diff --git a/python/packages/tools/README.md b/python/packages/tools/README.md
index 2cc7ed6ad22..8ea7b18e712 100644
--- a/python/packages/tools/README.md
+++ b/python/packages/tools/README.md
@@ -43,15 +43,20 @@ asyncio.run(main())
 ### Safety
 
 > **`LocalShellTool` is not a sandbox.** It runs commands directly on the
-> host with the agent process's privileges. The actual security boundary
-> is **approval-in-the-loop**. For untrusted input use a sandboxed
-> executor — see [`agent-framework-hyperlight`](#relationship-to-agent-framework-hyperlight).
+> host with the agent process's permissions. Human approval provides a check
+> before agent-requested commands run, but does not isolate the shell.
+> Production use needs separately enforced isolation and restricted permissions.
+> For untrusted input use a sandboxed executor, such as `DockerShellTool` for
+> shell commands — see also
+> [`agent-framework-hyperlight`](#relationship-to-agent-framework-hyperlight).
 
 Defenses (in priority order):
 
-- **Approval-in-the-loop** — every command surfaces as a
-  `user_input_request`; nothing runs without consent. Disabling this
-  requires `acknowledge_unsafe=True`.
+- **Approval-in-the-loop** — when an agent calls the tool returned by
+  `as_function()` with `approval_mode="always_require"` (the default),
+  commands require approval through a `user_input_request` before they run.
+  Disabling this requires `acknowledge_unsafe=True`. Direct calls to
+  `run()` do not request approval, regardless of the configured approval mode.
 - **Process-tree termination on timeout** via `psutil`, so child
   processes (`make`, watchers, network tools) cannot survive the timeout.
 - **Output truncation** to 64 KiB (head + tail with marker).
@@ -61,22 +66,32 @@ Defenses (in priority order):
   security boundary — operators are expected to supply patterns that
   match their workload (and they can be defeated by trivial obfuscation
   such as `\rm -rf /`, `${RM:=rm} -rf /`, `python -c "…"`, encoded
-  payloads, or PowerShell-native equivalents). Real isolation comes from
-  approval gating and the sandbox tier (`DockerShellTool`). See
+  payloads, or PowerShell-native equivalents). Isolation must be enforced
+  separately, for example by the sandbox tier (`DockerShellTool`);
+  neither approval nor command filters isolate the shell. See
   `tests/test_security.py` for the documented residual risk surface.
 
-Override with `ShellPolicy`:
+Command-text filtering with `ShellPolicy` and required human approval:
+
+> [!WARNING]
+> The filters below can allow embedded commands, including `$(...)` and
+> backticks, and do not enforce read-only access. Review the full command before
+> approving it. Approval does not isolate the shell: commands still run with
+> the application's permissions.
 
 ```python
 from agent_framework.tools import LocalShellTool, ShellPolicy
 
 shell = LocalShellTool(
     policy=ShellPolicy(allowlist=[r"^ls\b", r"^cat\b", r"^git status$"]),
-    approval_mode="never_require",
-    acknowledge_unsafe=True,  # required to bypass approval
+    approval_mode="always_require",
 )
 ```
 
+Expose this tool to the agent through `shell.as_function()` and handle its
+approval requests with human review. Calling `shell.run()` directly does not
+request approval.
+
 ### Cross-OS
 
 - **Windows**: `pwsh -NoProfile -Command -` (falls back to `powershell.exe`).
@@ -160,7 +175,7 @@ container runtime with `docker_binary="podman"`.
 |---|---|---|
 | Run *code* (untrusted) | `HyperlightCodeActProvider.execute_code` (`agent-framework-hyperlight`) | Hyperlight WASM microVM |
 | Run *shell* (untrusted) | `DockerShellTool` | OCI container (network-off, non-root, capabilities dropped) |
-| Run *shell* (trusted dev) | `LocalShellTool` | Approval-in-the-loop |
+| Run *shell* (trusted dev) | `LocalShellTool` | None; human approval for agent calls by default |
 
 ## Relationship to `agent-framework-hyperlight`
 
diff --git a/python/packages/tools/agent_framework_tools/shell/_tool.py b/python/packages/tools/agent_framework_tools/shell/_tool.py
index 55acd1af7ce..86fb896739c 100644
--- a/python/packages/tools/agent_framework_tools/shell/_tool.py
+++ b/python/packages/tools/agent_framework_tools/shell/_tool.py
@@ -108,8 +108,9 @@ class LocalShellTool:
             ``cd``-wandering in one call does not leak to the next. This is
             a **re-anchor**, not a hard confinement — a command that does
             ``cd /tmp && rm -rf .`` in one call can still touch ``/tmp``.
-            Use :class:`ShellPolicy` or a sandboxed executor for true
-            confinement.
+            :class:`ShellPolicy` only filters command text; it does not
+            restrict file access. Use an executor with separately enforced
+            isolation and restricted permissions to limit access.
         env: Seed environment. In stateless mode this replaces the child's
             environment unless ``clean_env=False``. In persistent mode the
             variables are exported before the session is used.
diff --git a/python/samples/02-agents/tools/README.md b/python/samples/02-agents/tools/README.md
index c4bfaf4bc4e..a15f85f56c9 100644
--- a/python/samples/02-agents/tools/README.md
+++ b/python/samples/02-agents/tools/README.md
@@ -68,11 +68,19 @@ take effect on the **next iteration** of the function-calling loop.
 
 | Path | Demonstrates |
 |------|--------------|
-| [`local_shell_with_allowlist.py`](local_shell_with_allowlist.py) | `LocalShellTool` restricted by a strict command allow-list. |
+| [`local_shell_with_allowlist.py`](local_shell_with_allowlist.py) | Command-text filtering without human approval; not safe for production as shown. |
 | [`local_shell_with_environment_provider.py`](local_shell_with_environment_provider.py) | `LocalShellTool` wired with a `ShellEnvironmentProvider`. |
 | [`local_code_interpreter/`](local_code_interpreter/) | Hyperlight-backed sandboxed code interpreter (standalone tool — *extra* pattern). |
 | [`monty_code_interpreter/`](monty_code_interpreter/) | Monty-backed sandboxed code interpreter (standalone tool — *extra* pattern). |
 
+> [!WARNING]
+> The allow-list shell sample is educational, not a safe production configuration.
+> Its filters can allow embedded shell commands and do not enforce read-only
+> access. Commands run with the application's permissions, without human approval.
+> Use only an isolated, disposable environment without secrets or valuable data.
+> Production use needs separately enforced isolation and restricted permissions;
+> human review can add a check, but does not isolate the shell.
+
 > [!TIP]
 > The `local_code_interpreter/` and `monty_code_interpreter/` samples show the
 > standalone-tool wiring and are provided as *extra* reference. For most
diff --git a/python/samples/02-agents/tools/local_shell_with_allowlist.py b/python/samples/02-agents/tools/local_shell_with_allowlist.py
index 90c85a0a812..3da1503bd9e 100644
--- a/python/samples/02-agents/tools/local_shell_with_allowlist.py
+++ b/python/samples/02-agents/tools/local_shell_with_allowlist.py
@@ -8,12 +8,19 @@
 from dotenv import load_dotenv
 
 """
-LocalShellTool with a strict allow-list (no approval loop).
-
-Every command must match one of the allow-list regexes and the deny-list
-still wins. Approval is disabled because the allow-list is doing the
-gating; this is the safest fully-automatic configuration of
-``LocalShellTool``.
+LocalShellTool with command-text filtering (no approval loop).
+
+WARNING: This is an educational example, not a safe production configuration.
+Do not copy this approval-disabled setup into production. Use only an isolated,
+disposable environment without secrets or valuable data.
+
+The allow-list checks command text, not what the shell will execute. Embedded
+commands in $(...) or backticks can pass these filters and run; patterns that
+only check the start of a command can also allow extra operations. These filters
+do not enforce read-only access or protect against malicious model instructions.
+Commands run with the application's permissions and can access or change files
+and other resources available to it. Human approval and separately enforced
+isolation are not replaced by an allow-list.
 """
 
 load_dotenv()
@@ -24,6 +31,7 @@ async def main() -> None:
 
     shell = LocalShellTool(
         mode="stateless",
+        # Unsafe for production as shown: these filters do not replace human approval or isolation.
         approval_mode="never_require",
         acknowledge_unsafe=True,
         policy=ShellPolicy(
@@ -40,10 +48,7 @@ async def main() -> None:
 
     agent = Agent(
         client=client,
-        instructions=(
-            "You can run a narrow set of read-only shell commands (ls, pwd, cat, "
-            "git status/log/diff, python --version). Anything else will be rejected."
-        ),
+        instructions=("Use only these shell commands: ls, pwd, cat, git status/log/diff, python --version."),
         tools=[client.get_shell_tool(func=shell.as_function())],
     )