feat(types): hybrid type resolution — closes #13

[Wolfvin]

Summary

Phase 1 Step 3 (final Phase 1 issue): import-aware CALLS edge refinement.

Adds a post-AST-pass type resolution layer (scripts/hybrid_type_resolver.py) that uses a per-file import registry to refine CALLS edges. Previously user.profile.update() was recorded as a call to update with no target type. Now the receiver type is resolved via the import registry, and the CALLS edge's target_id is refined to the correct target node.

Files

Created (5):

• scripts/hybrid_type_resolver.py (1383 lines) — import registry builder, receiver type resolver, CALLS edge refiner
• scripts/commands/resolve_types.py (94 lines) — new CLI command (auto-registered)
• tests/test_hybrid_type_resolver.py (601 lines) — 25 tests
• tests/fixtures/type_resolution/{main,models,cache}.py — synthetic fixture for disambiguation testing

Modified (7):

• scripts/commands/scan.py — calls refine_call_edges after graph population; adds type_resolution field to output
• scripts/graph_model.py — graph_stats() reports IMPORTS edges
• scripts/mcp_server.py — removed manual format field from architecture tool (was blocking compact enum injection, caused test regression)
• tests/test_graph_model.py — updated assertions for IMPORTS edges
• tests/test_compact_format.py — updated assertions for IMPORTS edges
• README.md — added resolve-types to Architecture command table
• SKILL-QUICK.md — added resolve-types, updated total commands 58→60, MCP tools 56→58
• CHANGELOG.md — added Hybrid Type Resolution (issue [FEATURE] Hybrid type resolution: import-aware call edge refinement #13) section under [8.2.0]

Test Results

• New tests (test_hybrid_type_resolver.py): 25/25 pass in 0.59s
• Full suite (excluding test_integration.py OOM): 4 pre-existing failures only (test_hybrid_engine.py confidence fields — present on main before any Phase 1 work, NOT caused by this PR)
• BOS fix verified: test_compact_format.py::TestMCPGraphSchemaTool::test_all_tools_have_format_enum regression fixed by removing manual format field from architecture tool definition.

Functional Verification (by BOS, independent)

• scan on clean_app: type_resolution: {edges_refined: 11, edges_unresolved: 55}
• resolve-types standalone: returns same stats + import_registry_size: 37
• IMPORTS edges in graph: 37 (from 37 import statements parsed)
• import_registry table: 37 rows
• Synthetic fixture (tests/fixtures/type_resolution/): user.profile.update() correctly refines to Profile.update even when Cache.update competitor exists (without type resolution, edge_resolver picks Cache.update alphabetically)

Non-Breaking

• Type resolution is best-effort: unresolvable edges left unchanged with resolution_attempted flag. No CALLS edge ever deleted.
• import_registry table is additive — no existing table modified.
• 11/97 CALLS edges refined on clean_app; 55 unresolved (self-referential or std-lib calls that don't need refinement).

Closes #13

Phase 1 (#21) checklist status after this PR:

• [ARCH] Replace flat registry with true graph data model (nodes + edges) #8 — True graph data model
• [FEATURE] get_architecture — single-call codebase overview for agents #19 — get_architecture
• [PERF] Token-efficient output: structured compact format for all MCP tools #17 — Token-efficient compact output
• [FEATURE] Hybrid type resolution: import-aware call edge refinement #13 — Hybrid type resolution (this PR)
• [FEATURE] Git-diff aware incremental re-index (replace file-watcher polling) #14 — Git-diff aware incremental re-index
• vuln DB staleness fix (no separate issue filed)

Follow-up #25 still open: incremental scan does not populate graph tables.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[sonarqubecloud]

Quality Gate failed

Failed conditions
6 Security Hotspots

See analysis details on SonarQube Cloud