Review: PR-A2 — SpEL/constant-ref routes + Route MCP tools

Verdict: Approved ✅

Clean, surgical implementation that finishes B2a per plan. Three-strategy ladder, kuzu helpers, and MCP wiring all match the proposal; scope discipline is excellent (zero leak into B2b/A3/brownfield/schema/ontology).

Scope discipline (out-of-scope checks)

Three-strategy ladder

ast_java.py route extraction now returns tuple[str, str, float, bool] (path, strategy, confidence, resolved) flowing through every emission site. Confidences match the locked plan (literal=1.0, SpEL=0.85, constant_ref=0.7).

The new _merge_http_route_with_class_base helper handles class+method composition correctly, including the subtle edge case where ${…} inside a non-string-literal stays constant_ref (not promoted to SpEL). Stickiness is right — string concat with a constant ref → constant_ref, never accidentally upgraded.

_route_id path_raw parameter — nice catch

Adding path_raw as a positional disambiguator prevents id collision when SpEL/const routes share an empty path_template. Backward compat is maintained via the PR-A1 test 11 update (still green). Comment in code explains intent clearly.

routes_skipped_unresolved semantics

Correctly narrowed — now only counts AST-level non-extractable cases (topicPattern, bindings). SpEL/constant_ref are emitted as unresolved Route nodes instead of dropped. Cases 8 & 9 were updated accordingly (previously asserted "skipped", now assert "emitted as unresolved") — correct migration.

Kuzu helpers (kuzu_queries.py)

• list_routes — sensible Cypher, all four filters (microservice / framework / path_prefix / method) AND-combined, limit clamped 1–500.
• find_route_handlers — returns list (correct for Feign arrays + class-level @RequestMapping arrays); includes confidence + strategy per handler.
• get_route_by_path — filters on path_template (the normalized form) which is consistent with what list_routes returns; microservice-scoped.

MCP tool wiring (server.py:933, 961, 985)

All three tools use asyncio.to_thread for sync graph calls, Pydantic Field validation, and proper DTO mapping (_route_dict_to_dto, RouteHandlerEntryDto). Tool descriptions correctly warn that SpEL-only routes have empty path_template and steer users to path_prefix for those. _INSTRUCTIONS updates (lines 53–55) are honest about deferred caller edges.

Tests

177 passed, 3 skipped

+7 new vs PR-A1 baseline (170). All planned tests landed:

• test_route_extraction.py::test_case12_get_mapping_spel_path — SpEL path
• test_case13_get_mapping_constant_ref — constant ref
• test_case14_request_mapping_string_concat_is_constant_ref — concat stickiness
• test_kuzu_queries.py::test_list_routes_filter_by_framework
• test_find_route_handlers_feign_array — uses new FeignTripleDup.java fixture
• test_get_route_by_path_microservice_isolated

Manual evidence reproduced

Ran python build_ast_graph.py --source-root tests/bank-chat-system --kuzu-path /tmp/check_kuzu_a2 --verbose:

[pass4] Route extraction: emitted=11, exposes=11, skipped_unresolved=0,
        routes_resolved_pct=81.8, by_framework={'spring_mvc': 9, 'kafka': 2}

✅ Matches your claim exactly.

Minor observations (non-blocking)

1. find_route_handlers returns plural by design — please ensure the B2b prompt expects an array when wiring (caller)-[:HTTP_CALLS]->(Route)<-[:EXPOSES]-(handler) traversal (multiple handlers per Route id is the common case for class-level mappings).
2. get_route_by_path requires microservice (no global lookup) — correct for multi-repo monorepos but worth noting in B2b prompt if a caller search needs to fall back across services.
3. _INSTRUCTIONS is getting long — fine for now, but consider sectioning by capability if you add 2–3 more tools in B2b.

Plan deltas needed

None. This PR matches PLAN-TIER1-COMPLETION.md §B2a precisely. Ready to merge.

Next up: PR-A3 (path-template registry + microservice scoping for routes) per the plan, then PR-A4/A5 (B2b: HTTP_CALLS edges + find_route_callers).