diff --git a/tests/tools/test_terminal_tool_description.py b/tests/tools/test_terminal_tool_description.py
new file mode 100644
index 0000000000000..3eb241917d91c
--- /dev/null
+++ b/tests/tools/test_terminal_tool_description.py
@@ -0,0 +1,28 @@
+"""Regression test for the terminal tool description wording (issue #7672).
+
+The description used to claim "Execute shell commands on a Linux environment",
+which is misleading for users running locally on Windows/macOS or in a non-Linux
+configured backend. Pin the invariant: the description must describe the
+environment as *configured* rather than hardcoding Linux.
+"""
+
+from __future__ import annotations
+
+from tools.terminal_tool import TERMINAL_TOOL_DESCRIPTION
+
+
+def test_terminal_description_does_not_hardcode_linux():
+    # The core of issue #7672: don't assert every user is on Linux.
+    assert "Linux environment" not in TERMINAL_TOOL_DESCRIPTION
+
+
+def test_terminal_description_mentions_configured_environment():
+    # It should frame the environment as setup-dependent, naming the real options.
+    assert "configured execution environment" in TERMINAL_TOOL_DESCRIPTION
+    assert "local machine" in TERMINAL_TOOL_DESCRIPTION
+    assert "Docker container" in TERMINAL_TOOL_DESCRIPTION
+
+
+def test_terminal_description_keeps_persistence_contract():
+    # The persistence guidance the rest of the prompt relies on must survive.
+    assert "persist between calls" in TERMINAL_TOOL_DESCRIPTION
diff --git a/tools/terminal_tool.py b/tools/terminal_tool.py
index 6a5a6af1fdfe0..962d1ca3ca414 100644
--- a/tools/terminal_tool.py
+++ b/tools/terminal_tool.py
@@ -932,7 +932,7 @@ def _transform_sudo_command(command: str | None) -> tuple[str | None, str | None
 
 
 # Tool description for LLM
-TERMINAL_TOOL_DESCRIPTION = """Execute shell commands on a Linux environment. Filesystem, current working directory, and exported environment variables persist between calls.
+TERMINAL_TOOL_DESCRIPTION = """Execute shell commands in the configured execution environment (local machine, Docker container, or cloud sandbox depending on setup — not necessarily Linux). Filesystem, current working directory, and exported environment variables persist between calls.
 
 Do NOT use cat/head/tail to read files — use read_file instead.
 Do NOT use grep/rg/find to search — use search_files instead.