archgraph-argo 0.24.1 → 0.24.3

package/argo/rules/archgraph.instructions.md

@@ -25,6 +25,7 @@ Non-negotiable red lines (MUST). Never skip, simplify, or silently violate them;
 9. Never duplicate: reuse is the default; a create blocked as an exact or semantic duplicate must be reused, or explicitly overridden with `onConflict: "allowDuplicate"` + justification. See `<GraphDeduplication>`.
 10. Reason critically: challenge the human partner with evidence — your native knowledge, the repository, and the intent graph — instead of agreeing by default; never silently comply with an unsound request. See `<CriticalReasoningGuideline>`.
 11. Never lose content silently: writes that reduce existing content must be lossless (merge/delta) or explicitly acknowledged (`acknowledgeLoss`), and destructive removals are tombstoned. See `<LosslessWrite>`.
+12. Federate by reference, never by copy: register/discover/authorize/read through the federation center (graph MCP surface); a read returns a reference, not a copy; cross-member access is denied by default. See `<FederationGuideline>`.
 </CoreRules>
 
 <Ontology>
@@ -170,6 +171,17 @@ Use `queryNeo4jGraph` for structural/type lookups. It never mutates the canonica
 3. Semantic/context reads (`getSystemArchitecture`, `getIntentElementContext`, `getArchitectureViewContext`) are the PRIORITY path; Cypher is the SECONDARY path per `<QueryPriorityGuideline>`.
 </GraphQueryGuideline>
 
+<FederationGuideline>
+When your project participates in a federation, act through the federation center (one central platform reached over the graph MCP):
+1. Register or exit: register your project with the center to join; deregister it to leave. Do this as the project's own Agent together with your human partner — no separate external approval is required.
+2. Discover before relying: query the center for the members that exist and what each offers (purpose, capabilities, open interfaces).
+3. Authorize before you read: read another member's opened content only after that member explicitly authorizes you; access is denied by default, assume no trust between members.
+4. Reference, not a copy: a read returns a reference, not a copy of another member's content; never persist a copy. The center holds only federation metadata (membership, interfaces, grants); each member's own graph stays the single source of truth.
+5. Never merge graphs: members stay independent; keep your own graph sovereign.
+6. One center: use the single federation center; if it is unreachable, fall back to locally cached discovery/grants and say so.
+7. Surface: do register / deregister / discover / authorize / read through the center's graph MCP tools; never reach the center by a file or SQL path.
+</FederationGuideline>
+
 <Attention>
 Confirm the ARGO MCP server is serving this repository's graph (design/KG/SystemArchitecture.json), not another graph. If it is not, stop and report to your human partner before doing anything else.
 </Attention>

package/argo/scripts/graph-rag/defaultSemanticRetrieval.js

@@ -85,32 +85,35 @@ const CHANNELS = Object.freeze([
     indexName: 'argo_production_semantic_view_vector',
   }),
 ]);
+// Project an explicit metadata allowlist instead of properties(node): Neo4j map
+// projection `.*` lets an existing property win, so properties(node) / node{.*, vector:null}
+// would still ship the 1536-dim vector; the read path only consumes canonicalIdentity, channel, searchText and score.
 const VECTOR_QUERY_CYPHER = [
   'CALL db.index.vector.queryNodes($indexName, $topK, $vector)',
   'YIELD node, score',
   'WHERE node.channel = $channel',
-  'RETURN properties(node) AS record, score',
+  'RETURN {canonicalIdentity: node.canonicalIdentity, channel: node.channel, searchText: node.searchText, dimensions: node.dimensions, provider: node.provider, model: node.model, modelVersion: node.modelVersion, canonicalVersion: node.canonicalVersion, contentVersion: node.contentVersion, indexVersion: node.indexVersion} AS record, score',
   'ORDER BY score DESC',
 ].join('\n');
 const VECTOR_QUERY_CYPHER_SCOPED = [
   'CALL db.index.vector.queryNodes($indexName, $topK, $vector)',
   'YIELD node, score',
   'WHERE node.channel = $channel AND node.canonicalIdentity IN $canonicalIdentities',
-  'RETURN properties(node) AS record, score',
+  'RETURN {canonicalIdentity: node.canonicalIdentity, channel: node.channel, searchText: node.searchText, dimensions: node.dimensions, provider: node.provider, model: node.model, modelVersion: node.modelVersion, canonicalVersion: node.canonicalVersion, contentVersion: node.contentVersion, indexVersion: node.indexVersion} AS record, score',
   'ORDER BY score DESC',
 ].join('\n');
 const LEXICAL_QUERY_CYPHER = [
   'CALL db.index.fulltext.queryNodes($indexName, $queryText, { limit: $topK })',
   'YIELD node, score',
   'WHERE node.channel = $channel',
-  'RETURN properties(node) AS record, score',
+  'RETURN {canonicalIdentity: node.canonicalIdentity, channel: node.channel, searchText: node.searchText, dimensions: node.dimensions, provider: node.provider, model: node.model, modelVersion: node.modelVersion, canonicalVersion: node.canonicalVersion, contentVersion: node.contentVersion, indexVersion: node.indexVersion} AS record, score',
   'ORDER BY score DESC',
 ].join('\n');
 const LEXICAL_QUERY_CYPHER_SCOPED = [
   'CALL db.index.fulltext.queryNodes($indexName, $queryText, { limit: $topK })',
   'YIELD node, score',
   'WHERE node.channel = $channel AND node.canonicalIdentity IN $canonicalIdentities',
-  'RETURN properties(node) AS record, score',
+  'RETURN {canonicalIdentity: node.canonicalIdentity, channel: node.channel, searchText: node.searchText, dimensions: node.dimensions, provider: node.provider, model: node.model, modelVersion: node.modelVersion, canonicalVersion: node.canonicalVersion, contentVersion: node.contentVersion, indexVersion: node.indexVersion} AS record, score',
   'ORDER BY score DESC',
 ].join('\n');
 const READINESS_QUERY_CYPHER = [
@@ -149,6 +152,12 @@ function resolveTopK() {
   const k = envNumber('ARGO_SEMANTIC_TOP_K', DEFAULT_TOP_K);
   return Number.isInteger(k) && k > 0 ? k : DEFAULT_TOP_K;
 }
+// The single production seed gate: a retrieved record becomes a seed only when
+// its TRUE vector similarity reaches the purpose-aware threshold. Exposed so
+// acceptance tests exercise the exact predicate the retrieval runtime uses.
+function isSeedAboveThreshold(score, threshold) {
+  return Number.isFinite(score) && Number.isFinite(threshold) && score >= threshold;
+}
 const testCompositionStorage = new AsyncLocalStorage();
 
 function createDefaultSemanticRetrieval(dependencies = {}) {
@@ -828,7 +837,7 @@ async function exhaustChannel({
     for (const raw of newlyVisible) {
       if (accepted.length >= effectiveMax) break;
       const record = normalizeVectorRecord(raw, channel);
-      if (record && record.score >= effectiveThreshold && !seen.has(record.id)) {
+      if (record && isSeedAboveThreshold(record.score, effectiveThreshold) && !seen.has(record.id)) {
         seen.add(record.id);
         accepted.push(Object.freeze(record));
       }
@@ -1238,8 +1247,11 @@ module.exports = {
   memoryThresholdFor,
   auditThresholdFor,
   resolveTopK,
+  isSeedAboveThreshold,
   scopeCanonicalIdentitiesForChannel,
   bareCanonicalId,
   normalizeVectorRecord,
+  exhaustChannel,
+  exhaustLexicalChannel,
   AUDIT_PURPOSES,
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "archgraph-argo",
-  "version": "0.24.1",
+  "version": "0.24.3",
   "description": "Deploy the ArchGraph ARGO toolchain, skills, and rules (schema, scripts, argo-init skill, global rule) with one command.",
   "license": "MIT",
   "bin": {