diff --git a/Intent.md b/Intent.md
index 35da2be..a1d0de1 100644
--- a/Intent.md
+++ b/Intent.md
@@ -35,6 +35,8 @@ The skill exists to answer:
 
 The evaluation must happen in 3 stages.
 
+Before moving to the next stage, wait until every agent in the current stage has finished its research and returned its output.
+
 ### Stage 1: Broad Scan
 
 Launch 5 subagents.
@@ -110,4 +112,3 @@ The skill is successful if it can:
 - support strong ideas with real reasoning,
 - surface useful market and execution risks,
 - and help the user decide whether the idea deserves an MVP.
-
diff --git a/claude/idea-evaluator/SKILL.md b/claude/idea-evaluator/SKILL.md
index b1e82d3..9fd29cd 100644
--- a/claude/idea-evaluator/SKILL.md
+++ b/claude/idea-evaluator/SKILL.md
@@ -20,6 +20,8 @@ Treat the task like a structured decision engine:
 
 Run the evaluation in 3 stages.
 
+Before moving to the next stage, wait until every agent in the current stage has finished its research and returned its output.
+
 ### Stage 1: Broad Scan
 
 Launch 5 subagents.
@@ -87,4 +89,3 @@ Include:
 - positives
 - negatives
 - MVP question
-
diff --git a/codex/idea-evaluator/SKILL.md b/codex/idea-evaluator/SKILL.md
index b1e82d3..9fd29cd 100644
--- a/codex/idea-evaluator/SKILL.md
+++ b/codex/idea-evaluator/SKILL.md
@@ -20,6 +20,8 @@ Treat the task like a structured decision engine:
 
 Run the evaluation in 3 stages.
 
+Before moving to the next stage, wait until every agent in the current stage has finished its research and returned its output.
+
 ### Stage 1: Broad Scan
 
 Launch 5 subagents.
@@ -87,4 +89,3 @@ Include:
 - positives
 - negatives
 - MVP question
-
diff --git a/openclaw/idea-evaluator/SKILL.md b/openclaw/idea-evaluator/SKILL.md
index b1e82d3..9fd29cd 100644
--- a/openclaw/idea-evaluator/SKILL.md
+++ b/openclaw/idea-evaluator/SKILL.md
@@ -20,6 +20,8 @@ Treat the task like a structured decision engine:
 
 Run the evaluation in 3 stages.
 
+Before moving to the next stage, wait until every agent in the current stage has finished its research and returned its output.
+
 ### Stage 1: Broad Scan
 
 Launch 5 subagents.
@@ -87,4 +89,3 @@ Include:
 - positives
 - negatives
 - MVP question
-