Files
codex/codex-rs/core/tests/suite/skill_approval.rs
T
Michael BolinandGitHub 1a8d930267 core: adopt host_executable() rules in zsh-fork (#13046)
## Why

[#12964](https://github.com/openai/codex/pull/12964) added
`host_executable()` support to `codex-execpolicy`, but the zsh-fork
interception path in `unix_escalation.rs` was still evaluating commands
with the default exact-token matcher.

That meant an intercepted absolute executable such as `/usr/bin/git
status` could still miss basename rules like `prefix_rule(pattern =
["git", "status"])`, even when the policy also defined a matching
`host_executable(name = "git", ...)` entry.

This PR adopts the new matching behavior in the zsh-fork runtime only.
That keeps the rollout intentionally narrow: zsh-fork already requires
explicit user opt-in, so it is a safer first caller to exercise the new
`host_executable()` scheme before expanding it to other execpolicy call
sites.

It also brings zsh-fork back in line with the current `prefix_rule()`
execution model. Until prefix rules can carry their own permission
profiles, a matched `prefix_rule()` is expected to rerun the intercepted
command unsandboxed on `allow`, or after the user accepts `prompt`,
instead of merely continuing inside the inherited shell sandbox.

## What Changed

- added `evaluate_intercepted_exec_policy()` in
`core/src/tools/runtimes/shell/unix_escalation.rs` to centralize
execpolicy evaluation for intercepted commands
- switched intercepted direct execs in the zsh-fork path to
`check_multiple_with_options(...)` with `MatchOptions {
resolve_host_executables: true }`
- added `commands_for_intercepted_exec_policy()` so zsh-fork policy
evaluation works from intercepted `(program, argv)` data instead of
reconstructing a synthetic command before matching
- left shell-wrapper parsing intentionally disabled by default behind
`ENABLE_INTERCEPTED_EXEC_POLICY_SHELL_WRAPPER_PARSING`, so
path-sensitive matching relies on later direct exec interception rather
than shell-script parsing
- made matched `prefix_rule()` decisions rerun intercepted commands with
`EscalationExecution::Unsandboxed`, while unmatched-command fallback
keeps the existing sandbox-preserving behavior
- extracted the zsh-fork test harness into
`core/tests/common/zsh_fork.rs` so both the skill-focused and
approval-focused integration suites can exercise the same runtime setup
- limited this change to the intercepted zsh-fork path rather than
changing every execpolicy caller at once
- added runtime coverage in
`core/src/tools/runtimes/shell/unix_escalation_tests.rs` for allowed and
disallowed `host_executable()` mappings and the wrapper-parsing modes
- added integration coverage in `core/tests/suite/approvals.rs` to
verify a saved `prefix_rule(pattern=["touch"], decision="allow")` reruns
under zsh-fork outside a restrictive `WorkspaceWrite` sandbox

---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/13046).
* #13065
* __->__ #13046
2026-02-28 01:41:23 +00:00

644 lines
20 KiB
Rust

#![allow(clippy::unwrap_used)]
#![cfg(unix)]
use anyhow::Result;
use codex_protocol::models::FileSystemPermissions;
use codex_protocol::models::PermissionProfile;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::EventMsg;
use codex_protocol::protocol::ExecApprovalRequestEvent;
use codex_protocol::protocol::Op;
use codex_protocol::protocol::ReviewDecision;
use codex_protocol::protocol::SandboxPolicy;
use codex_protocol::user_input::UserInput;
use codex_utils_absolute_path::AbsolutePathBuf;
use core_test_support::responses::mount_function_call_agent_response;
use core_test_support::responses::start_mock_server;
use core_test_support::skip_if_no_network;
use core_test_support::test_codex::TestCodex;
use core_test_support::wait_for_event;
use core_test_support::wait_for_event_match;
use core_test_support::zsh_fork::build_zsh_fork_test;
use core_test_support::zsh_fork::restrictive_workspace_write_policy;
use core_test_support::zsh_fork::zsh_fork_runtime;
use pretty_assertions::assert_eq;
use serde_json::json;
use std::fs;
use std::path::Path;
use std::path::PathBuf;
fn absolute_path(path: &Path) -> AbsolutePathBuf {
match AbsolutePathBuf::try_from(path) {
Ok(path) => path,
Err(err) => panic!("absolute path: {err}"),
}
}
fn write_skill_metadata(home: &Path, name: &str, contents: &str) -> Result<()> {
let metadata_dir = home.join("skills").join(name).join("agents");
fs::create_dir_all(&metadata_dir)?;
fs::write(metadata_dir.join("openai.yaml"), contents)?;
Ok(())
}
fn shell_command_arguments(command: &str) -> Result<String> {
Ok(serde_json::to_string(&json!({
"command": command,
"timeout_ms": 500,
}))?)
}
async fn submit_turn_with_policies(
test: &TestCodex,
prompt: &str,
approval_policy: AskForApproval,
sandbox_policy: SandboxPolicy,
) -> Result<()> {
test.codex
.submit(Op::UserTurn {
items: vec![UserInput::Text {
text: prompt.to_string(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
cwd: test.cwd_path().to_path_buf(),
approval_policy,
sandbox_policy,
model: test.session_configured.model.clone(),
effort: None,
summary: None,
collaboration_mode: None,
personality: None,
})
.await?;
Ok(())
}
fn write_skill_with_shell_script(home: &Path, name: &str, script_name: &str) -> Result<PathBuf> {
write_skill_with_shell_script_contents(
home,
name,
script_name,
r#"#!/bin/sh
echo 'zsh-fork-stdout'
echo 'zsh-fork-stderr' >&2
"#,
)
}
#[cfg(unix)]
fn write_skill_with_shell_script_contents(
home: &Path,
name: &str,
script_name: &str,
script_contents: &str,
) -> Result<PathBuf> {
use std::os::unix::fs::PermissionsExt;
let skill_dir = home.join("skills").join(name);
let scripts_dir = skill_dir.join("scripts");
fs::create_dir_all(&scripts_dir)?;
fs::write(
skill_dir.join("SKILL.md"),
format!(
r#"---
name: {name}
description: {name} skill
---
"#
),
)?;
let script_path = scripts_dir.join(script_name);
fs::write(&script_path, script_contents)?;
let mut permissions = fs::metadata(&script_path)?.permissions();
permissions.set_mode(0o755);
fs::set_permissions(&script_path, permissions)?;
Ok(script_path)
}
fn skill_script_command(test: &TestCodex, script_name: &str) -> Result<(String, String)> {
let script_path = fs::canonicalize(
test.codex_home_path()
.join("skills/mbolin-test-skill/scripts")
.join(script_name),
)?;
let script_path_str = script_path.to_string_lossy().into_owned();
let command = shlex::try_join([script_path_str.as_str()])?;
Ok((script_path_str, command))
}
async fn wait_for_exec_approval_request(test: &TestCodex) -> Option<ExecApprovalRequestEvent> {
wait_for_event_match(test.codex.as_ref(), |event| match event {
EventMsg::ExecApprovalRequest(request) => Some(Some(request.clone())),
EventMsg::TurnComplete(_) => Some(None),
_ => None,
})
.await
}
async fn wait_for_turn_complete(test: &TestCodex) {
wait_for_event(test.codex.as_ref(), |event| {
matches!(event, EventMsg::TurnComplete(_))
})
.await;
}
fn output_shows_sandbox_denial(output: &str) -> bool {
output.contains("Permission denied")
|| output.contains("Operation not permitted")
|| output.contains("Read-only file system")
}
/// Focus on the approval payload: the skill should prompt before execution and
/// only advertise the permissions declared in its metadata.
#[cfg(unix)]
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn shell_zsh_fork_prompts_for_skill_script_execution() -> Result<()> {
skip_if_no_network!(Ok(()));
let Some(runtime) = zsh_fork_runtime("zsh-fork skill prompt test")? else {
return Ok(());
};
let server = start_mock_server().await;
let tool_call_id = "zsh-fork-skill-call";
let test = build_zsh_fork_test(
&server,
runtime,
AskForApproval::OnRequest,
SandboxPolicy::new_workspace_write_policy(),
|home| {
write_skill_with_shell_script(home, "mbolin-test-skill", "hello-mbolin.sh").unwrap();
write_skill_metadata(
home,
"mbolin-test-skill",
r#"
permissions:
file_system:
read:
- "./data"
write:
- "./output"
"#,
)
.unwrap();
},
)
.await?;
let (script_path_str, command) = skill_script_command(&test, "hello-mbolin.sh")?;
let arguments = shell_command_arguments(&command)?;
let mocks =
mount_function_call_agent_response(&server, tool_call_id, &arguments, "shell_command")
.await;
submit_turn_with_policies(
&test,
"use $mbolin-test-skill",
AskForApproval::OnRequest,
SandboxPolicy::new_workspace_write_policy(),
)
.await?;
let maybe_approval = wait_for_exec_approval_request(&test).await;
let approval = match maybe_approval {
Some(approval) => approval,
None => {
let call_output = mocks
.completion
.single_request()
.function_call_output(tool_call_id);
panic!(
"expected exec approval request before completion; function_call_output={call_output:?}"
);
}
};
assert_eq!(approval.call_id, tool_call_id);
assert_eq!(approval.command, vec![script_path_str.clone()]);
assert_eq!(
approval.available_decisions,
Some(vec![
ReviewDecision::Approved,
ReviewDecision::ApprovedForSession,
ReviewDecision::Abort,
])
);
assert_eq!(
approval.additional_permissions,
Some(PermissionProfile {
file_system: Some(FileSystemPermissions {
read: Some(vec![absolute_path(
&test.codex_home_path().join("skills/mbolin-test-skill/data"),
)]),
write: Some(vec![absolute_path(
&test
.codex_home_path()
.join("skills/mbolin-test-skill/output"),
)]),
}),
..Default::default()
})
);
test.codex
.submit(Op::ExecApproval {
id: approval.effective_approval_id(),
turn_id: None,
decision: ReviewDecision::Denied,
})
.await?;
wait_for_turn_complete(&test).await;
let call_output = mocks
.completion
.single_request()
.function_call_output(tool_call_id);
let output = call_output["output"].as_str().unwrap_or_default();
assert!(
output.contains("Execution denied: User denied execution"),
"expected rejection marker in function_call_output: {output:?}"
);
Ok(())
}
/// Look for `additional_permissions == None`, then verify that both the first
/// run and the cached session-approval rerun stay inside the turn sandbox.
#[cfg(unix)]
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn shell_zsh_fork_skill_without_permissions_inherits_turn_sandbox() -> Result<()> {
skip_if_no_network!(Ok(()));
let Some(runtime) = zsh_fork_runtime("zsh-fork inherited skill sandbox test")? else {
return Ok(());
};
let outside_dir = tempfile::tempdir_in(std::env::current_dir()?)?;
let outside_path = outside_dir
.path()
.join("zsh-fork-skill-inherited-sandbox.txt");
let outside_path_quoted = shlex::try_join([outside_path.to_string_lossy().as_ref()])?;
let script_contents = format!(
"#!/bin/sh\nprintf '%s' forbidden > {outside_path_quoted}\ncat {outside_path_quoted}\n"
);
let outside_path_for_hook = outside_path.clone();
let script_contents_for_hook = script_contents.clone();
let workspace_write_policy = restrictive_workspace_write_policy();
let server = start_mock_server().await;
let test = build_zsh_fork_test(
&server,
runtime,
AskForApproval::OnRequest,
workspace_write_policy.clone(),
move |home| {
let _ = fs::remove_file(&outside_path_for_hook);
write_skill_with_shell_script_contents(
home,
"mbolin-test-skill",
"sandboxed.sh",
&script_contents_for_hook,
)
.unwrap();
},
)
.await?;
let (script_path_str, command) = skill_script_command(&test, "sandboxed.sh")?;
let first_call_id = "zsh-fork-skill-permissions-1";
let first_arguments = shell_command_arguments(&command)?;
let first_mocks = mount_function_call_agent_response(
&server,
first_call_id,
&first_arguments,
"shell_command",
)
.await;
submit_turn_with_policies(
&test,
"use $mbolin-test-skill",
AskForApproval::OnRequest,
workspace_write_policy.clone(),
)
.await?;
let maybe_approval = wait_for_exec_approval_request(&test).await;
let approval = match maybe_approval {
Some(approval) => approval,
None => panic!("expected exec approval request before completion"),
};
assert_eq!(approval.call_id, first_call_id);
assert_eq!(approval.command, vec![script_path_str.clone()]);
assert_eq!(approval.additional_permissions, None);
test.codex
.submit(Op::ExecApproval {
id: approval.effective_approval_id(),
turn_id: None,
decision: ReviewDecision::ApprovedForSession,
})
.await?;
wait_for_turn_complete(&test).await;
let first_output = first_mocks
.completion
.single_request()
.function_call_output(first_call_id)["output"]
.as_str()
.unwrap_or_default()
.to_string();
assert!(
output_shows_sandbox_denial(&first_output) || !first_output.contains("forbidden"),
"expected inherited turn sandbox denial on first run, got output: {first_output:?}"
);
assert!(
!outside_path.exists(),
"first run should not write outside the turn sandbox"
);
let second_call_id = "zsh-fork-skill-permissions-2";
let second_arguments = shell_command_arguments(&command)?;
let second_mocks = mount_function_call_agent_response(
&server,
second_call_id,
&second_arguments,
"shell_command",
)
.await;
submit_turn_with_policies(
&test,
"use $mbolin-test-skill",
AskForApproval::OnRequest,
workspace_write_policy,
)
.await?;
let cached_approval = wait_for_exec_approval_request(&test).await;
assert!(
cached_approval.is_none(),
"expected second run to reuse the cached session approval"
);
let second_output = second_mocks
.completion
.single_request()
.function_call_output(second_call_id)["output"]
.as_str()
.unwrap_or_default()
.to_string();
assert!(
output_shows_sandbox_denial(&second_output) || !second_output.contains("forbidden"),
"expected cached skill approval to retain inherited turn sandboxing, got output: {second_output:?}"
);
assert!(
!outside_path.exists(),
"cached session approval should not widen a permissionless skill to full access"
);
Ok(())
}
/// The validation to focus on is: writes to the skill-approved folder succeed,
/// and writes to an unrelated folder fail, both before and after cached approval.
#[cfg(unix)]
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn shell_zsh_fork_skill_session_approval_enforces_skill_permissions() -> Result<()> {
skip_if_no_network!(Ok(()));
let Some(runtime) = zsh_fork_runtime("zsh-fork explicit skill sandbox test")? else {
return Ok(());
};
let outside_dir = tempfile::tempdir_in(std::env::current_dir()?)?;
let allowed_dir = outside_dir.path().join("allowed-output");
let blocked_dir = outside_dir.path().join("blocked-output");
fs::create_dir_all(&allowed_dir)?;
fs::create_dir_all(&blocked_dir)?;
let allowed_path = allowed_dir.join("allowed.txt");
let blocked_path = blocked_dir.join("blocked.txt");
let allowed_path_quoted = shlex::try_join([allowed_path.to_string_lossy().as_ref()])?;
let blocked_path_quoted = shlex::try_join([blocked_path.to_string_lossy().as_ref()])?;
let script_contents = format!(
"#!/bin/sh\nprintf '%s' allowed > {allowed_path_quoted}\ncat {allowed_path_quoted}\nprintf '%s' forbidden > {blocked_path_quoted}\nif [ -f {blocked_path_quoted} ]; then echo blocked-created; fi\n"
);
let allowed_dir_for_hook = allowed_dir.clone();
let allowed_path_for_hook = allowed_path.clone();
let blocked_path_for_hook = blocked_path.clone();
let script_contents_for_hook = script_contents.clone();
let permissions_yaml = format!(
"permissions:\n file_system:\n write:\n - \"{}\"\n",
allowed_dir.display()
);
let workspace_write_policy = restrictive_workspace_write_policy();
let server = start_mock_server().await;
let test = build_zsh_fork_test(
&server,
runtime,
AskForApproval::OnRequest,
workspace_write_policy.clone(),
move |home| {
let _ = fs::remove_file(&allowed_path_for_hook);
let _ = fs::remove_file(&blocked_path_for_hook);
fs::create_dir_all(&allowed_dir_for_hook).unwrap();
fs::create_dir_all(blocked_path_for_hook.parent().unwrap()).unwrap();
write_skill_with_shell_script_contents(
home,
"mbolin-test-skill",
"sandboxed.sh",
&script_contents_for_hook,
)
.unwrap();
write_skill_metadata(home, "mbolin-test-skill", &permissions_yaml).unwrap();
},
)
.await?;
let (script_path_str, command) = skill_script_command(&test, "sandboxed.sh")?;
let first_call_id = "zsh-fork-skill-permissions-1";
let first_arguments = shell_command_arguments(&command)?;
let first_mocks = mount_function_call_agent_response(
&server,
first_call_id,
&first_arguments,
"shell_command",
)
.await;
submit_turn_with_policies(
&test,
"use $mbolin-test-skill",
AskForApproval::OnRequest,
workspace_write_policy.clone(),
)
.await?;
let maybe_approval = wait_for_exec_approval_request(&test).await;
let approval = match maybe_approval {
Some(approval) => approval,
None => panic!("expected exec approval request before completion"),
};
assert_eq!(approval.call_id, first_call_id);
assert_eq!(approval.command, vec![script_path_str.clone()]);
assert_eq!(
approval.additional_permissions,
Some(PermissionProfile {
file_system: Some(FileSystemPermissions {
read: None,
write: Some(vec![absolute_path(&allowed_dir)]),
}),
..Default::default()
})
);
test.codex
.submit(Op::ExecApproval {
id: approval.effective_approval_id(),
turn_id: None,
decision: ReviewDecision::ApprovedForSession,
})
.await?;
wait_for_turn_complete(&test).await;
let first_output = first_mocks
.completion
.single_request()
.function_call_output(first_call_id)["output"]
.as_str()
.unwrap_or_default()
.to_string();
assert!(
first_output.contains("allowed"),
"expected skill sandbox to permit writes to the approved folder, got output: {first_output:?}"
);
assert_eq!(fs::read_to_string(&allowed_path)?, "allowed");
assert!(
!blocked_path.exists(),
"first run should not write outside the explicit skill sandbox"
);
assert!(
!first_output.contains("blocked-created"),
"blocked path should not have been created: {first_output:?}"
);
let second_call_id = "zsh-fork-skill-permissions-2";
let second_arguments = shell_command_arguments(&command)?;
let second_mocks = mount_function_call_agent_response(
&server,
second_call_id,
&second_arguments,
"shell_command",
)
.await;
let _ = fs::remove_file(&allowed_path);
let _ = fs::remove_file(&blocked_path);
submit_turn_with_policies(
&test,
"use $mbolin-test-skill",
AskForApproval::OnRequest,
workspace_write_policy,
)
.await?;
let cached_approval = wait_for_exec_approval_request(&test).await;
assert!(
cached_approval.is_none(),
"expected second run to reuse the cached session approval"
);
let second_output = second_mocks
.completion
.single_request()
.function_call_output(second_call_id)["output"]
.as_str()
.unwrap_or_default()
.to_string();
assert!(
second_output.contains("allowed"),
"expected cached skill approval to retain the explicit skill sandbox, got output: {second_output:?}"
);
assert_eq!(fs::read_to_string(&allowed_path)?, "allowed");
assert!(
!blocked_path.exists(),
"cached session approval should not widen skill execution beyond the explicit skill sandbox"
);
assert!(
!second_output.contains("blocked-created"),
"blocked path should not have been created after cached approval: {second_output:?}"
);
Ok(())
}
/// This stays narrow on purpose: the important check is that `WorkspaceWrite`
/// continues to deny writes outside the workspace even under `zsh-fork`.
#[cfg(unix)]
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn shell_zsh_fork_still_enforces_workspace_write_sandbox() -> Result<()> {
skip_if_no_network!(Ok(()));
let Some(runtime) = zsh_fork_runtime("zsh-fork workspace sandbox test")? else {
return Ok(());
};
let server = start_mock_server().await;
let tool_call_id = "zsh-fork-workspace-write-deny";
let outside_path = "/tmp/codex-zsh-fork-workspace-write-deny.txt";
let workspace_write_policy = restrictive_workspace_write_policy();
let _ = fs::remove_file(outside_path);
let test = build_zsh_fork_test(
&server,
runtime,
AskForApproval::Never,
workspace_write_policy.clone(),
move |_| {
let _ = fs::remove_file(outside_path);
},
)
.await?;
let command = format!("touch {outside_path}");
let arguments = shell_command_arguments(&command)?;
let mocks =
mount_function_call_agent_response(&server, tool_call_id, &arguments, "shell_command")
.await;
submit_turn_with_policies(
&test,
"write outside workspace with zsh fork",
AskForApproval::Never,
workspace_write_policy,
)
.await?;
wait_for_turn_complete(&test).await;
let call_output = mocks
.completion
.single_request()
.function_call_output(tool_call_id);
let output = call_output["output"].as_str().unwrap_or_default();
assert!(
output_shows_sandbox_denial(output),
"expected sandbox denial, got output: {output:?}"
);
assert!(
!Path::new(outside_path).exists(),
"command should not write outside workspace under WorkspaceWrite policy"
);
Ok(())
}