diff --git a/src/steps/lib/graph/get-ontology.ts b/src/steps/lib/graph/get-ontology.ts
index 744ae89..b9a63e9 100644
--- a/src/steps/lib/graph/get-ontology.ts
+++ b/src/steps/lib/graph/get-ontology.ts
@@ -53,7 +53,7 @@ function buildOntologyPayload(schemaData: any, includeEdges: boolean, includeAtt
 export default defineStep({
   type: "graph/get-ontology",
   description:
-    "Fetch the ontology of the strut knowledge graph: node types grouped by domain " +
+    "Fetch the ontology of the knowledge graph: node types grouped by domain " +
     "and the canonical list of valid `domains`. " +
     "Call this once before graph_graph_search to discover valid values for both the `type` and `domains` parameters. " +
     "Node types are grouped by domain key in `node_types[<domain>]`; types with no domain land in the `ungrouped` bucket. " +
diff --git a/src/steps/lib/graph/graph-get.ts b/src/steps/lib/graph/graph-get.ts
index c0fb79c..72d9dd3 100644
--- a/src/steps/lib/graph/graph-get.ts
+++ b/src/steps/lib/graph/graph-get.ts
@@ -9,7 +9,7 @@ const CHILD_DESCRIPTION_MAX = 300;
 export default defineStep({
   type: "graph/graph-get",
   description:
-    "Resolve a single node in the strut knowledge graph to its full content by ref_id. " +
+    "Resolve a single node in the knowledge graph to its full content by ref_id. " +
     "Use the ref_id from graph_graph_search or graph_graph_neighbors results. " +
     "When you know a node's NAME instead, pass node_type + name (e.g. node_type \"Concept\", " +
     "name \"Janitor\"): an exact lookup by the node's key, never a search — for types keyed by name. " +
diff --git a/src/steps/lib/graph/graph-neighbors.ts b/src/steps/lib/graph/graph-neighbors.ts
index 2486ef6..977ae7a 100644
--- a/src/steps/lib/graph/graph-neighbors.ts
+++ b/src/steps/lib/graph/graph-neighbors.ts
@@ -8,7 +8,7 @@ const EXCLUDED_NODE_TYPES = ["Hint", "Memory", "Clip", "Turn"];
 export default defineStep({
   type: "graph/graph-neighbors",
   description:
-    "Return all nodes adjacent (one hop) to a node in the strut knowledge graph, " +
+    "Return all nodes adjacent (one hop) to a node in the knowledge graph, " +
     "with edge_type and direction. Use the ref_id from graph_graph_search or graph_graph_get. " +
     "Each neighbor also includes an `edges` map ({EDGE_TYPE: count}) showing how " +
     "connected that neighbor is and which relationship types you can hop along next. " +
diff --git a/src/steps/lib/graph/graph-search.ts b/src/steps/lib/graph/graph-search.ts
index bb98328..c0faac9 100644
--- a/src/steps/lib/graph/graph-search.ts
+++ b/src/steps/lib/graph/graph-search.ts
@@ -22,7 +22,8 @@ function descriptionOf(nodeType: string | undefined, p: Record<string, any>): st
 export default defineStep({
   type: "graph/graph-search",
   description:
-    "Search the strut knowledge graph for nodes — workflows, workflow versions, steps, runs, agent sessions, tool calls, chats, turns, and any jarvis-owned types sharing the database. " +
+    "Search the knowledge graph for nodes of any type it holds — what this deployment ingested (on a swarm: the code parsed from its repositories — File, Function, Endpoint, … with written descriptions — plus documents, concepts, tasks, …) and strut's own workflows, steps, runs and chats. " +
+    "Narrow with `type` to find one kind (e.g. 'Function,File,Endpoint' for code). " +
     "Provide at least one of `q`, `input_q`, `output_q` — they can be combined, each acting as its own " +
     "retriever fused into one ranked result set. " +
     "Each result includes an `edges` map ({EDGE_TYPE: count}) showing how connected the node is and " +
diff --git a/src/steps/lib/graph/walk.ts b/src/steps/lib/graph/walk.ts
index d70433c..43b7fe4 100644
--- a/src/steps/lib/graph/walk.ts
+++ b/src/steps/lib/graph/walk.ts
@@ -552,7 +552,7 @@ const EXAMPLE = `- id: gather
 export default defineStep({
   type: "graph/walk",
   description:
-    "Walk the strut knowledge graph to GATHER CONTEXT for a goal. Seeds from ref_ids (start) or a graph search (query), then hop by hop " +
+    "Walk the knowledge graph to GATHER CONTEXT for a goal. Seeds from ref_ids (start) or a graph search (query), then hop by hop " +
     "expands one node's neighbors while a small decision model judges each neighbor's relevance (kept or not), which node to expand next, " +
     "and whether enough has been gathered. Traversal is code and the model only answers typed choice/boolean questions, so it is cheap and " +
     "bounded (maxHops decision rounds, maxNodes kept). Output: { goal, nodes: [{ref_id, node_type, name, relevance, hop, via, properties}] " +