feat: salsa integration (Phase 5) + dogfood tracking (#121)

* chore: track batch 2 features as rivet artifacts (FEAT-093 to FEAT-105)

13 features: rowan Phases 1-3, 6 domain schemas, MCP expansion,
schema validate, docs refresh, pre-commit hook.

573 total artifacts.

* feat(salsa): wire schema-driven rowan parser into salsa DB (Phase 5)

parse_artifacts_v2() tracked function uses extract_schema_driven()
from the rowan HIR layer. Schema is a transitive salsa dependency —
schema changes invalidate all artifact extraction, source changes
only re-extract that file.

Feature flag 'rowan-yaml' (default on). Debug builds log warnings
if old and new parsers produce different artifact IDs.

build_store() now takes schema_set parameter for the new code path.
All 7 db.store() call sites updated.

Author: avrabe

artifacts/v031-phase2-features.yaml

@@ -0,0 +1,157 @@
+# Features shipped in v0.3.1 second batch (2026-04-02)
+artifacts:
+  # ── Rowan parser phases ──────────────────────────────────────────────
+
+  - id: FEAT-093
+    type: feature
+    title: Rowan YAML CST parser (Phase 1)
+    status: approved
+    description: >
+      Lossless, span-preserving YAML parser using rowan. 28 SyntaxKind
+      variants, hand-written lexer and recursive-descent parser with
+      indent tracking. Round-trip guarantee for all project YAML files.
+    links:
+      - type: satisfies
+        target: REQ-002
+
+  - id: FEAT-094
+    type: feature
+    title: HIR extraction from rowan CST (Phase 2)
+    status: approved
+    description: >
+      Walks rowan CST to extract Vec<SpannedArtifact> with byte-accurate
+      spans for every field. Cross-validated against serde-based parser.
+    links:
+      - type: satisfies
+        target: REQ-002
+
+  - id: FEAT-095
+    type: feature
+    title: Schema-driven YAML extraction (Phase 3)
+    status: approved
+    description: >
+      Uses yaml-section and shorthand-links schema metadata to parse
+      section-based YAML formats. One function replaces the 861-line
+      hardcoded STPA adapter. Handles generic and section formats.
+    links:
+      - type: satisfies
+        target: REQ-002
+      - type: satisfies
+        target: REQ-010
+
+  # ── Domain schemas ─────────────────────────────────────────────────
+
+  - id: FEAT-096
+    type: feature
+    title: DO-178C airborne software schema
+    status: approved
+    description: >
+      14 artifact types covering PSAC through SAS with DAL-based
+      traceability rules for aviation software certification.
+    links:
+      - type: satisfies
+        target: REQ-010
+
+  - id: FEAT-097
+    type: feature
+    title: EN 50128 railway software safety schema
+    status: approved
+    description: >
+      14 artifact types with SIL-dependent rules, tool qualification,
+      independent assessment for railway software.
+    links:
+      - type: satisfies
+        target: REQ-010
+
+  - id: FEAT-098
+    type: feature
+    title: IEC 61508 functional safety schema
+    status: approved
+    description: >
+      15 artifact types covering the full IEC 61508 V-model with
+      SIL-based traceability and independent assessment rules.
+    links:
+      - type: satisfies
+        target: REQ-010
+
+  - id: FEAT-099
+    type: feature
+    title: IEC 62304 medical device software schema
+    status: approved
+    description: >
+      13 artifact types with class-conditional verification rules
+      for medical device software (Class A/B/C).
+    links:
+      - type: satisfies
+        target: REQ-010
+
+  - id: FEAT-100
+    type: feature
+    title: ISO/PAS 8800 AI safety lifecycle schema
+    status: approved
+    description: >
+      12 artifact types for AI element safety in road vehicles.
+      Bridges to STPA-AI for ML controller traceability.
+    links:
+      - type: satisfies
+        target: REQ-010
+
+  - id: FEAT-101
+    type: feature
+    title: ISO 21448 SOTIF schema
+    status: approved
+    description: >
+      8 artifact types covering functional insufficiency hazards,
+      triggering conditions, and known/unknown unsafe scenarios.
+    links:
+      - type: satisfies
+        target: REQ-010
+
+  # ── Other features ─────────────────────────────────────────────────
+
+  - id: FEAT-102
+    type: feature
+    title: MCP server expanded to 9 tools
+    status: approved
+    description: >
+      Added rivet_get, rivet_coverage, rivet_schema, rivet_embed,
+      rivet_snapshot_capture, rivet_add to the MCP server with
+      proper JSON Schema inputSchema definitions.
+    links:
+      - type: satisfies
+        target: REQ-001
+
+  - id: FEAT-103
+    type: feature
+    title: rivet schema validate command
+    status: approved
+    description: >
+      Validates loaded schemas are well-formed: checks link types
+      exist, target types exist, traceability rule references valid.
+      Found 3 real errors in AADL cross-schema references.
+    links:
+      - type: satisfies
+        target: REQ-010
+
+  - id: FEAT-104
+    type: feature
+    title: Documentation refresh with 7 new topics
+    status: approved
+    description: >
+      New docs: embed-syntax, schemas-overview, schema/eu-ai-act,
+      schema/safety-case, schema/stpa-ai, schema/stpa-sec,
+      schema/research. Updated CLI and YAML references.
+    links:
+      - type: satisfies
+        target: REQ-001
+
+  - id: FEAT-105
+    type: feature
+    title: Pre-commit hook for formatting and clippy
+    status: approved
+    description: >
+      scripts/pre-commit runs cargo fmt and clippy before every commit.
+      Prevents CI format failures.
+    links:
+      - type: satisfies
+        target: REQ-001

rivet-cli/src/main.rs

@@ -6540,7 +6540,7 @@ fn cmd_lsp(cli: &Cli) -> Result<bool> {
     };
 
     // Build supplementary state for rendering
-    let store = db.store(source_set);
+    let store = db.store(source_set, schema_set);
     let render_schema = db.schema(schema_set);
     let mut render_graph = rivet_core::links::LinkGraph::build(&store, &render_schema);
 
@@ -6617,7 +6617,7 @@ fn cmd_lsp(cli: &Cli) -> Result<bool> {
                 match method {
                     "textDocument/hover" => {
                         let params: HoverParams = serde_json::from_value(req.params.clone())?;
-                        let store = db.store(source_set);
+                        let store = db.store(source_set, schema_set);
                         let result = lsp_hover(&params, &store);
                         connection.sender.send(Message::Response(Response {
                             id: req.id,
@@ -6628,7 +6628,7 @@ fn cmd_lsp(cli: &Cli) -> Result<bool> {
                     "textDocument/definition" => {
                         let params: GotoDefinitionParams =
                             serde_json::from_value(req.params.clone())?;
-                        let store = db.store(source_set);
+                        let store = db.store(source_set, schema_set);
                         let result = lsp_goto_definition(&params, &store);
                         connection.sender.send(Message::Response(Response {
                             id: req.id,
@@ -6638,7 +6638,7 @@ fn cmd_lsp(cli: &Cli) -> Result<bool> {
                     }
                     "textDocument/completion" => {
                         let params: CompletionParams = serde_json::from_value(req.params.clone())?;
-                        let store = db.store(source_set);
+                        let store = db.store(source_set, schema_set);
                         let schema = db.schema(schema_set);
                         let result = lsp_completion(&params, &store, &schema);
                         connection.sender.send(Message::Response(Response {
@@ -6923,7 +6923,7 @@ fn cmd_lsp(cli: &Cli) -> Result<bool> {
                                 // and append document [[ID]] reference validation so
                                 // broken wiki-links in markdown files are reported.
                                 let mut new_diagnostics = db.diagnostics(source_set, schema_set);
-                                let new_store = db.store(source_set);
+                                let new_store = db.store(source_set, schema_set);
                                 new_diagnostics
                                     .extend(validate::validate_documents(&doc_store, &new_store));
                                 lsp_publish_salsa_diagnostics(
@@ -6939,7 +6939,7 @@ fn cmd_lsp(cli: &Cli) -> Result<bool> {
                                 );
 
                                 // Rebuild render state
-                                render_store = db.store(source_set);
+                                render_store = db.store(source_set, schema_set);
                                 let render_schema = db.schema(schema_set);
                                 render_graph = rivet_core::links::LinkGraph::build(
                                     &render_store,
@@ -6986,7 +6986,7 @@ fn cmd_lsp(cli: &Cli) -> Result<bool> {
                                         // including document [[ID]] reference validation.
                                         let mut diagnostics =
                                             db.diagnostics(source_set, schema_set);
-                                        let store = db.store(source_set);
+                                        let store = db.store(source_set, schema_set);
                                         diagnostics.extend(validate::validate_documents(
                                             &doc_store, &store,
                                         ));

rivet-core/Cargo.toml

@@ -9,7 +9,8 @@ rust-version.workspace = true
 
 
 [features]
-default = ["aadl"]
+default = ["aadl", "rowan-yaml"]
+rowan-yaml = []
 oslc = ["dep:reqwest", "dep:urlencoding"]
 wasm = ["dep:wasmtime", "dep:wasmtime-wasi"]
 aadl = ["dep:spar-hir", "dep:spar-analysis"]

rivet-core/src/db.rs

@@ -21,6 +21,8 @@ use crate::model::Artifact;
 use crate::schema::{Schema, SchemaFile};
 use crate::store::Store;
 use crate::validate::Diagnostic;
+#[cfg(feature = "rowan-yaml")]
+use crate::yaml_hir;
 
 // ── Salsa inputs ────────────────────────────────────────────────────────
 
@@ -109,6 +111,30 @@ pub fn parse_artifacts(db: &dyn salsa::Database, source: SourceFile) -> Vec<Arti
     }
 }
 
+/// Parse artifacts from a single source file using the schema-driven rowan parser.
+///
+/// Uses `yaml_hir::extract_schema_driven` which reads `yaml-section` metadata
+/// from the schema to discover sections and auto-convert shorthand links.
+///
+/// This is a salsa tracked function — results are memoized and only
+/// recomputed when the `SourceFile` content or `SchemaInputSet` changes.
+#[cfg(feature = "rowan-yaml")]
+#[salsa::tracked]
+pub fn parse_artifacts_v2(
+    db: &dyn salsa::Database,
+    source: SourceFile,
+    schema_set: SchemaInputSet,
+) -> Vec<Artifact> {
+    let content = source.content(db);
+    let path = source.path(db);
+    let source_path = std::path::Path::new(&path);
+
+    let schema = build_schema(db, schema_set);
+    let parsed = yaml_hir::extract_schema_driven(&content, &schema, Some(source_path));
+
+    parsed.artifacts.into_iter().map(|sa| sa.artifact).collect()
+}
+
 /// Collect parse errors from all source files as diagnostics.
 ///
 /// Each file that fails to parse produces a `yaml-parse-error` diagnostic
@@ -234,9 +260,9 @@ pub fn validate_all(
 ///
 /// Both `validate_all` and this function call `build_pipeline`, which is a
 /// plain (non-tracked) helper. The tracked functions that `build_pipeline`
-/// delegates to (`parse_artifacts`) are individually cached by salsa, so
-/// the repeated calls do NOT re-parse source files — only the lightweight
-/// store/schema assembly runs twice.
+/// delegates to (`parse_artifacts` / `parse_artifacts_v2`) are individually
+/// cached by salsa, so the repeated calls do NOT re-parse source files —
+/// only the lightweight store/schema assembly runs twice.
 #[salsa::tracked]
 pub fn evaluate_conditional_rules(
     db: &dyn salsa::Database,
@@ -280,18 +306,53 @@ fn build_pipeline(
     source_set: SourceFileSet,
     schema_set: SchemaInputSet,
 ) -> (Store, Schema, LinkGraph) {
-    let store = build_store(db, source_set);
+    let store = build_store(db, source_set, schema_set);
     let schema = build_schema(db, schema_set);
     let graph = LinkGraph::build(&store, &schema);
     (store, schema, graph)
 }
 
 /// Build an artifact `Store` from all source file inputs.
-fn build_store(db: &dyn salsa::Database, source_set: SourceFileSet) -> Store {
+///
+/// When the `rowan-yaml` feature is enabled, uses the schema-driven rowan
+/// parser (`parse_artifacts_v2`) which reads `yaml-section` metadata from
+/// the schema. In debug builds, both parsers run and their output is
+/// compared as a cross-check.
+fn build_store(
+    db: &dyn salsa::Database,
+    source_set: SourceFileSet,
+    schema_set: SchemaInputSet,
+) -> Store {
+    #[cfg(not(feature = "rowan-yaml"))]
+    let _ = schema_set;
+
     let sources = source_set.files(db);
     let mut store = Store::new();
     for source in sources {
-        for artifact in parse_artifacts(db, source) {
+        #[cfg(feature = "rowan-yaml")]
+        let artifacts = {
+            let new_arts = parse_artifacts_v2(db, source, schema_set);
+
+            #[cfg(debug_assertions)]
+            {
+                let old_arts = parse_artifacts(db, source);
+                let new_ids: Vec<&str> = new_arts.iter().map(|a| a.id.as_str()).collect();
+                let old_ids: Vec<&str> = old_arts.iter().map(|a| a.id.as_str()).collect();
+                if old_ids != new_ids {
+                    log::warn!(
+                        "parser mismatch for {}: old={old_ids:?} new={new_ids:?}",
+                        source.path(db),
+                    );
+                }
+            }
+
+            new_arts
+        };
+
+        #[cfg(not(feature = "rowan-yaml"))]
+        let artifacts = parse_artifacts(db, source);
+
+        for artifact in artifacts {
             // Use upsert to avoid panics on duplicate IDs across files.
             store.upsert(artifact);
         }
@@ -378,9 +439,9 @@ impl RivetDatabase {
         false
     }
 
-    /// Get the current store (computed from source inputs).
-    pub fn store(&self, source_set: SourceFileSet) -> Store {
-        build_store(self, source_set)
+    /// Get the current store (computed from source and schema inputs).
+    pub fn store(&self, source_set: SourceFileSet, schema_set: SchemaInputSet) -> Store {
+        build_store(self, source_set, schema_set)
     }
 
     /// Get the current merged schema (computed from schema inputs).
@@ -618,9 +679,10 @@ artifacts:
     fn adding_artifact_appears_in_store() {
         let mut db = RivetDatabase::new();
         let sources = db.load_sources(&[("reqs.yaml", SOURCE_REQ)]);
+        let schemas = db.load_schemas(&[("test", TEST_SCHEMA)]);
 
         // Initially: 1 artifact (REQ-001).
-        let store = db.store(sources);
+        let store = db.store(sources, schemas);
         assert_eq!(store.len(), 1);
         assert!(store.contains("REQ-001"));
 
@@ -636,7 +698,7 @@ artifacts:
 "#;
         db.update_source(sources, "reqs.yaml", combined.to_string());
 
-        let store = db.store(sources);
+        let store = db.store(sources, schemas);
         assert_eq!(store.len(), 2);
         assert!(store.contains("REQ-001"));
         assert!(store.contains("REQ-002"));