fix(security): backend-only PII redaction; fix fmt CI failure

Resolves all three findings from the second automated review and
fixes the cargo fmt --check CI failure (formatting drift in analysis.rs
from a prior merge).

[BLOCKER 1 + BLOCKER 2 + WARNING]
Frontend no longer performs any PII scanning or redaction. All three
concerns stemmed from the same root cause: outMessage was derived
on the frontend and used for display, DB storage (via lastUserMsgRef
and the chat bubble), and the AI payload — causing the original message
to be silently replaced before the backend received it.

Fix: frontend sends the original message verbatim. Backend is now the
sole authority. chat_message auto-redacts the typed message text using
PiiDetector + apply_redactions() before building the full payload, logs
the PII types via tracing::warn, and stores only the redacted form in
ai_messages and the audit log. The redacted form is returned to the
caller as ChatResponse.user_message (Option<String>, absent from direct
provider calls).

Frontend uses message (original) for the chat bubble and
lastUserMsgRef — resolution steps show natural language, not
[Password] tokens. The AI and DB see only the redacted version.

CI fix: cargo fmt applied to analysis.rs; all format checks now pass.

src-tauri/src/ai/anthropic.rs

@@ -116,6 +116,7 @@ impl Provider for AnthropicProvider {
             content,
             model,
             usage,
+            user_message: None,
             tool_calls: None,
         })
     }

src-tauri/src/ai/gemini.rs

@@ -119,6 +119,7 @@ impl Provider for GeminiProvider {
             content,
             model: config.model.clone(),
             usage,
+            user_message: None,
             tool_calls: None,
         })
     }

src-tauri/src/ai/mistral.rs

@@ -84,6 +84,7 @@ impl Provider for MistralProvider {
             content,
             model: config.model.clone(),
             usage,
+            user_message: None,
             tool_calls: None,
         })
     }

src-tauri/src/ai/mod.rs

@@ -30,6 +30,10 @@ pub struct ChatResponse {
     pub content: String,
     pub model: String,
     pub usage: Option<TokenUsage>,
+    /// The user message as it was stored in the DB (may be auto-redacted).
+    /// Set by chat_message; absent from direct provider calls.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub user_message: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub tool_calls: Option<Vec<ToolCall>>,
 }

src-tauri/src/ai/ollama.rs

@@ -100,6 +100,7 @@ impl Provider for OllamaProvider {
             content,
             model: config.model.clone(),
             usage,
+            user_message: None,
             tool_calls: None,
         })
     }

src-tauri/src/ai/openai.rs

@@ -197,6 +197,7 @@ impl OpenAiProvider {
             content,
             model: config.model.clone(),
             usage,
+            user_message: None,
             tool_calls,
         })
     }
@@ -397,6 +398,7 @@ impl OpenAiProvider {
             content,
             model: config.model.clone(),
             usage: None, // This custom REST contract doesn't provide token usage in response
+            user_message: None,
             tool_calls,
         })
     }

src-tauri/src/commands/ai.rs

@@ -234,9 +234,29 @@ pub async fn chat_message(
             .collect()
     };
 
-    // Load attachment files from DB, scan for PII, and embed clean content into the message.
+    // Auto-redact PII in both the typed message and any file attachments.
-    // File content never passes through the frontend — the backend is the single source of truth.
+    // The backend is the sole authority for redaction; the frontend sends original content.
     let full_message = {
+        // Step 1: redact the typed user message text.
+        let base = {
+            let spans = crate::pii::PiiDetector::new().detect(&message);
+            if spans.is_empty() {
+                message.clone()
+            } else {
+                let types: std::collections::HashSet<&str> =
+                    spans.iter().map(|s| s.pii_type.as_str()).collect();
+                let mut type_list: Vec<&str> = types.into_iter().collect();
+                type_list.sort_unstable();
+                warn!(
+                    pii_types = ?type_list,
+                    pii_count = spans.len(),
+                    "PII detected in typed chat message — auto-redacting before AI send"
+                );
+                crate::pii::apply_redactions(&message, &spans)
+            }
+        };
+
+        // Step 2: load attachment files from DB, scan, and embed clean content.
         let files: Vec<(String, String)> = if let Some(ref ids) = log_file_ids {
             let db = state.db.lock().map_err(|e| e.to_string())?;
             let mut v = Vec::new();
@@ -257,7 +277,7 @@ pub async fn chat_message(
             vec![]
         };
 
-        let mut msg = message.clone();
+        let mut msg = base;
         for (file_name, file_path) in &files {
             let content = std::fs::read_to_string(file_path).unwrap_or_default();
             let preview = &content[..content.len().min(8000)];
@@ -409,9 +429,11 @@ pub async fn chat_message(
     }
 
     // Save both user message and response to DB
+    let stored_user_message;
     {
         let db = state.db.lock().map_err(|e| e.to_string())?;
         let user_msg = AiMessage::new(conversation_id.clone(), "user".to_string(), full_message);
+        stored_user_message = user_msg.content.clone();
         let asst_msg = AiMessage::new(
             conversation_id,
             "assistant".to_string(),
@@ -468,7 +490,10 @@ pub async fn chat_message(
         }
     }
 
-    Ok(final_response)
+    Ok(crate::ai::ChatResponse {
+        user_message: Some(stored_user_message),
+        ..final_response
+    })
 }
 
 #[tauri::command]

src/lib/tauriCommands.ts

@@ -34,6 +34,8 @@ export interface ChatResponse {
   content: string;
   model: string;
   usage?: TokenUsage;
+  /** What was stored in the DB — may be auto-redacted. Use this for display and history. */
+  user_message?: string;
 }
 
 export interface AnalysisResult {

src/pages/Triage/index.tsx

@@ -8,7 +8,6 @@ import { useSessionStore } from "@/stores/sessionStore";
 import { useSettingsStore } from "@/stores/settingsStore";
 import {
   chatMessageCmd,
-  scanTextForPiiCmd,
   getIssueCmd,
   getIssueMessagesCmd,
   uploadLogFileCmd,
@@ -135,29 +134,10 @@ export default function Triage() {
     setIsLoading(true);
     setError(null);
 
-    // Auto-redact PII in typed message text before sending to AI.
-    // Spans are replaced in reverse-start-offset order to preserve byte positions.
-    let outMessage = message;
-    if (message.trim()) {
-      try {
-        const textResult = await scanTextForPiiCmd(message);
-        if (textResult.total_pii_found > 0) {
-          const sorted = [...textResult.detections].sort((a, b) => b.start - a.start);
-          let redacted = message;
-          for (const span of sorted) {
-            redacted = redacted.slice(0, span.start) + span.replacement + redacted.slice(span.end);
-          }
-          outMessage = redacted;
-        }
-      } catch {
-        // Non-fatal: if the scan fails, send original
-      }
-    }
-
     const displayContent =
       pendingFiles.length > 0
-        ? `${outMessage}${outMessage ? "\n" : ""}📎 ${pendingFiles.map((f) => f.name).join(", ")}`
+        ? `${message}${message ? "\n" : ""}📎 ${pendingFiles.map((f) => f.name).join(", ")}`
-        : outMessage;
+        : message;
 
     const userMsg: TriageMessage = {
       id: `user-${Date.now()}`,
@@ -167,7 +147,7 @@ export default function Triage() {
       why_level: currentWhyLevel,
       created_at: Date.now(),
     };
-    lastUserMsgRef.current = outMessage;
+    lastUserMsgRef.current = message;
     addMessage(userMsg);
     const logFileIds = pendingFiles.map((f) => f.logFileId);
     setPendingFiles([]);
@@ -184,7 +164,8 @@ export default function Triage() {
 
       // Use the active domain for the system prompt
       const systemPrompt = activeDomain ? getDomainPrompt(activeDomain) : undefined;
-      const response = await chatMessageCmd(id, outMessage, logFileIds, provider, systemPrompt);
+      // Backend auto-redacts PII in both message text and attachments before sending to AI.
+      const response = await chatMessageCmd(id, message, logFileIds, provider, systemPrompt);
       const assistantMsg: TriageMessage = {
         id: `asst-${Date.now()}`,
         issue_id: id,