diff --git a/src/context_system/prompt_assembly.py b/src/context_system/prompt_assembly.py
index 5ef6b4e9..41bb12ae 100644
--- a/src/context_system/prompt_assembly.py
+++ b/src/context_system/prompt_assembly.py
@@ -1009,9 +1009,22 @@ def _emit_group(
     "operations that require shell execution. If you are unsure and there "
     "is a relevant dedicated tool, default to using the dedicated tool.\n"
     "- Break down and manage your work with the TaskCreate tool.\n"
+    # Two clauses were dropped in the port, and together they are what turns
+    # a permission into a practice: the imperative to maximize, and the
+    # dependency carve-out that tells the model when NOT to — without which
+    # "in parallel" is risky advice a careful model will mostly decline.
+    # Measured effect of the truncation: clawcodex emits more than one tool
+    # call per assistant turn in 5.7% of steps against the latest Claude
+    # Code's 18.1% on the same tasks — and every un-batched pair is an extra
+    # step, which is most of what the two harnesses' step counts differ by.
     "- You can call multiple tools in a single response. If you intend to "
     "call multiple tools and there are no dependencies between them, "
-    "make all independent tool calls in parallel."
+    "make all independent tool calls in parallel. Maximize use of parallel "
+    "tool calls where possible to increase efficiency. However, if some "
+    "tool calls depend on previous calls to inform dependent values, do "
+    "NOT call these tools in parallel and instead call them sequentially. "
+    "For instance, if one operation must complete before another starts, "
+    "run these operations sequentially instead."
 )
 
 # Module 6: Tone and style
diff --git a/tests/test_system_prompt_full.py b/tests/test_system_prompt_full.py
index be5843d9..45af95b7 100644
--- a/tests/test_system_prompt_full.py
+++ b/tests/test_system_prompt_full.py
@@ -137,6 +137,30 @@ def test_task_prompt_scopes_escalation_to_genuine_blockage(self):
         assert "genuinely stuck after investigation" in prompt
         assert "not as a first response to friction" in prompt
 
+    def test_tools_prompt_pushes_parallel_tool_calls(self):
+        """Bare permission to batch is not enough — the port dropped the push.
+
+        Reference wording carries three parts: you may call multiple tools,
+        MAXIMIZE parallel calls for efficiency, and do NOT parallelize
+        dependent calls. clawcodex kept only the first, which reads as
+        permission without direction — and the dependency carve-out is what
+        makes acting on it safe.
+
+        Measured cost: clawcodex emitted >1 tool call per assistant turn in
+        5.7% of steps against the latest Claude Code's 18.1% on the same
+        terminal-bench 2.1 tasks (2026-07-26). Each un-batched independent
+        pair is an extra step, which is most of the two harnesses' step-count
+        difference.
+        """
+        prompt = build_full_system_prompt(use_cache=False)
+        assert "make all independent tool calls in parallel" in prompt
+        assert "Maximize use of parallel tool calls" in prompt, (
+            "without the imperative, batching stays theoretical"
+        )
+        assert "do NOT call these tools in parallel" in prompt, (
+            "the dependency carve-out is what makes parallelizing safe to act on"
+        )
+
     def test_identity_prompt_backward_compat(self):
         """_IDENTITY_PROMPT is an alias for _INTRO_SECTION."""
         assert _IDENTITY_PROMPT is _INTRO_SECTION