fix: preserve real prefix when middle ID segment is a reserved word

understand-anything-plugin/packages/core/src/__tests__/normalize-graph.test.ts

@@ -31,6 +31,38 @@ describe("normalizeNodeId", () => {
     ).toBe("file:src/foo.ts");
   });
 
+  it("keeps a real prefix when a different reserved word is a middle segment", () => {
+    // Regression: "endpoint:service:x" is a valid prefix followed by a real
+    // path segment that happens to be a reserved word. The outer "endpoint"
+    // prefix must be preserved, not dropped in favour of "service".
+    expect(
+      normalizeNodeId("endpoint:service:getUser", { type: "endpoint" }),
+    ).toBe("endpoint:service:getUser");
+  });
+
+  it("is idempotent for IDs whose middle segment is a reserved word", () => {
+    const once = normalizeNodeId("endpoint:service:getUser", {
+      type: "endpoint",
+    });
+    expect(normalizeNodeId(once, { type: "endpoint" })).toBe(once);
+  });
+
+  it("strips a project-name prefix that collides with a reserved word", () => {
+    // Regression: when the project name is itself a reserved word ("service")
+    // and the node's real prefix follows ("file"), the spurious outer prefix
+    // must be dropped so the canonical "file:src/foo.ts" form is used — not
+    // left as "service:file:src/foo.ts", which would dangle edges that
+    // reference the canonical ID.
+    expect(
+      normalizeNodeId("service:file:src/foo.ts", { type: "file" }),
+    ).toBe("file:src/foo.ts");
+  });
+
+  it("is idempotent when a reserved-word project prefix is stripped", () => {
+    const once = normalizeNodeId("service:file:src/foo.ts", { type: "file" });
+    expect(normalizeNodeId(once, { type: "file" })).toBe(once);
+  });
+
   it("strips project-name prefix when valid prefix follows", () => {
     expect(
       normalizeNodeId("my-project:file:src/foo.ts", { type: "file" }),
@@ -438,6 +470,48 @@ describe("normalizeBatchOutput", () => {
     expect(result.edges[0].source).toBe("file:src/bare.ts");
     expect(result.edges[0].target).toBe("file:src/target.ts");
   });
+
+  it("repairs an edge endpoint whose project prefix collides with a reserved word", () => {
+    // Regression: an edge endpoint "service:file:src/foo.ts" refers to the
+    // canonical node "file:src/foo.ts", but inferTypeFromId reads the spurious
+    // reserved-word project prefix "service" as the type. The fallback must
+    // still resolve it to the existing node rather than drop the edge.
+    const result = normalizeBatchOutput({
+      nodes: [
+        {
+          id: "file:src/foo.ts",
+          type: "file",
+          name: "foo.ts",
+          filePath: "src/foo.ts",
+          summary: "Target",
+          tags: [],
+          complexity: "simple",
+        },
+        {
+          id: "file:src/bar.ts",
+          type: "file",
+          name: "bar.ts",
+          filePath: "src/bar.ts",
+          summary: "Source",
+          tags: [],
+          complexity: "simple",
+        },
+      ],
+      edges: [
+        {
+          source: "file:src/bar.ts",
+          target: "service:file:src/foo.ts",
+          type: "imports",
+          direction: "forward",
+          weight: 0.7,
+        },
+      ],
+    });
+
+    expect(result.edges).toHaveLength(1);
+    expect(result.edges[0].target).toBe("file:src/foo.ts");
+    expect(result.stats.danglingEdgesDropped).toBe(0);
+  });
 });
 
 describe("normalizeBatchOutput integration", () => {

understand-anything-plugin/packages/core/src/analyzer/normalize-graph.ts

@@ -27,8 +27,17 @@ const TYPE_TO_PREFIX: Record<string, string> = {
 /**
  * Strips all non-valid prefixes from an ID, returning the bare path
  * and the first valid prefix found (if any).
+ *
+ * `expectedPrefix` is the canonical prefix for the node's declared type
+ * (e.g. "file" for a file node). It disambiguates a reserved word that
+ * appears before the expected prefix — a spurious project-name prefix that
+ * happens to collide with a reserved word — from a reserved word that is a
+ * legitimate middle path segment.
  */
-function stripToValidPrefix(id: string): { prefix: string | null; path: string } {
+function stripToValidPrefix(
+  id: string,
+  expectedPrefix?: string,
+): { prefix: string | null; path: string } {
   let remaining = id;
 
   // Peel off colon-separated segments until we find a valid prefix or run out
@@ -38,11 +47,22 @@ function stripToValidPrefix(id: string): { prefix: string | null; path: string }
 
     const segment = remaining.slice(0, colonIdx);
     if (VALID_PREFIXES.has(segment)) {
-      // Check for double valid prefix (e.g., "file:file:src/foo.ts")
+      // Collapse the outer prefix only when the next segment is either:
+      //   - the SAME reserved word — a true duplicate ("file:file:src/foo.ts"), or
+      //   - the node's expected prefix — a spurious project-name prefix that
+      //     collides with a reserved word ("service:file:src/foo.ts" for a file
+      //     node), which must resolve to the canonical "file:src/foo.ts".
+      // A different reserved word that is NOT the expected prefix
+      // ("endpoint:service:x" for an endpoint node) is a real path segment and
+      // must be preserved.
       const rest = remaining.slice(colonIdx + 1);
       const innerColonIdx = rest.indexOf(":");
-      if (innerColonIdx > 0 && VALID_PREFIXES.has(rest.slice(0, innerColonIdx))) {
-        // Double-prefixed — skip the outer, recurse on inner
+      const innerSegment = innerColonIdx > 0 ? rest.slice(0, innerColonIdx) : "";
+      if (
+        innerColonIdx > 0 &&
+        (innerSegment === segment || innerSegment === expectedPrefix)
+      ) {
+        // Skip the outer prefix, recurse on the inner one
         remaining = rest;
         continue;
       }
@@ -69,7 +89,7 @@ export function normalizeNodeId(
   if (!trimmed) return trimmed;
 
   const expectedPrefix = TYPE_TO_PREFIX[node.type];
-  const { prefix, path } = stripToValidPrefix(trimmed);
+  const { prefix, path } = stripToValidPrefix(trimmed, expectedPrefix);
 
   if (prefix) {
     // For step nodes with filePath, reconstruct as step:flowSlug:filePath:stepSlug.
@@ -191,6 +211,40 @@ function inferTypeFromId(id: string): string {
   return "file";
 }
 
+/**
+ * Best-effort repair of an edge endpoint that matches no node ID.
+ *
+ * Tries the prefix-inferred type first (preserving the common case), then
+ * each subsequent leading reserved-word segment as a candidate type. This
+ * recovers a reserved-word project prefix — e.g. an edge endpoint
+ * `service:file:src/foo.ts` pointing at the canonical node `file:src/foo.ts`,
+ * where `inferTypeFromId` would treat the spurious `service` as the type and
+ * fail to strip it, leaving the edge dangling. Returns the original id
+ * unchanged when nothing resolves to an existing node.
+ */
+function resolveEdgeEndpoint(id: string, validNodeIds: Set<string>): string {
+  const candidateTypes: string[] = [inferTypeFromId(id)];
+
+  // Add each leading valid-prefix segment's type as an additional candidate,
+  // so a spurious outer reserved word can be skipped in favour of the real one.
+  let rest = id;
+  while (true) {
+    const colonIdx = rest.indexOf(":");
+    if (colonIdx <= 0) break;
+    const segment = rest.slice(0, colonIdx);
+    if (!(segment in PREFIX_TO_TYPE)) break;
+    const type = PREFIX_TO_TYPE[segment];
+    if (!candidateTypes.includes(type)) candidateTypes.push(type);
+    rest = rest.slice(colonIdx + 1);
+  }
+
+  for (const type of candidateTypes) {
+    const normalized = normalizeNodeId(id, { type });
+    if (validNodeIds.has(normalized)) return normalized;
+  }
+  return id;
+}
+
 /**
  * Normalizes a merged batch output: fixes node IDs and numeric complexity,
  * rewrites edge references, deduplicates nodes and edges, and drops dangling edges.
@@ -280,18 +334,14 @@ export function normalizeBatchOutput(data: {
     let newSource = idMap.get(oldSource) ?? oldSource;
     let newTarget = idMap.get(oldTarget) ?? oldTarget;
 
-    // Fallback: if endpoint not found in idMap, normalize it directly
-    // (handles cross-variant malformed IDs between nodes and edges).
-    // Try the edge's implied type first (from prefix), then fall back to "file".
+    // Fallback: if an endpoint isn't found in idMap, repair it directly
+    // (handles cross-variant malformed IDs between nodes and edges, including
+    // reserved-word project prefixes that inferTypeFromId alone can't resolve).
     if (!validNodeIds.has(newSource)) {
-      const inferredType = inferTypeFromId(newSource);
-      const normalized = normalizeNodeId(newSource, { type: inferredType });
-      if (validNodeIds.has(normalized)) newSource = normalized;
+      newSource = resolveEdgeEndpoint(newSource, validNodeIds);
     }
     if (!validNodeIds.has(newTarget)) {
-      const inferredType = inferTypeFromId(newTarget);
-      const normalized = normalizeNodeId(newTarget, { type: inferredType });
-      if (validNodeIds.has(normalized)) newTarget = normalized;
+      newTarget = resolveEdgeEndpoint(newTarget, validNodeIds);
     }
 
     if (newSource !== oldSource || newTarget !== oldTarget) {