-ops in #2643
• feat(tui): slash-command autocomplete in insert mode by @bug-ops in #2644
• fix(index): wire workspace_root through IndexConfig and optimize memory usage by @bug-ops in #2648
• feat(memory): persona memory layer and multi-agent consistency protocol by @bug-ops in #2649
• fix(index): offload directory walk to spawn_blocking to unblock TUI startup by @bug-ops in #2651
• fix(memory): wire persona extraction write-path into agent loop by @bug-ops in #2652
• fix(tui): start rendering immediately, defer agent wiring to phase 2 by @bug-ops in #2654
• fix(memory): await persona extraction on shutdown to prevent task cancellation by @bug-ops in #2655
• fix(tui): start TUI rendering before heavy agent initialization by @bug-ops in #2657
• fix(tui): clear status bar after init, switch tui_status! to send().await by @bug-ops in #2659
• fix(tui): fix status race condition and blocking skill hash loop by @bug-ops in #2661
• fix(memory): move embed_missing to background, add early Ctrl+C handler by @bug-ops in #2663
• perf(index): add memory batching and max_file_bytes to reduce peak RAM by @bug-ops in #2665
• perf(llm): add embed_batch() to LlmProvider, eliminate per-chunk sequential HTTP calls by @bug-ops in #2667
• fix(index): add embed_provider to IndexConfig to prevent guardrail contention by @bug-ops in #2669
• fix(index,memory,tui): fix 35GB RAM growth, guardrail false-positives, missing status indicators by @bug-ops in #2671
• fix(router): exclude embed=true providers from inference routing pool by @bug-ops in #2673
• fix(skills): replace panic in RoutingHead::update() with graceful no-op by @bug-ops in #2676
• fix(memory): guard upsert_persona_fact against stale source_conversation_id by @bug-ops in #2678
• fix(llm,index): embed retry/backoff on 429 + embed_concurrency cap (#2677, #2679) by @bug-ops in #2680
• fix(llm): use OLLAMA_CHAT_MODEL/OLLAMA_EMBED_MODEL env vars in ollama.rs integration tests by @bug-ops in #2682
• release: v0.18.4 by @bug-ops in #2683

Full Changelog: v0.18.3...v0.18.4