feat: true interactive mode — run claude directly, verdict via file, no script/capture

prompts/evaluator/_base.md

@@ -42,16 +42,18 @@ Evaluate story **`{{CURRENT_STORY_ID}}`**. This is the story the generator just
 
 ## Verdict Format
 
-You MUST end your response with EXACTLY ONE of these verdict blocks:
+You MUST do TWO things when delivering your verdict:
 
-### If the story genuinely passes all criteria:
+### 1. Write the verdict to a file
 
+Write your verdict to `{{LOOP_DIR}}/.verdict` using the Write tool. This file is how the loop harness reads your decision.
+
+**If PASS:**
 ```
 <verdict>PASS</verdict>
 ```
 
-### If any criterion is not met or issues are found:
-
+**If REJECT:**
 ```
 <verdict>REJECT</verdict>
 <rejection_reason>
@@ -61,6 +63,10 @@ Be concrete — "the function doesn't handle null input" not "there might be edg
 </rejection_reason>
 ```
 
+### 2. Also include the verdict in your response
+
+End your response with the same verdict block so it's visible in the terminal output.
+
 ## What Warrants Rejection
 
 - ANY acceptance criterion not actually met (not "mostly met" — MET)