Session 249

Prose pending.

Purpose

Wire CodebaseTools (Prolog codebase-intelligence) to volks + production KB lifecycle, per reports/proposal-prolog-next-impact.md §4

Edit 1: AGENT_TOOLS volk exposure
- tool_registry.py AGENT_TOOLS: added codebase_impact/codebase_module_deps/codebase_find_cycles/codebase_dep_path to U:Ar, U:Cd, U:SD
- codebase_impact/codebase_module_deps only to U:BA (analysis needs blast-radius/dep-shape, not cycles/dep-path)
- U:Cp untouched (stays empty)
Edit 2: production KB lifecycle
- codebase_query_tools.py: new _build_production_kb() resolves ORCHESTRATA_CODEBASE_KB_DIR/ORCHESTRATA_AST_SNAPSHOT (default .claude/run/codebase_kb, plans/ast-ultimate.json, project-root-anchored)
- mtime-based attach-vs-reindex; missing snapshot / failed reindex raises RuntimeError caught by _run_query -> structured ToolResult error, never a crash
Wiring seam for tests + future daemon
- register_native_tools()/ToolExecutor gained codebase_kb/codebase_kb_environ passthrough params (default None = real production lifecycle)
- BaseAgent gained _codebase_kb_environ() override seam (mirrors _get_workspace) so the real tool loop can be pointed at a fixture KB in tests without touching real env

src/orchestrata_ultimate/agents/tool_registry.py (2053 lines) - prolog-codeintel volk wiring
src/orchestrata_ultimate/tools/codebase_query_tools.py (456 lines) - prolog-codeintel volk wiring
src/orchestrata_ultimate/agents/volks/base.py (557 lines) - prolog-codeintel volk wiring
tests/tools/test_codebase_query_tools.py (540 lines) - prolog-codeintel volk wiring
tests/agents/test_tool_registry.py (1146 lines) - prolog-codeintel volk wiring
tests/agents/volks/test_base_tool_loop.py (530 lines) - prolog-codeintel volk wiring
tests/agents/volks/codedirector/test_volk.py (964 lines) - prolog-codeintel volk wiring

Split utilization-plan P1: proof-tree explanations (#3 in ranked shortlist) is the next Prolog item; the 'KG deduction for speed' half of utilization-plan #1 is dropped as falsified: reports/proposal-prolog-next-impact.md §3.1: the benchmark showed Prolog LOSES to warm in-memory Python recursion on every workload (W1 ~370x, W3 1.7x, W4 3.3x) -- the only real win is a persisted/warm in-process engine for interactive queries, which the now-wired CodebaseTools already realizes. Backlog entry per report §4 follow-up item.