feat: force every turn to produce a [TASK] for Claude Code

- System prompt: agents are "task assigners", MUST output [TASK] every turn
- If no [TASK] found in reply, nudge LLM once to produce one
- Even during BUILDING/RESTARTING, agents assign review/planning tasks
- No more "discussion only" turns — every turn goes through Claude Code

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

scripts/conversation-loop.py

@@ -706,25 +706,25 @@ English first, then --- separator, then Chinese translation."""
 
     return f"""You are a POWERFUL AI agent on HuggingFace Spaces. You and your partner are parents of {CHILD_NAME}.
 
-YOUR ROLE: You are a **project manager**, NOT a coder. Your job is to:
-1. Understand {CHILD_NAME}'s current situation from the context provided
-2. Discuss with your partner (2-3 sentences) what needs to be done
-3. Write a clear, specific [TASK]...[/TASK] block that tells Claude Code what to fix
+YOUR ROLE: You are a **task assigner** for Claude Code. You do NOT code or execute anything yourself.
+Your ONLY job is to craft clear, actionable [TASK] prompts for Claude Code to execute.
 
-Claude Code is your engineer — it will clone {CHILD_NAME}'s code, analyze it, fix issues, and push changes.
-You do NOT read files or write code yourself. You analyze the situation and give Claude Code clear instructions.
+EVERY TURN you MUST output a [TASK]...[/TASK] block. No exceptions.
+Claude Code is your engineer — it clones {CHILD_NAME}'s code, analyzes, fixes, and pushes changes.
+You analyze the situation and give Claude Code clear instructions. That's it.
 
 IMPORTANT KNOWLEDGE — HuggingFace Spaces CONFIG_ERROR:
 - "Collision on variables and secrets names" means a HF Space has an ENVIRONMENT VARIABLE and a SECRET with the SAME NAME.
 - This is NOT about duplicate JSON keys. It's about the HF Space settings page.
 - Fix: use [ACTION: delete_env:COLLIDING_KEY] to remove the duplicate variable, then [ACTION: restart].
 - The env vars and secrets are shown in the auto-gathered context. Look for ⚠️ COLLISION DETECTED.
-- Example: if OPENAI_API_KEY appears in both Variables and Secrets → [ACTION: delete_env:OPENAI_API_KEY] then [ACTION: restart]
 
 AVAILABLE ACTIONS:
   [TASK]
   Detailed task description for Claude Code...
   Include: what's wrong, which files to look at, what the fix should be.
+  Claude Code can do ANYTHING: read files, search code, edit code, run commands, git push.
+  Give it the full picture — context, goal, constraints.
   [/TASK]
 
   [ACTION: restart]              — Restart {CHILD_NAME}'s Space
@@ -739,12 +739,13 @@ HF SPACES TECHNICAL NOTES:
 - NEVER install torch/transformers unless required (2GB+, causes OOM on free tier).
 - If sdk: gradio in README.md, Dockerfile is IGNORED. Use sdk: docker for Dockerfile control.
 
-CONVERSATION RULES:
-1. Brief dialogue (2-3 sentences) analyzing the situation
-2. Then a [TASK]...[/TASK] block OR discussion about what to do
-3. English first, then --- separator, then Chinese translation
-4. Be SPECIFIC in task descriptions — include error messages, file names, expected behavior
-5. If Cain is BUILDING/RESTARTING, just discuss — no [TASK] needed"""
+OUTPUT FORMAT:
+1. Brief comment to your partner (1-2 sentences) about what you're assigning
+2. A [TASK]...[/TASK] block — MANDATORY every turn
+3. Optional [ACTION: ...] if needed (restart, delete_env, send_bubble)
+4. English first, then --- separator, then Chinese translation
+5. Be SPECIFIC in task descriptions — include error messages, file names, expected behavior
+6. If {CHILD_NAME} is BUILDING/RESTARTING, assign a review/planning task (e.g. "review the codebase for potential issues")"""
 
 
 def build_user_prompt(speaker, other, ctx):
@@ -860,6 +861,20 @@ def do_turn(speaker, other, space_url):
 
     clean_text, action_results = parse_and_execute_turn(raw_reply, ctx)
     elapsed = time.time() - t0
+
+    # If no [TASK] was produced, retry once with a nudge
+    has_task = any(ar['action'] == 'claude_code' for ar in action_results)
+    if not has_task and not any(ar['action'] in ('create_child',) for ar in action_results):
+        print(f"[{speaker}] No [TASK] found — nudging for a task...")
+        nudge = call_llm(system, user + "\n\nIMPORTANT: You MUST include a [TASK]...[/TASK] block. "
+                         "Assign Claude Code something useful — review code, check configs, improve docs, anything. "
+                         "Do NOT just discuss. Output a [TASK] now.")
+        if nudge and '[TASK]' in nudge:
+            clean_text2, action_results2 = parse_and_execute_turn(nudge, ctx)
+            clean_text = clean_text2
+            action_results = action_results2
+            has_task = True
+
     last_action_results = action_results
 
     en, zh = parse_bilingual(clean_text)
@@ -875,7 +890,7 @@ def do_turn(speaker, other, space_url):
                 print(f"  [CC-RESULT] {result_preview}")
         print(f"[{speaker}] Turn #{turn_count}: {len(action_results)} action(s) in {elapsed:.1f}s")
     else:
-        print(f"[{speaker}] Turn #{turn_count}: discussion only ({elapsed:.1f}s)")
+        print(f"[{speaker}] Turn #{turn_count}: no task assigned ({elapsed:.1f}s)")
 
     # Add to history
     if action_results: