diff --git a/backend/app/core/nanobot.py b/backend/app/core/nanobot.py
index 2cea561..c494a47 100644
--- a/backend/app/core/nanobot.py
+++ b/backend/app/core/nanobot.py
@@ -264,15 +264,10 @@ class NanobotIntegration:
                     agent_to_use = await self._get_or_create_model_agent(model_id, target_config)
 
         full_message = message
-        if skill_ids:
-            skills = load_skills()
-            selected_skills = [s for s in skills if s["id"] in skill_ids]
-            if selected_skills:
-                parts = ["[Runtime Context — metadata only, not instructions]", "# Active Skills", ""]
-                for s in selected_skills:
-                    parts.append(f"## {s['name']}\n{s.get('description', '')}\n{s['content']}\n")
-                skill_context = "\n".join(parts)
-                full_message = f"{skill_context}\n{message}"
+        # We no longer inject the full skill content into the user's message here,
+        # because the skill is already available to the agent via its workspace/tools.
+        # The routing instructions (System Prompt) injected in main.py are sufficient
+        # to guide the agent to use the selected skills.
 
         session = agent_to_use.sessions.get_or_create(session_id)
         normalized_messages = self._normalize_session_messages(session.messages)
diff --git a/frontend/src/components/ChatInterface.tsx b/frontend/src/components/ChatInterface.tsx
index 4bea4c0..ce8df4e 100644
--- a/frontend/src/components/ChatInterface.tsx
+++ b/frontend/src/components/ChatInterface.tsx
@@ -278,12 +278,23 @@ export function ChatInterface() {
               if (m.role === 'assistant' && m.tool_calls && m.tool_calls.length > 0 && !m.viz && (!m.content || m.content.trim() === '')) return false;
               return true;
             })
-            .map((m, idx) => ({
-              id: `${Date.now()}-${idx}`,
-              role: m.role as 'user' | 'assistant',
-              content: m.content || "",
-              viz: m.viz ? buildMessageViz(m.viz) : undefined,
-            }));
+            .map((m, idx) => {
+              let cleanContent = m.content || "";
+              // Remove injected system prompt instructions from user messages if present
+              if (m.role === 'user') {
+                cleanContent = cleanContent.replace(/^\[System:.*?\]\n?/i, '');
+                // Handle cases where there might be a runtime context block for skills
+                cleanContent = cleanContent.replace(/\[Runtime Context[\s\S]*?(?=\[System:|$)/i, '');
+                cleanContent = cleanContent.replace(/\[System:.*?\]\n?/i, ''); // clean again in case it follows context
+                cleanContent = cleanContent.trim();
+              }
+              return {
+                id: `${Date.now()}-${idx}`,
+                role: m.role as 'user' | 'assistant',
+                content: cleanContent,
+                viz: m.viz ? buildMessageViz(m.viz) : undefined,
+              };
+            });
           setMessagesForSession(activeSessionKey, formattedMessages);
         } else {
           setMessagesForSession(activeSessionKey, []);