Roadmap

Current Status

IntentWeave ships three composable layers of code intelligence. Layer 1 (CARI) is production-ready with 1532 tests passing across 80 test files and 16 packages. Layer 2 (Selective Enrichment) is shipped — iw index enrich is live. Layer 3 (Intent Verification) is shipped — iw intent check runs all three enforcement domains (structural / behavioral / documentary) at $0 after a one-time LLM rule-extraction step.

Architecture

Layer 3 — Intent Verification          (shipped)
  iw intent check: structural + behavioral + documentary domains
  Insights Book: 15+ chapter HTML deliverable
Layer 2 — Selective Semantic Enrichment (shipped)
  iw index enrich — budget-controlled LLM on CARI-selected targets
Layer 1 — CARI                         (production-ready, $0)
  AST + keywords + git + SQLite → 60+ query modes

Recently Shipped

Dependency Maintenance & Rust Indexer Design (v0.13.0)

✅ Rust Indexer Port design analysis — full architecture plan for packages/cari-native/ Rust binary. Benchmarked: KWG = 69% of 47 s build on 595 files. Phase R1 targets 10–20× speedup (47 s → 3–5 s). Phase R2: NAPI-RS warm incremental daemon. Library choices finalised: oxc_parser, pulldown-cmark, rayon, rusqlite, gix.
✅ TypeScript 5.9 — upgraded from ^5.6.0
✅ Prettier 3.8, Turbo 2.9, Vitest 2.1.9 — toolchain upgrades
✅ All 16 packages at 0.13.0

Intent Engine — Three-Domain Enforcement (v0.12.0)

✅ iw intent check --domain structural/behavioral/documentary/all — single-pass enforcement across all three domains
✅ Behavioral domain — Mermaid-based rules in rules.yaml: sequenceDiagram, stateDiagram-v2, flowchart. Zero-dep custom parser, $0 enforcement
✅ Documentary domain — four built-in CARI-backed checks (doc.coverage.low, doc.terminology, doc.orphaned-section, doc.completeness.low) with no configuration required
✅ .iw/config.yaml — per-domain CI thresholds; promote warn → error per domain

Insights Book — 15+ Chapter HTML Deliverable (v0.12.0)

✅ iw index export --book — self-contained HTML with Executive Summary, Recommendations, Rules Catalog, Layer Architecture, Documentation & Source, Architecture (D3), Code Structure, Code Health, Violations, Coverage, Living Score, Priority Files, Tech Debt, Test Coverage, Call Graph, per-ADR Cytoscape.js chapters
✅ Layer Sankey SVG — two-column bezier flow visualization in Coverage chapter
✅ Cross-chapter navigation — domain pills and “Go to chapter” buttons

Call Graph (v0.12.0)

✅ symbol_calls / property_accesses — full call-graph extraction from AST
✅ iw index calls — query call edges by caller file or callee name
✅ iw index trace — BFS call-path tracing (forward and backward)
✅ iw index rule-coverage — flag packages with zero behavioral rules
✅ cari_calls / cari_trace MCP tools

Rule System Expansion (v0.12.0)

✅ type: variable_assignment, type: cypher, type: property_chain_length
✅ --baseline regression gating (13.5), import_pattern: "**" (13.6)
✅ symbol_name scope modifier, context_import, except_symbol, count_mode: per_file
✅ Import violation line numbers

Signal-Layer Checks (v0.12.0)

✅ @deprecated caller detection, @internal / _ enforcement
✅ as any inventory, naming conventions, comment-to-code ratio
✅ Decorator-derived layer assignment, ADR conformance trend, test description alignment

Prescriptive Architecture (§17, v0.11.0)

✅ allowed: entries in rules.yaml with edge-level rationale
✅ Layer geometry rendering for SVG layout
✅ Rule-expressed element overlay with violation heat-map
✅ ASCII conformance diagram in iw intent check output
✅ LLM-assisted spec synthesis from ADR prose

Plugin Architecture (v0.9.0)

✅ Plugin system — PluginRegistry with auto-discovery, 3 capability types (LLM, persistence, language)
✅ CypherLite — zero-dependency Cypher→SQL transpiler for SQLite-backed KG queries
✅ plugin-kg-lite — lightweight KG backend using CypherLite + SQLite (no Neo4j needed)
✅ plugin-kg — full Neo4j KG backend via PersistenceCapability
✅ plugin-swift / plugin-python — tree-sitter Swift and Python language plugins
✅ plugin-llm — OpenAI-based LLM capability

Architecture Visualization (5.x / 10.x)

✅ Multi-view community detection — structural, semantic, temporal modes
✅ Vertical slice detection and hierarchical sub-layering
✅ Architecture report — D3-powered HTML with Layers, Communities, Dependencies views
✅ Focused architecture view — Graphviz WASM report centered on a target entity
✅ LLM layer naming — descriptive layer and directory names via OpenAI
✅ Layer inference + layer check (5.1a/b)

CARI Core

✅ 60+ query modes: retrieve, connections, check, report, clones, structural clones, circular imports, unused exports, hotspot priority, TODOs, module coverage, orphaned sections, doc completeness, cross-group drift, mentions, annotations, test coverage, hubs, communities, surprises, rationale, terminology, dep depth, boundary violations, layers, focus, arch-check, living score, calls, trace, rule coverage, and more
✅ 58 MCP tools (6 KG + 52 CARI/intent) for GitHub Copilot
✅ CariIndex facade, Entity Bridge, Library API

CI & Automation

✅ Watch mode — iw index watch continuous re-indexing on file changes
✅ CI GitHub Action — uses: intentweave/doc-health-action@v1
✅ Git hooks — iw hook install/uninstall/status
✅ REST API v1.0.0 with OpenAPI/Swagger UI

Next Up

Language Support

✅ Swift AST extraction — @intentweave/swift-parser + @intentweave/plugin-swift
✅ Python AST extraction — @intentweave/python-parser + @intentweave/plugin-python
Planned: Go AST extraction (tree-sitter)
Planned: Rust AST extraction (tree-sitter)
Planned: Generic keyword-only fallback for unsupported extensions

Selective Enrichment — Remaining Use Cases

Use Case	Status	Notes
Completion backfill	Planned	Requires generating new doc content, not just extracting triples
Architecture narrative	Planned	`iw index narrative` — LLM prose from layer/community data

Semantic Clone Detection (Type 3)

LLM-powered similarity scoring to catch behaviourally equivalent but structurally different snippets. Depends on enrichment pipeline.

Decision Lifecycle Tracking

Track ADRs through state changes (proposed → accepted → deprecated) and detect unimplemented decisions.

Performance — Rust Indexer (Phase R1)

Replace the CARI build pipeline (AX + KWG stages) with a native Rust binary that writes the same SQLite schema. The 57 TypeScript query files are unchanged.

Design complete (v0.13.0): benchmarks measured, library choices finalised, schema compatibility contract defined.

Phase	Scope	Target speedup
R1-a/b	KWG: Markdown + co-occurrence	8–15× KWG (32 s → 2–4 s)
R1-c	AX: `oxc_parser` + tree-sitter	10–20× AX (11 s → 0.5–1 s)
R1-d	TCG: `gix` crate	1–2× (low priority)
R1-f	CLI shim + `cargo-dist` packaging	—
R1 total	Full build binary	10–20× (47 s → 3–5 s)

Phase R2: NAPI-RS bridge for warm incremental daemon (optional, later).

Ecosystem Integrations

Native plugins and exports for Docusaurus, Starlight, and Obsidian.

Developer Experience

VS Code extension with inline drift warnings (planned)
Drift alerts pushed to editor/Slack when iw index watch detects doc-breaking changes

Contributing

Have ideas? See Contributing or open a GitHub Discussion.