mirror of
https://github.com/openai/codex.git
synced 2026-11-06 00:30:29 +00:00
Refactor execpolicy fallback evaluation (#7544)
## Refactor of the `execpolicy` crate To illustrate why we need this refactor, consider an agent attempting to run `apple | rm -rf ./`. Suppose `apple` is allowed by `execpolicy`. Before this PR, `execpolicy` would consider `apple` and `pear` and only render one rule match: `Allow`. We would skip any heuristics checks on `rm -rf ./` and immediately approve `apple | rm -rf ./` to run. To fix this, we now thread a `fallback` evaluation function into `execpolicy` that runs when no `execpolicy` rules match a given command. In our example, we would run `fallback` on `rm -rf ./` and prevent `apple | rm -rf ./` from being run without approval.
This commit is contained in:
1 parent
e925a380dc
commit
3d35cb4619
27 files changed
+538
-257
No files matched your search
@@ -17,6 +17,7 @@ use crate::render::highlight::highlight_bash_to_lines;
|
||||
use crate::render::renderable::ColumnRenderable;
|
||||
use crate::render::renderable::Renderable;
|
||||
use codex_core::protocol::ElicitationAction;
|
||||
use codex_core::protocol::ExecPolicyAmendment;
|
||||
use codex_core::protocol::FileChange;
|
||||
use codex_core::protocol::Op;
|
||||
use codex_core::protocol::ReviewDecision;
|
||||
@@ -43,7 +44,7 @@ pub(crate) enum ApprovalRequest {
|
||||
command: Vec<String>,
|
||||
reason: Option<String>,
|
||||
risk: Option<SandboxCommandAssessment>,
|
||||
allow_prefix: Option<Vec<String>>,
|
||||
proposed_execpolicy_amendment: Option<ExecPolicyAmendment>,
|
||||
},
|
||||
ApplyPatch {
|
||||
id: String,
|
||||
@@ -105,8 +106,11 @@ impl ApprovalOverlay {
|
||||
header: Box<dyn Renderable>,
|
||||
) -> (Vec<ApprovalOption>, SelectionViewParams) {
|
||||
let (options, title) = match &variant {
|
||||
ApprovalVariant::Exec { allow_prefix, .. } => (
|
||||
exec_options(allow_prefix.clone()),
|
||||
ApprovalVariant::Exec {
|
||||
proposed_execpolicy_amendment,
|
||||
..
|
||||
} => (
|
||||
exec_options(proposed_execpolicy_amendment.clone()),
|
||||
"Would you like to run the following command?".to_string(),
|
||||
),
|
||||
ApprovalVariant::ApplyPatch { .. } => (
|
||||
@@ -337,7 +341,7 @@ impl From<ApprovalRequest> for ApprovalRequestState {
|
||||
command,
|
||||
reason,
|
||||
risk,
|
||||
allow_prefix,
|
||||
proposed_execpolicy_amendment,
|
||||
} => {
|
||||
let reason = reason.filter(|item| !item.is_empty());
|
||||
let has_reason = reason.is_some();
|
||||
@@ -360,7 +364,7 @@ impl From<ApprovalRequest> for ApprovalRequestState {
|
||||
variant: ApprovalVariant::Exec {
|
||||
id,
|
||||
command,
|
||||
allow_prefix,
|
||||
proposed_execpolicy_amendment,
|
||||
},
|
||||
header: Box::new(Paragraph::new(header).wrap(Wrap { trim: false })),
|
||||
}
|
||||
@@ -437,7 +441,7 @@ enum ApprovalVariant {
|
||||
Exec {
|
||||
id: String,
|
||||
command: Vec<String>,
|
||||
allow_prefix: Option<Vec<String>>,
|
||||
proposed_execpolicy_amendment: Option<ExecPolicyAmendment>,
|
||||
},
|
||||
ApplyPatch {
|
||||
id: String,
|
||||
@@ -470,7 +474,7 @@ impl ApprovalOption {
|
||||
}
|
||||
}
|
||||
|
||||
fn exec_options(allow_prefix: Option<Vec<String>>) -> Vec<ApprovalOption> {
|
||||
fn exec_options(proposed_execpolicy_amendment: Option<ExecPolicyAmendment>) -> Vec<ApprovalOption> {
|
||||
vec![
|
||||
ApprovalOption {
|
||||
label: "Yes, proceed".to_string(),
|
||||
@@ -486,10 +490,10 @@ fn exec_options(allow_prefix: Option<Vec<String>>) -> Vec<ApprovalOption> {
|
||||
},
|
||||
]
|
||||
.into_iter()
|
||||
.chain(allow_prefix.map(|prefix| ApprovalOption {
|
||||
.chain(proposed_execpolicy_amendment.map(|prefix| ApprovalOption {
|
||||
label: "Yes, and don't ask again for commands with this prefix".to_string(),
|
||||
decision: ApprovalDecision::Review(ReviewDecision::ApprovedAllowPrefix {
|
||||
allow_prefix: prefix,
|
||||
decision: ApprovalDecision::Review(ReviewDecision::ApprovedExecpolicyAmendment {
|
||||
proposed_execpolicy_amendment: prefix,
|
||||
}),
|
||||
display_shortcut: None,
|
||||
additional_shortcuts: vec![key_hint::plain(KeyCode::Char('p'))],
|
||||
@@ -556,7 +560,7 @@ mod tests {
|
||||
command: vec!["echo".to_string(), "hi".to_string()],
|
||||
reason: Some("reason".to_string()),
|
||||
risk: None,
|
||||
allow_prefix: None,
|
||||
proposed_execpolicy_amendment: None,
|
||||
}
|
||||
}
|
||||
|
||||
@@ -590,7 +594,7 @@ mod tests {
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn exec_prefix_option_emits_allow_prefix() {
|
||||
fn exec_prefix_option_emits_execpolicy_amendment() {
|
||||
let (tx, mut rx) = unbounded_channel::<AppEvent>();
|
||||
let tx = AppEventSender::new(tx);
|
||||
let mut view = ApprovalOverlay::new(
|
||||
@@ -599,7 +603,9 @@ mod tests {
|
||||
command: vec!["echo".to_string()],
|
||||
reason: None,
|
||||
risk: None,
|
||||
allow_prefix: Some(vec!["echo".to_string()]),
|
||||
proposed_execpolicy_amendment: Some(ExecPolicyAmendment::new(vec![
|
||||
"echo".to_string(),
|
||||
])),
|
||||
},
|
||||
tx,
|
||||
);
|
||||
@@ -609,8 +615,10 @@ mod tests {
|
||||
if let AppEvent::CodexOp(Op::ExecApproval { decision, .. }) = ev {
|
||||
assert_eq!(
|
||||
decision,
|
||||
ReviewDecision::ApprovedAllowPrefix {
|
||||
allow_prefix: vec!["echo".to_string()]
|
||||
ReviewDecision::ApprovedExecpolicyAmendment {
|
||||
proposed_execpolicy_amendment: ExecPolicyAmendment::new(vec![
|
||||
"echo".to_string()
|
||||
])
|
||||
}
|
||||
);
|
||||
saw_op = true;
|
||||
@@ -619,7 +627,7 @@ mod tests {
|
||||
}
|
||||
assert!(
|
||||
saw_op,
|
||||
"expected approval decision to emit an op with allow prefix"
|
||||
"expected approval decision to emit an op with command prefix"
|
||||
);
|
||||
}
|
||||
|
||||
@@ -633,7 +641,7 @@ mod tests {
|
||||
command,
|
||||
reason: None,
|
||||
risk: None,
|
||||
allow_prefix: None,
|
||||
proposed_execpolicy_amendment: None,
|
||||
};
|
||||
|
||||
let view = ApprovalOverlay::new(exec_request, tx);
|
||||
|
||||
@@ -570,7 +570,7 @@ mod tests {
|
||||
command: vec!["echo".into(), "ok".into()],
|
||||
reason: None,
|
||||
risk: None,
|
||||
allow_prefix: None,
|
||||
proposed_execpolicy_amendment: None,
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -1074,7 +1074,7 @@ impl ChatWidget {
|
||||
command: ev.command,
|
||||
reason: ev.reason,
|
||||
risk: ev.risk,
|
||||
allow_prefix: ev.allow_prefix,
|
||||
proposed_execpolicy_amendment: ev.proposed_execpolicy_amendment,
|
||||
};
|
||||
self.bottom_pane.push_approval_request(request);
|
||||
self.request_redraw();
|
||||
|
||||
@@ -23,6 +23,7 @@ use codex_core::protocol::ExecApprovalRequestEvent;
|
||||
use codex_core::protocol::ExecCommandBeginEvent;
|
||||
use codex_core::protocol::ExecCommandEndEvent;
|
||||
use codex_core::protocol::ExecCommandSource;
|
||||
use codex_core::protocol::ExecPolicyAmendment;
|
||||
use codex_core::protocol::ExitedReviewModeEvent;
|
||||
use codex_core::protocol::FileChange;
|
||||
use codex_core::protocol::Op;
|
||||
@@ -688,7 +689,7 @@ fn exec_approval_emits_proposed_command_and_decision_history() {
|
||||
"this is a test reason such as one that would be produced by the model".into(),
|
||||
),
|
||||
risk: None,
|
||||
allow_prefix: None,
|
||||
proposed_execpolicy_amendment: None,
|
||||
parsed_cmd: vec![],
|
||||
};
|
||||
chat.handle_codex_event(Event {
|
||||
@@ -733,7 +734,7 @@ fn exec_approval_decision_truncates_multiline_and_long_commands() {
|
||||
"this is a test reason such as one that would be produced by the model".into(),
|
||||
),
|
||||
risk: None,
|
||||
allow_prefix: None,
|
||||
proposed_execpolicy_amendment: None,
|
||||
parsed_cmd: vec![],
|
||||
};
|
||||
chat.handle_codex_event(Event {
|
||||
@@ -784,7 +785,7 @@ fn exec_approval_decision_truncates_multiline_and_long_commands() {
|
||||
cwd: std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
|
||||
reason: None,
|
||||
risk: None,
|
||||
allow_prefix: None,
|
||||
proposed_execpolicy_amendment: None,
|
||||
parsed_cmd: vec![],
|
||||
};
|
||||
chat.handle_codex_event(Event {
|
||||
@@ -1993,7 +1994,11 @@ fn approval_modal_exec_snapshot() {
|
||||
"this is a test reason such as one that would be produced by the model".into(),
|
||||
),
|
||||
risk: None,
|
||||
allow_prefix: Some(vec!["echo".into(), "hello".into(), "world".into()]),
|
||||
proposed_execpolicy_amendment: Some(ExecPolicyAmendment::new(vec![
|
||||
"echo".into(),
|
||||
"hello".into(),
|
||||
"world".into(),
|
||||
])),
|
||||
parsed_cmd: vec![],
|
||||
};
|
||||
chat.handle_codex_event(Event {
|
||||
@@ -2040,7 +2045,11 @@ fn approval_modal_exec_without_reason_snapshot() {
|
||||
cwd: std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
|
||||
reason: None,
|
||||
risk: None,
|
||||
allow_prefix: Some(vec!["echo".into(), "hello".into(), "world".into()]),
|
||||
proposed_execpolicy_amendment: Some(ExecPolicyAmendment::new(vec![
|
||||
"echo".into(),
|
||||
"hello".into(),
|
||||
"world".into(),
|
||||
])),
|
||||
parsed_cmd: vec![],
|
||||
};
|
||||
chat.handle_codex_event(Event {
|
||||
@@ -2254,7 +2263,10 @@ fn status_widget_and_approval_modal_snapshot() {
|
||||
"this is a test reason such as one that would be produced by the model".into(),
|
||||
),
|
||||
risk: None,
|
||||
allow_prefix: Some(vec!["echo".into(), "hello world".into()]),
|
||||
proposed_execpolicy_amendment: Some(ExecPolicyAmendment::new(vec![
|
||||
"echo".into(),
|
||||
"hello world".into(),
|
||||
])),
|
||||
parsed_cmd: vec![],
|
||||
};
|
||||
chat.handle_codex_event(Event {
|
||||
|
||||
@@ -409,7 +409,7 @@ pub fn new_approval_decision_cell(
|
||||
],
|
||||
)
|
||||
}
|
||||
ApprovedAllowPrefix { .. } => {
|
||||
ApprovedExecpolicyAmendment { .. } => {
|
||||
let snippet = Span::from(exec_snippet(&command)).dim();
|
||||
(
|
||||
"✔ ".green(),
|
||||
@@ -418,7 +418,7 @@ pub fn new_approval_decision_cell(
|
||||
"approved".bold(),
|
||||
" codex to run ".into(),
|
||||
snippet,
|
||||
" and added its prefix to your allow list".bold(),
|
||||
" and applied the execpolicy amendment".bold(),
|
||||
],
|
||||
)
|
||||
}
|
||||
|
||||
Reference in new issue
Block a user