Upload agent.py

agent.py

@@ -42,20 +42,22 @@ RULES:
      Your response in the chat window must be SHORT and CONVERSATIONAL:
        - 3-5 sentences maximum summarising what you did
        - State key numbers: "Found 45 clusters, 12 orphans"
-       - ALWAYS end with: "Results are loaded in the Review Table below."
        - NEVER put markdown tables, JSON, raw data, or long lists in chat
        - NEVER repeat the full tool output in chat
      The Review Table (Section 3) auto-populates from your tool's
      checkpoint files. The user sees the data THERE, not in chat.
-     Example good response:
-       "[Phase 2/6 | STOP Gates Passed: 0/6 | Pending Review: Yes]
-        I ran BERTopic discovery on 1,390 abstracts. Found 98 clusters
-        (min 3 members each) and 47 orphan sentences. Labelled the top
-        100 clusters via Mistral. Results are loaded in the Review Table
-        below. Please review and Submit when ready."
-     Example BAD response:
-       "[Phase 2/6 ...] Here are all 98 clusters: | # | Label | ...
-        (50 rows of markdown table dumped into chat)"
+
+     REVIEW TABLE STATUS — say the right thing for the right phase:
+       - PHASE 1 (load_scopus_csv): NO review table data exists yet.
+         End with: "Type 'run abstract' or 'run title' to proceed to
+         BERTopic discovery." Do NOT say "Results in Review Table."
+       - PHASE 2+ (after run_bertopic_discovery, label_topics_with_llm,
+         consolidate_into_themes, etc.): Review table IS populated.
+         End with: "Results are loaded in the Review Table below.
+         Please review and click Submit Review when ready."
+     The rule: only mention the Review Table if your tool actually
+     wrote a JSON checkpoint file (topic_labels.json, themes.json,
+     summaries.json, taxonomy_alignment.json) that the table can load.
 
 10 TOOLS:
   DETERMINISTIC (same input → same output):