diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 0968bf9b6d..1207395ed6 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -62,9 +62,17 @@ _INTENT_SIGNAL = re.compile(
     # appear frequently in direct answers / explanations.
     r"\b(i['\u2019](ll|m going to|m gonna)|i am (going to|gonna)|i will|i shall|let me|allow me)\b"
     r"|"
-    # Step/plan framing: "First ...", "Step 1:", "Here's my plan", bare
-    # "Plan:" / "Approach:" as the first line of a structured reply.
-    r"\b(?:first\b|step \d+:?|here['\u2019]?s (?:my |the |a )?(?:plan|approach)|(?:plan|approach):)"
+    # Step/plan framing: "First ...", "Step 1:", "Here's my plan",
+    # "Here is the plan", "Here are my steps".
+    r"\b(?:first\b|step \d+:?|"
+    r"here['\u2019]?s (?:my |the |a )?(?:plan|approach)|"
+    r"here (?:is|are) (?:my |the |a )?(?:plan|approach|steps))"
+    r"|"
+    # Bare "Plan:" / "Approach:" (optionally preceded by a determiner
+    # like "My" / "The" / "Our") anchored to start of line so direct
+    # answers like "Here is a lesson plan:" or "meal plan:" do not trip
+    # the re-prompt path.
+    r"(?:^|\r?\n)[ \t]*(?:(?:my|the|our|a|this|that)\s+)?(?:plan|approach):"
     r"|"
     # "Now I" / "Next I" patterns
     r"\b(?:now i|next i)\b"
@@ -112,21 +120,25 @@ _NUMBERED_LIST_ARTIFACT = re.compile(
 )
 
 # Markers that a numbered list is a plan (still re-promptable), not a
-# final answer. Explicit "plan:" / "approach:" / "Here's my plan", OR
-# intent phrasing followed shortly by a plan / tool-action verb. The
+# final answer. The intent alternatives mirror _INTENT_SIGNAL above so
+# every recognised intent phrase can disqualify a numbered list. The
 # apostrophe in ``i['’]ll`` is required (no ``?``) so the regex does not
 # accidentally match the word "ill". The verb set is intentionally
 # conservative: ambiguous verbs like "write", "create", "make", "build"
 # are omitted because real answer lists use them ("1. Write a poem",
-# "1. Create directory").
+# "1. Create directory"). ``plan:`` / ``approach:`` is anchored to the
+# start of a line so "lesson plan:" / "meal plan:" do not trip the guard.
 _PLAN_LIST_FRAMING = re.compile(
-    r"\b(?:here['’]?s (?:my |the |a )?(?:plan|approach)|step \d+|"
-    r"i['’]ll|i will|i am going to|let me|now i|next i)\b"
+    r"\b(?:here['’]?s (?:my |the |a )?(?:plan|approach)|"
+    r"here (?:is|are) (?:my |the |a )?(?:plan|approach|steps)|"
+    r"step \d+|"
+    r"i['’](?:ll|m going to|m gonna)|i am (?:going to|gonna)|"
+    r"i will|i shall|let me|allow me|now i|next i)\b"
     r"[\s\S]{0,80}"
     r"\b(?:search|look up|call|use|fetch|browse|run|execute|"
     r"check|find|open|verify|compare|summari[sz]e|think|respond|"
     r"answer|analy[sz]e|explore|outline|gather|query|reason)\b"
-    r"|\b(?:plan|approach):",
+    r"|(?:^|\r?\n)[ \t]*(?:(?:my|the|our|a|this|that)\s+)?(?:plan|approach):",
     re.IGNORECASE,
 )
 
diff --git a/studio/backend/tests/test_llama_cpp_reprompt_guard.py b/studio/backend/tests/test_llama_cpp_reprompt_guard.py
index 22f5d11185..955dc423eb 100644
--- a/studio/backend/tests/test_llama_cpp_reprompt_guard.py
+++ b/studio/backend/tests/test_llama_cpp_reprompt_guard.py
@@ -376,6 +376,58 @@ def test_plan_framing_requires_apostrophe_in_ill():
         assert got == expected, f"{content!r} expected reprompt={expected} got {got}"
 
 
+def test_reprompts_on_all_intent_form_numbered_action_plans():
+    """``_PLAN_LIST_FRAMING`` must mirror every intent form that
+    ``_INTENT_SIGNAL`` accepts so numbered action plans phrased with
+    ``Allow me``, ``I'm going to``, ``I'm gonna``, ``I am gonna``,
+    ``I shall``, ``Now I``, ``Next I`` also re-prompt instead of being
+    silently classified as completed answers."""
+    samples = [
+        "Allow me to do this:\n1. search the docs\n2. fetch the result",
+        "I'm going to do this:\n1. search the docs\n2. fetch the result",
+        "I'm gonna do this:\n1. search the docs\n2. fetch the result",
+        "I am gonna do this:\n1. search the docs\n2. fetch the result",
+        "I shall do this:\n1. search the docs\n2. fetch the result",
+        "Now I will do these:\n1. search\n2. summarise",
+        "Next I will do these:\n1. fetch\n2. compare",
+    ]
+    for s in samples:
+        assert _would_reprompt(s), s
+
+
+def test_plan_colon_intent_is_line_anchored():
+    """``Plan:`` / ``Approach:`` only counts as an intent marker when it
+    is at the start of a line. Without this anchor, normal direct
+    answers containing phrases like ``lesson plan:``, ``meal plan:``,
+    ``migration plan:``, or ``My approach:`` would trigger the
+    re-prompt path and risk wiping a valid response."""
+    # These mid-line "plan:" / "approach:" mentions are NOT intent signals.
+    # The qualifier before "plan" is a content noun ("lesson", "meal",
+    # "migration") rather than a generic determiner ("my", "the", ...).
+    direct_answers = [
+        "Here is a lesson plan:\n1. Warm-up\n2. Group practice\n3. Assessment",
+        "I prepared a meal plan: rice, beans, eggs.",
+        "Quick approach: top-down then bottom-up.",
+    ]
+    for s in direct_answers:
+        assert not _INTENT_SIGNAL.search(s), s
+        assert not _would_reprompt(s), s
+    # "Plan:" / "Approach:" with optional generic determiner at the start
+    # of a line IS an intent signal.
+    plan_starts = [
+        "Plan:\n1. search\n2. summarise",
+        "Approach:\n1. fetch\n2. compare",
+        "  Plan:\n1. think\n2. respond",  # leading indent OK
+        "Lorem ipsum\nPlan:\n1. step\n2. step",  # plan: on a later line
+        "My plan:\n1. search\n2. summarise",
+        "The plan:\n1. look up\n2. compare",
+        "Our approach:\n1. fetch\n2. verify",
+    ]
+    for s in plan_starts:
+        assert _INTENT_SIGNAL.search(s), s
+        assert _would_reprompt(s), s
+
+
 # ── Cross-platform line endings ────────────────────────────────────