fix: roundtable early exit, iteration cap, and ghost-agent prevention

chapter_roundtable.yml:
- Reduce max_iterations 9 → 5 (3 rounds of 3 editors is enough; 9 was
  burning credits in a retry loop after credit exhaustion)
- Add explicit 'Once any participant outputs CONSENSUS REACHED, the
  debate is over' — prevents continuation into wasted rounds

planning.yml:
- Add ANTI-HALLUCINATION RULE FOR AGENTS block: explicitly names the
  known ghost agents (Worldbuilder, Prose Engine, Plot Architect, etc.)
  and forbids their use; maps task types to canonical CLP agents so
  planning LLM has unambiguous fallback assignments

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

templates/chapter_roundtable.yml

@@ -24,7 +24,7 @@ steps:
   - type: think
     rotate_participants: true
     loop:
-      max_iterations: 9
+      max_iterations: 5
     hint: |
       You are {agent.name}, an editor in the editorial roundtable for this chapter.
       {agent.identity}
@@ -61,6 +61,9 @@ steps:
       - [Specific change 3 if needed]
       VERDICT: [PASS / REVISE / REWRITE]
 
+      Once any participant outputs CONSENSUS REACHED, the debate is over.
+      Do NOT continue to another round after CONSENSUS REACHED has been written.
+
   - type: package
     schema:
       consensus_critique: string

templates/planning.yml

@@ -56,6 +56,17 @@ steps:
       No ambiguous assignments. No placeholder names.
       Do NOT invent task type names. Each task_type MUST appear verbatim in the TEAM ROSTER.
 
+      ANTI-HALLUCINATION RULE FOR AGENTS:
+        NEVER invent agent names. Only assign tasks to agents that exist in the TEAM ROSTER above.
+        Invented agents like "Worldbuilder", "Prose Engine", "Plot Architect", "Voice Sculptor",
+        "Concept Analyst", "Logic Warden", "Heartstring Tensioner", or ANY name not in the roster
+        will cause routing failure — the task will run with no agent context and produce nothing.
+        If no existing agent perfectly fits a task, assign it to the closest match:
+          - Writing/prose tasks → Iris
+          - Planning/strategy tasks → Nova or Selene
+          - Research tasks → Forrest
+          - Editorial/review tasks → Devon, Lane, or Cora
+
       BOARDROOM RULE:
         - When the request is for a strategic debate, a structured discussion, a decision-making
           session, or a formal deliberation WITHOUT producing new content artifacts, create ONE task: