diff --git a/src/agent/stirrup_agent/runner.py b/src/agent/stirrup_agent/runner.py
index 46272278..bef1edc2 100644
--- a/src/agent/stirrup_agent/runner.py
+++ b/src/agent/stirrup_agent/runner.py
@@ -55,21 +55,10 @@
 _CONTEXT_SUMMARIZATION_CUTOFF = 0.75
 _CODE_EXEC_SYSTEM_PROMPT = """\
 Code execution:
-- MCP tools and their definitions are authoritative for domain data and semantics.
-  Never use code to query backing services or bypass an available MCP tool.
-- Do not overuse code_exec. Answer directly from MCP results, domain knowledge,
-  and basic reasoning or arithmetic when sufficient. Use code_exec only for
-  necessary computation, data processing, workspace inspection, or validation.
-  Never use it for planning, comments, placeholders, or empty scripts.
 - Prefer one complete script that inspects, analyzes, and verifies. Do not repeat
   equivalent experiments; correct failures directly.
 - Stay inside the execution workspace and use relative paths. Workspace state
   persists across code_exec calls.
-- For artifacts, inspect only the schema, counts, a small sample, or the specific
-  rows or fields needed, then process in place. If an artifact exceeds 200 KiB,
-  never print it in full; extract and process the relevant subset in bounded
-  batches. Avoid large record lists and verbose diagnostics. Reuse snapshots
-  unless domain state has changed.
 """
 _DOCKER_CODE_EXEC_SYSTEM_PROMPT = """\
 The Docker execution workspace is /workspace. Host filesystem paths are not
diff --git a/src/agent/stirrup_agent/tests/test_runner.py b/src/agent/stirrup_agent/tests/test_runner.py
index c64430c2..307f0a10 100644
--- a/src/agent/stirrup_agent/tests/test_runner.py
+++ b/src/agent/stirrup_agent/tests/test_runner.py
@@ -130,6 +130,19 @@ def test_stirrup_runner_uses_shared_prompt_when_code_is_disabled():
     assert runner._build_system_prompt() == AGENT_SYSTEM_PROMPT
 
 
+def test_stirrup_runner_uses_minimal_code_exec_guidance():
+    runner = StirrupAgentRunner(code_backend="local")
+
+    prompt = runner._build_system_prompt()
+    normalized_prompt = " ".join(prompt.split())
+
+    assert "Prefer one complete script" in normalized_prompt
+    assert "Keep all reads and writes inside the workspace" in normalized_prompt
+    assert "MCP tools and their definitions are authoritative" not in normalized_prompt
+    assert "Do not overuse code_exec" not in normalized_prompt
+    assert "If an artifact exceeds" not in normalized_prompt
+
+
 def test_stirrup_runner_forwards_temperature_to_litellm_client():
     runner = StirrupAgentRunner(
         model="watsonx/ibm/granite-4-h-small",