feat(eval): log agent response, judge scores, dimensions, and feedback

Add visibility into the eval loop: agent response preview (first 300
chars), dimension scores, judge feedback, and raw judge JSON response.
Essential for understanding why scenarios pass/fail and what to change.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: NetDevAutomate

packages/studyctl/src/studyctl/eval/judge/llm.py

@@ -50,12 +50,13 @@ def score(self, scenario: Scenario, response: str) -> JudgeResult:
         # 2. Build prompt from template
         prompt = self._build_prompt(scenario, response)
 
-        # 3. Call LLM
+        # 3. Call LLM judge
         try:
             raw = self.client.chat(
                 [{"role": "user", "content": prompt}],
                 temperature=0.1,
             )
+            logger.info("Judge raw response: %s", raw[:500])
         except LLMClientError as exc:
             logger.warning("LLM call failed for scenario %s: %s", scenario.id, exc)
             return JudgeResult(

packages/studyctl/src/studyctl/eval/orchestrator.py

@@ -42,6 +42,8 @@ def run_evaluation(
         try:
             target.setup(scenario)
             response = target.run(scenario)
+            if response:
+                logger.info("  Agent response (%d chars): %s", len(response), response[:300])
             score = judge.score(scenario, response)
         except EvalTimeout:
             logger.warning("Scenario %s timed out", scenario.id)
@@ -62,6 +64,13 @@ def run_evaluation(
             score.weighted_score,
             "PASS" if score.passed else "FAIL",
         )
+        # Log dimensions and feedback for visibility
+        if score.dimensions:
+            dims = "  ".join(f"{k}={v}" for k, v in score.dimensions.items())
+            logger.info("  Dimensions: %s", dims)
+        if score.feedback:
+            for fb in score.feedback:
+                logger.info("  Feedback: %s", fb)
 
         # Capture persona_hash from target if available (set during setup)
         if not persona_hash: