refactor: skills-based prompt architecture

.env.example

@@ -98,6 +98,11 @@ SEARXNG_SECRET=
 RCA_OPTIMIZE_COSTS=false
 GEMINI_DISABLE_THINKING=
 
+# Optional: Token budget for auto-loaded integration skills during background RCA.
+# Defaults to 12000. Set higher to include more integration context, lower to save tokens.
+RCA_SKILLS_TOKEN_BUDGET=
+RCA_TOKEN_BUDGET=
+
 # Model overrides — defaults vary by setting; see each variable below.
 # Use provider/model format (e.g., openai/gpt-4.1, google/gemini-2.5-pro, ollama/llama3.1).
 

client/src/app/chat/components/ChatClient.tsx

@@ -374,7 +374,7 @@ export default function ChatClient({ initialSessionId, shouldStartNewChat, initi
     // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [handleChatSessionSelect, handleNewChat, currentSessionId]);
 
-  // Auto-send initial message from URL
+  // Auto-send initial message from URL or sessionStorage (e.g., Next Steps execution)
   useEffect(() => {
     if (initialMessage && chatWebSocket.isReady && userId && !isLoadingSessionMessages && !isSending && !initialMessageSentRef.current) {
       initialMessageSentRef.current = true;
@@ -384,6 +384,9 @@ export default function ChatClient({ initialSessionId, shouldStartNewChat, initi
             sessionStorage.removeItem('pendingChatMessage');
             const sessionIdToUse = currentSessionId;
             window.history.replaceState({}, '', `/chat${sessionIdToUse ? `?sessionId=${sessionIdToUse}` : ''}`);
+          } else {
+            // Reset so the effect can retry when conditions are met again
+            initialMessageSentRef.current = false;
           }
         });
       }, 500);

client/src/app/incidents/components/SuggestionModal.tsx

@@ -89,6 +89,7 @@ export default function SuggestionModal({
     if (!suggestion.command || isExecuting) return;
     setIsExecuting(true);
 
+    // Mark suggestion as executed (non-blocking — don't let this prevent navigation)
     try {
       const res = await fetch(`/api/incidents/suggestions/${suggestion.id}/mark-executed`, {
         method: 'POST',
@@ -97,13 +98,9 @@ export default function SuggestionModal({
       });
       if (!res.ok) {
         console.error('Failed to mark suggestion as executed:', res.status);
-        setIsExecuting(false);
-        return;
       }
     } catch (err) {
       console.error('Failed to mark suggestion as executed:', err);
-      setIsExecuting(false);
-      return;
     }
 
     const message = `Execute this command and report the output:\n\n\`\`\`\n${suggestion.command}\n\`\`\`\n\nRun ONLY this command. Report the output, then stop. Do not run follow-up commands or investigate further.`;

client/src/components/tool-calls/CommandLogo.tsx

@@ -258,6 +258,24 @@ const logos = {
       <path strokeLinecap="round" strokeLinejoin="round" d="M12 6.253v13m0-13C10.832 5.477 9.246 5 7.5 5S4.168 5.477 3 6.253v13C4.168 18.477 5.754 18 7.5 18s3.332.477 4.5 1.253m0-13C13.168 5.477 14.754 5 16.5 5c1.747 0 3.332.477 4.5 1.253v13C19.832 18.477 18.247 18 16.5 18c-1.746 0-3.332.477-4.5 1.253" />
     </svg>
   ),
+  loadSkill: (
+    <svg
+      className="w-4 h-4 min-w-4 min-h-4 text-teal-500 dark:text-teal-400"
+      fill="none"
+      stroke="currentColor"
+      strokeWidth="2"
+      viewBox="0 0 24 24"
+      xmlns="http://www.w3.org/2000/svg"
+    >
+      <circle cx="12" cy="12" r="3" />
+      <line x1="12" y1="3" x2="12" y2="9" />
+      <line x1="5.5" y1="17.5" x2="9.5" y2="14" />
+      <line x1="18.5" y1="17.5" x2="14.5" y2="14" />
+      <circle cx="12" cy="3" r="1.5" fill="currentColor" />
+      <circle cx="5.5" cy="17.5" r="1.5" fill="currentColor" />
+      <circle cx="18.5" cy="17.5" r="1.5" fill="currentColor" />
+    </svg>
+  ),
   rcaUpdate: (
     <svg
       className="w-4 h-4 min-w-4 min-h-4 text-red-600 dark:text-red-400"
@@ -305,6 +323,11 @@ const getLogoForCommand = (command: string | any, toolName: string, provider?: s
     return 'tailscale'
   }
 
+  // Load skill tool
+  if (tool === 'load_skill') {
+    return 'loadSkill'
+  }
+
   // Web search tool
   if (tool === 'web_search') {
     return 'web'

client/src/components/tool-calls/ToolExecutionWidget.tsx

@@ -196,6 +196,16 @@ const ToolExecutionWidget = ({ tool, className, sendMessage, sendRaw, onToolUpda
     command = "Knowledge Base"
   }
 
+  if (tool.tool_name === "load_skill") {
+    try {
+      const parsed = JSON.parse(normalizedInput)
+      const skillId = parsed.skill_id || parsed.kwargs?.skill_id || ''
+      command = skillId ? `Loading ${skillId} skill` : "Loading integration skill"
+    } catch {
+      command = "Loading integration skill"
+    }
+  }
+
   // terminal_exec parsing - extract command from input or output
   if (tool.tool_name === "terminal_exec") {
     try {

client/src/components/tool-calls/tool-output-renderer.tsx

@@ -35,6 +35,10 @@ const isIacTool = (toolName: string) => {
   return toolName.includes('iac') || toolName.includes('terraform')
 }
 
+const isLoadSkillTool = (toolName: string) => {
+  return toolName === 'load_skill'
+}
+
 const isWebSearchTool = (toolName: string) => {
   return toolName.includes('web_search')
 }
@@ -97,6 +101,19 @@ export function RenderOutput({
 
   // If not JSON, check tool type for appropriate rendering
   if (!parsed) {
+    // load_skill - compact one-liner, no raw content
+    if (isLoadSkillTool(toolName)) {
+      const alreadyLoaded = output.includes('already loaded')
+      return (
+        <div className="flex items-center gap-2 py-1">
+          <span className={`inline-block w-1.5 h-1.5 rounded-full ${alreadyLoaded ? 'bg-gray-400 dark:bg-gray-500' : 'bg-teal-500 dark:bg-teal-400'}`} />
+          <span className="text-xs text-gray-500 dark:text-gray-400">
+            {alreadyLoaded ? 'Already in context' : 'Integration guidance ready'}
+          </span>
+        </div>
+      )
+    }
+
     // IAC tools with HCL content - use Monaco Editor
     if (isIacTool(toolName) && isHclContent(output)) {
       const trimmedOutput = output.trim()

server/chat/backend/agent/agent.py

@@ -205,16 +205,15 @@ async def agentic_tool_flow(self, state: State) -> State:
                     logging.error(f"Invalid user_id format: '{state.user_id}' - must be a non-empty string")
                     # Don't fail completely, but log the error
 
-                # Get connected providers from database (always fetch from DB, no preferences stored)
+                # Get verified providers (cloud + SkillRegistry-validated integrations)
                 provider_preference = getattr(state, 'provider_preference', None)
                 if provider_preference is None:
                     try:
-                        from utils.auth.stateless_auth import get_connected_providers
-                        provider_preference = get_connected_providers(state.user_id)
+                        from chat.background.rca_prompt_builder import get_user_providers
+                        provider_preference = get_user_providers(state.user_id)
                     except Exception as e:
                         logging.error(f"Error getting connected providers: {e}")
                         provider_preference = []
-                    # Update state so downstream code can access it
                     state.provider_preference = provider_preference
 
                 selected_project_id = getattr(state, 'selected_project_id', None)