diff --git a/codex-rs/app-server-protocol/schema/json/ClientRequest.json b/codex-rs/app-server-protocol/schema/json/ClientRequest.json index 93d08f7cb7..c015d664c5 100644 --- a/codex-rs/app-server-protocol/schema/json/ClientRequest.json +++ b/codex-rs/app-server-protocol/schema/json/ClientRequest.json @@ -5,6 +5,15 @@ "description": "A path that is guaranteed to be absolute and normalized (though it is not guaranteed to be canonicalized or exist on the filesystem).\n\nIMPORTANT: When deserializing an `AbsolutePathBuf`, a base path must be set using [AbsolutePathBufGuard::new]. If no base path is set, the deserialization will fail unless the path being deserialized is already absolute.", "type": "string" }, + "ApprovalSourceMetadata": { + "enum": [ + "guardian", + "user", + "policy", + "unknown" + ], + "type": "string" + }, "ApprovalsReviewer": { "description": "Configures who approval requests are routed to for review. Examples include sandbox escapes, blocked network access, MCP approval prompts, and ARC escalations. Defaults to `user`. `guardian_subagent` uses a carefully prompted subagent to gather relevant context and apply a risk-based decision framework before approving or denying the request.", "enum": [ @@ -2024,6 +2033,16 @@ }, "ResponseItemMetadata": { "properties": { + "approval_source": { + "anyOf": [ + { + "$ref": "#/definitions/ApprovalSourceMetadata" + }, + { + "type": "null" + } + ] + }, "is_tool_call_escalated": { "type": [ "boolean", diff --git a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json index 5708304428..c604ea661d 100644 --- a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json +++ b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json @@ -5296,6 +5296,15 @@ "AppToolsConfig": { "type": "object" }, + "ApprovalSourceMetadata": { + "enum": [ + "guardian", + "user", + "policy", + "unknown" + ], + "type": "string" + }, "ApprovalsReviewer": { "description": "Configures who approval requests are routed to for review. Examples include sandbox escapes, blocked network access, MCP approval prompts, and ARC escalations. Defaults to `user`. `guardian_subagent` uses a carefully prompted subagent to gather relevant context and apply a risk-based decision framework before approving or denying the request.", "enum": [ @@ -10747,6 +10756,16 @@ }, "ResponseItemMetadata": { "properties": { + "approval_source": { + "anyOf": [ + { + "$ref": "#/definitions/v2/ApprovalSourceMetadata" + }, + { + "type": "null" + } + ] + }, "is_tool_call_escalated": { "type": [ "boolean", diff --git a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json index f39fd1055c..0d9c191839 100644 --- a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json +++ b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json @@ -536,6 +536,15 @@ "AppToolsConfig": { "type": "object" }, + "ApprovalSourceMetadata": { + "enum": [ + "guardian", + "user", + "policy", + "unknown" + ], + "type": "string" + }, "ApprovalsReviewer": { "description": "Configures who approval requests are routed to for review. Examples include sandbox escapes, blocked network access, MCP approval prompts, and ARC escalations. Defaults to `user`. `guardian_subagent` uses a carefully prompted subagent to gather relevant context and apply a risk-based decision framework before approving or denying the request.", "enum": [ @@ -7515,6 +7524,16 @@ }, "ResponseItemMetadata": { "properties": { + "approval_source": { + "anyOf": [ + { + "$ref": "#/definitions/ApprovalSourceMetadata" + }, + { + "type": "null" + } + ] + }, "is_tool_call_escalated": { "type": [ "boolean", diff --git a/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json b/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json index adee557c4b..c5f8ec4b5f 100644 --- a/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json +++ b/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json @@ -1,6 +1,15 @@ { "$schema": "http://json-schema.org/draft-07/schema#", "definitions": { + "ApprovalSourceMetadata": { + "enum": [ + "guardian", + "user", + "policy", + "unknown" + ], + "type": "string" + }, "ContentItem": { "oneOf": [ { @@ -872,6 +881,16 @@ }, "ResponseItemMetadata": { "properties": { + "approval_source": { + "anyOf": [ + { + "$ref": "#/definitions/ApprovalSourceMetadata" + }, + { + "type": "null" + } + ] + }, "is_tool_call_escalated": { "type": [ "boolean", diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json index 226ec47f75..d73afba5ad 100644 --- a/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json +++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json @@ -1,6 +1,15 @@ { "$schema": "http://json-schema.org/draft-07/schema#", "definitions": { + "ApprovalSourceMetadata": { + "enum": [ + "guardian", + "user", + "policy", + "unknown" + ], + "type": "string" + }, "ApprovalsReviewer": { "description": "Configures who approval requests are routed to for review. Examples include sandbox escapes, blocked network access, MCP approval prompts, and ARC escalations. Defaults to `user`. `guardian_subagent` uses a carefully prompted subagent to gather relevant context and apply a risk-based decision framework before approving or denying the request.", "enum": [ @@ -938,6 +947,16 @@ }, "ResponseItemMetadata": { "properties": { + "approval_source": { + "anyOf": [ + { + "$ref": "#/definitions/ApprovalSourceMetadata" + }, + { + "type": "null" + } + ] + }, "is_tool_call_escalated": { "type": [ "boolean", diff --git a/codex-rs/app-server-protocol/schema/typescript/ApprovalSourceMetadata.ts b/codex-rs/app-server-protocol/schema/typescript/ApprovalSourceMetadata.ts new file mode 100644 index 0000000000..37da210732 --- /dev/null +++ b/codex-rs/app-server-protocol/schema/typescript/ApprovalSourceMetadata.ts @@ -0,0 +1,5 @@ +// GENERATED CODE! DO NOT MODIFY BY HAND! + +// This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually. + +export type ApprovalSourceMetadata = "guardian" | "user" | "policy" | "unknown"; diff --git a/codex-rs/app-server-protocol/schema/typescript/ResponseItemMetadata.ts b/codex-rs/app-server-protocol/schema/typescript/ResponseItemMetadata.ts index deb3c845bd..52a7bc1185 100644 --- a/codex-rs/app-server-protocol/schema/typescript/ResponseItemMetadata.ts +++ b/codex-rs/app-server-protocol/schema/typescript/ResponseItemMetadata.ts @@ -1,6 +1,7 @@ // GENERATED CODE! DO NOT MODIFY BY HAND! // This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually. +import type { ApprovalSourceMetadata } from "./ApprovalSourceMetadata"; import type { ReviewDecisionMetadata } from "./ReviewDecisionMetadata"; import type { SandboxPolicyMetadata } from "./SandboxPolicyMetadata"; import type { UserMessageType } from "./UserMessageType"; @@ -9,4 +10,4 @@ export type ResponseItemMetadata = { user_message_type?: UserMessageType, /** * Client-visible UUID generated by Codex for this item. */ -uuid?: string, sandbox_policy?: SandboxPolicyMetadata, is_tool_call_escalated?: boolean, review_decision?: ReviewDecisionMetadata, }; +uuid?: string, sandbox_policy?: SandboxPolicyMetadata, is_tool_call_escalated?: boolean, review_decision?: ReviewDecisionMetadata, approval_source?: ApprovalSourceMetadata, }; diff --git a/codex-rs/app-server-protocol/schema/typescript/index.ts b/codex-rs/app-server-protocol/schema/typescript/index.ts index a963406a5c..e3124672e4 100644 --- a/codex-rs/app-server-protocol/schema/typescript/index.ts +++ b/codex-rs/app-server-protocol/schema/typescript/index.ts @@ -3,6 +3,7 @@ export type { AbsolutePathBuf } from "./AbsolutePathBuf"; export type { ApplyPatchApprovalParams } from "./ApplyPatchApprovalParams"; export type { ApplyPatchApprovalResponse } from "./ApplyPatchApprovalResponse"; +export type { ApprovalSourceMetadata } from "./ApprovalSourceMetadata"; export type { AuthMode } from "./AuthMode"; export type { ClientInfo } from "./ClientInfo"; export type { ClientNotification } from "./ClientNotification"; diff --git a/codex-rs/core/src/codex.rs b/codex-rs/core/src/codex.rs index cd11b99b6e..5d9a107a65 100644 --- a/codex-rs/core/src/codex.rs +++ b/codex-rs/core/src/codex.rs @@ -88,6 +88,7 @@ use codex_protocol::items::TurnItem; use codex_protocol::items::UserMessageItem; use codex_protocol::items::build_hook_prompt_message; use codex_protocol::mcp::CallToolResult; +use codex_protocol::models::ApprovalSourceMetadata; use codex_protocol::models::BaseInstructions; use codex_protocol::models::PermissionProfile; use codex_protocol::models::ResponseItemMessageMetadata; @@ -209,6 +210,7 @@ use crate::file_watcher::FileWatcher; use crate::file_watcher::FileWatcherEvent; use crate::git_info::get_git_repo_root; use crate::guardian::GuardianReviewSessionManager; +use crate::guardian::routes_approval_to_guardian; use crate::hook_runtime::PendingInputHookDisposition; use crate::hook_runtime::inspect_pending_input; use crate::hook_runtime::record_additional_contexts; @@ -1052,6 +1054,13 @@ fn user_message_metadata_patch( } } +fn approval_source_for_turn(turn_context: &TurnContext) -> ApprovalSourceMetadata { + if routes_approval_to_guardian(turn_context) { + ApprovalSourceMetadata::Guardian + } else { + ApprovalSourceMetadata::User + } +} fn response_item_tool_call_id(item: &ResponseItem) -> Option<&str> { match item { ResponseItem::LocalShellCall { @@ -1135,10 +1144,12 @@ fn stamp_tool_approval_metadata_with_snapshot( Some(outcome) => { metadata.is_tool_call_escalated = Some(true); metadata.review_decision = outcome.review_decision; + metadata.approval_source = Some(outcome.approval_source); } None if !has_pending_approval => { metadata.is_tool_call_escalated = Some(false); metadata.review_decision = None; + metadata.approval_source = None; } None => { return stamp_tool_metadata_on_response_item(response_item, metadata); @@ -3017,6 +3028,7 @@ impl Session { effective_approval_id.clone(), PendingApprovalMetadata { call_id: call_id.clone(), + approval_source: approval_source_for_turn(turn_context), }, ); ts.insert_pending_approval(effective_approval_id.clone(), tx_approve) @@ -3088,6 +3100,7 @@ impl Session { approval_id.clone(), PendingApprovalMetadata { call_id: call_id.clone(), + approval_source: approval_source_for_turn(turn_context), }, ); ts.insert_pending_approval(approval_id.clone(), tx_approve) @@ -3385,16 +3398,16 @@ impl Session { .remove_pending_approval_call_id(approval_id) .unwrap_or_else(|| PendingApprovalMetadata { call_id: approval_id.to_string(), + approval_source: ApprovalSourceMetadata::Unknown, }); drop(ts); drop(active); self.record_call_approval_outcome( pending_approval.call_id, - ApprovalOutcomeMetadata::reviewed(decision), + ApprovalOutcomeMetadata::reviewed(decision, pending_approval.approval_source), ) .await; } - pub(crate) async fn record_call_approval_outcome( &self, call_id: String, diff --git a/codex-rs/core/src/codex_tests.rs b/codex-rs/core/src/codex_tests.rs index 7854e3222b..7b657fda44 100644 --- a/codex-rs/core/src/codex_tests.rs +++ b/codex-rs/core/src/codex_tests.rs @@ -4418,52 +4418,100 @@ async fn task_finish_emits_prompt_queued_metadata_for_injected_user_input_when_f #[test] fn review_decision_metadata_mapping_is_stable() { assert_eq!( - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::Approved).review_decision, + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::Approved, + codex_protocol::models::ApprovalSourceMetadata::User, + ) + .review_decision, Some(codex_protocol::models::ReviewDecisionMetadata::Approved) ); assert_eq!( - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::Denied).review_decision, + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::Denied, + codex_protocol::models::ApprovalSourceMetadata::User, + ) + .review_decision, Some(codex_protocol::models::ReviewDecisionMetadata::Denied) ); assert_eq!( - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::Abort).review_decision, + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::Abort, + codex_protocol::models::ApprovalSourceMetadata::User, + ) + .review_decision, Some(codex_protocol::models::ReviewDecisionMetadata::Abort) ); assert_eq!( - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::ApprovedForSession).review_decision, + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::ApprovedForSession, + codex_protocol::models::ApprovalSourceMetadata::User, + ) + .review_decision, Some(codex_protocol::models::ReviewDecisionMetadata::ApprovedForSession) ); assert_eq!( - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::ApprovedExecpolicyAmendment { - proposed_execpolicy_amendment: codex_protocol::approvals::ExecPolicyAmendment::new( - vec!["echo".to_string(), "hi".to_string()], - ), - }) + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::ApprovedExecpolicyAmendment { + proposed_execpolicy_amendment: codex_protocol::approvals::ExecPolicyAmendment::new( + vec!["echo".to_string(), "hi".to_string()], + ), + }, + codex_protocol::models::ApprovalSourceMetadata::User, + ) .review_decision, Some(codex_protocol::models::ReviewDecisionMetadata::ApprovedWithAmendment) ); assert_eq!( - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::NetworkPolicyAmendment { - network_policy_amendment: codex_protocol::approvals::NetworkPolicyAmendment { - host: "example.com".to_string(), - action: codex_protocol::protocol::NetworkPolicyRuleAction::Allow, + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::NetworkPolicyAmendment { + network_policy_amendment: codex_protocol::approvals::NetworkPolicyAmendment { + host: "example.com".to_string(), + action: NetworkPolicyRuleAction::Allow, + }, }, - }) + codex_protocol::models::ApprovalSourceMetadata::User, + ) .review_decision, Some(codex_protocol::models::ReviewDecisionMetadata::ApprovedWithNetworkPolicyAllow) ); assert_eq!( - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::NetworkPolicyAmendment { - network_policy_amendment: codex_protocol::approvals::NetworkPolicyAmendment { - host: "example.com".to_string(), - action: codex_protocol::protocol::NetworkPolicyRuleAction::Deny, + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::NetworkPolicyAmendment { + network_policy_amendment: codex_protocol::approvals::NetworkPolicyAmendment { + host: "example.com".to_string(), + action: NetworkPolicyRuleAction::Deny, + }, }, - }) + codex_protocol::models::ApprovalSourceMetadata::User, + ) .review_decision, Some(codex_protocol::models::ReviewDecisionMetadata::DeniedWithNetworkPolicyDeny) ); } +#[tokio::test] +async fn approval_source_metadata_mapping_is_stable() { + let (_session, mut turn_context) = make_session_and_context().await; + + assert_eq!( + approval_source_for_turn(&turn_context), + codex_protocol::models::ApprovalSourceMetadata::User + ); + + turn_context + .approval_policy + .set(crate::protocol::AskForApproval::OnRequest) + .expect("test setup should allow updating approval policy"); + Arc::get_mut(&mut turn_context.config) + .expect("single turn config ref") + .approvals_reviewer = codex_protocol::config_types::ApprovalsReviewer::GuardianSubagent; + + assert_eq!( + approval_source_for_turn(&turn_context), + codex_protocol::models::ApprovalSourceMetadata::Guardian + ); +} + #[test] fn sandbox_policy_metadata_mapping_is_stable() { assert_eq!( @@ -4527,6 +4575,7 @@ async fn assert_next_emitted_function_call_metadata( expected_sandbox_policy: codex_protocol::models::SandboxPolicyMetadata, expected_escalated: bool, expected_review_decision: Option, + expected_approval_source: Option, ) { let event = tokio::time::timeout(std::time::Duration::from_secs(2), rx.recv()) .await @@ -4542,6 +4591,7 @@ async fn assert_next_emitted_function_call_metadata( .. } if metadata.is_tool_call_escalated == Some(expected_escalated) && metadata.review_decision == expected_review_decision + && metadata.approval_source == expected_approval_source && metadata.sandbox_policy == Some(expected_sandbox_policy) ) )); @@ -4553,7 +4603,10 @@ async fn tool_call_metadata_stamps_escalated_review_decision_when_feature_enable sess.record_call_approval_outcome( "call-1".to_string(), - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::Denied), + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::Denied, + codex_protocol::models::ApprovalSourceMetadata::User, + ), ) .await; sess.record_response_item_and_emit_turn_item(tc.as_ref(), function_call_item("call-1")) @@ -4563,6 +4616,7 @@ async fn tool_call_metadata_stamps_escalated_review_decision_when_feature_enable expected_sandbox_policy, true, Some(codex_protocol::models::ReviewDecisionMetadata::Denied), + Some(codex_protocol::models::ApprovalSourceMetadata::User), ) .await; } @@ -4573,7 +4627,62 @@ async fn tool_call_metadata_stamps_non_escalated_false_when_feature_enabled() { sess.record_response_item_and_emit_turn_item(tc.as_ref(), function_call_item("call-2")) .await; - assert_next_emitted_function_call_metadata(&rx, expected_sandbox_policy, false, None).await; + assert_next_emitted_function_call_metadata(&rx, expected_sandbox_policy, false, None, None) + .await; +} + +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] +async fn tool_call_metadata_stamps_guardian_direct_review_when_feature_enabled() { + let (sess, tc, rx, expected_sandbox_policy) = setup_tool_call_metadata_runtime_test().await; + + sess.record_call_approval_outcome( + "call-guardian-runtime-1".to_string(), + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::Denied, + codex_protocol::models::ApprovalSourceMetadata::Guardian, + ), + ) + .await; + sess.record_response_item_and_emit_turn_item( + tc.as_ref(), + function_call_item("call-guardian-runtime-1"), + ) + .await; + assert_next_emitted_function_call_metadata( + &rx, + expected_sandbox_policy, + true, + Some(codex_protocol::models::ReviewDecisionMetadata::Denied), + Some(codex_protocol::models::ApprovalSourceMetadata::Guardian), + ) + .await; +} + +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] +async fn tool_call_metadata_stamps_policy_source_without_review_decision_when_feature_enabled() { + let (sess, tc, rx, expected_sandbox_policy) = setup_tool_call_metadata_runtime_test().await; + + sess.record_call_approval_outcome( + "call-policy-runtime-1".to_string(), + ApprovalOutcomeMetadata { + review_decision: None, + approval_source: codex_protocol::models::ApprovalSourceMetadata::Policy, + }, + ) + .await; + sess.record_response_item_and_emit_turn_item( + tc.as_ref(), + function_call_item("call-policy-runtime-1"), + ) + .await; + assert_next_emitted_function_call_metadata( + &rx, + expected_sandbox_policy, + true, + None, + Some(codex_protocol::models::ApprovalSourceMetadata::Policy), + ) + .await; } #[tokio::test(flavor = "multi_thread", worker_threads = 2)] @@ -4629,6 +4738,7 @@ async fn handle_output_item_done_stamps_tool_call_metadata_when_feature_enabled( } if call_id == "call-stream-1" && metadata.is_tool_call_escalated == Some(false) && metadata.review_decision.is_none() + && metadata.approval_source.is_none() && metadata.sandbox_policy == Some(expected_sandbox_policy.clone()) ) })); @@ -4671,7 +4781,10 @@ async fn tool_call_metadata_can_be_restamped_after_approval_outcome() { .await; sess.record_call_approval_outcome( "call-restamp-1".to_string(), - ApprovalOutcomeMetadata::reviewed(&ReviewDecision::Denied), + ApprovalOutcomeMetadata::reviewed( + &ReviewDecision::Denied, + codex_protocol::models::ApprovalSourceMetadata::User, + ), ) .await; @@ -4701,6 +4814,89 @@ async fn tool_call_metadata_can_be_restamped_after_approval_outcome() { } if metadata.is_tool_call_escalated == Some(true) && metadata.review_decision == Some(codex_protocol::models::ReviewDecisionMetadata::Denied) + && metadata.approval_source + == Some(codex_protocol::models::ApprovalSourceMetadata::User) + && metadata.sandbox_policy == Some(expected_sandbox_policy) + )); +} + +#[tokio::test] +async fn tool_call_metadata_snapshot_stamps_guardian_approval_source() { + let (_session, turn_context_raw) = make_session_and_context().await; + let turn_context = Arc::new(turn_context_raw); + let expected_sandbox_policy = sandbox_policy_to_metadata(turn_context.sandbox_policy.get()); + let item = ResponseItem::FunctionCall { + id: None, + name: "shell".to_string(), + namespace: None, + arguments: "{}".to_string(), + call_id: "call-guardian-1".to_string(), + metadata: None, + }; + let snapshot = ToolApprovalMetadataSnapshot { + approval_outcomes_by_call_id: HashMap::from([( + "call-guardian-1".to_string(), + ApprovalOutcomeMetadata { + review_decision: Some(codex_protocol::models::ReviewDecisionMetadata::Denied), + approval_source: codex_protocol::models::ApprovalSourceMetadata::Guardian, + }, + )]), + pending_approval_call_ids: HashSet::new(), + }; + + let stamped = + stamp_tool_approval_metadata_with_snapshot(turn_context.as_ref(), item, Some(&snapshot)); + + assert!(matches!( + stamped, + ResponseItem::FunctionCall { + metadata: Some(metadata), + .. + } if metadata.is_tool_call_escalated == Some(true) + && metadata.review_decision + == Some(codex_protocol::models::ReviewDecisionMetadata::Denied) + && metadata.approval_source + == Some(codex_protocol::models::ApprovalSourceMetadata::Guardian) + && metadata.sandbox_policy == Some(expected_sandbox_policy) + )); +} + +#[tokio::test] +async fn tool_call_metadata_snapshot_stamps_policy_approval_source_without_review_decision() { + let (_session, turn_context_raw) = make_session_and_context().await; + let turn_context = Arc::new(turn_context_raw); + let expected_sandbox_policy = sandbox_policy_to_metadata(turn_context.sandbox_policy.get()); + let item = ResponseItem::FunctionCall { + id: None, + name: "shell".to_string(), + namespace: None, + arguments: "{}".to_string(), + call_id: "call-policy-1".to_string(), + metadata: None, + }; + let snapshot = ToolApprovalMetadataSnapshot { + approval_outcomes_by_call_id: HashMap::from([( + "call-policy-1".to_string(), + ApprovalOutcomeMetadata { + review_decision: None, + approval_source: codex_protocol::models::ApprovalSourceMetadata::Policy, + }, + )]), + pending_approval_call_ids: HashSet::new(), + }; + + let stamped = + stamp_tool_approval_metadata_with_snapshot(turn_context.as_ref(), item, Some(&snapshot)); + + assert!(matches!( + stamped, + ResponseItem::FunctionCall { + metadata: Some(metadata), + .. + } if metadata.is_tool_call_escalated == Some(true) + && metadata.review_decision.is_none() + && metadata.approval_source + == Some(codex_protocol::models::ApprovalSourceMetadata::Policy) && metadata.sandbox_policy == Some(expected_sandbox_policy) )); } diff --git a/codex-rs/core/src/mcp_tool_call.rs b/codex-rs/core/src/mcp_tool_call.rs index 3e7f0cb84f..105f85eefd 100644 --- a/codex-rs/core/src/mcp_tool_call.rs +++ b/codex-rs/core/src/mcp_tool_call.rs @@ -31,9 +31,11 @@ use crate::protocol::EventMsg; use crate::protocol::McpInvocation; use crate::protocol::McpToolCallBeginEvent; use crate::protocol::McpToolCallEndEvent; +use crate::state::ApprovalOutcomeMetadata; use crate::state_db; use codex_features::Feature; use codex_protocol::mcp::CallToolResult; +use codex_protocol::models::ApprovalSourceMetadata; use codex_protocol::openai_models::InputModality; use codex_protocol::protocol::AskForApproval; use codex_protocol::protocol::ReviewDecision; @@ -545,6 +547,11 @@ async fn maybe_request_mcp_tool_approval( monitor_reason.clone(), ) .await; + sess.record_call_approval_outcome( + call_id.to_string(), + ApprovalOutcomeMetadata::reviewed(&decision, ApprovalSourceMetadata::Guardian), + ) + .await; let decision = mcp_tool_approval_decision_from_guardian(decision); apply_mcp_tool_approval_decision( sess, diff --git a/codex-rs/core/src/state/turn.rs b/codex-rs/core/src/state/turn.rs index 1e2b7a24bb..388ee43efc 100644 --- a/codex-rs/core/src/state/turn.rs +++ b/codex-rs/core/src/state/turn.rs @@ -10,6 +10,7 @@ use tokio_util::sync::CancellationToken; use tokio_util::task::AbortOnDropHandle; use codex_protocol::dynamic_tools::DynamicToolResponse; +use codex_protocol::models::ApprovalSourceMetadata; use codex_protocol::models::ResponseInputItem; use codex_protocol::models::ReviewDecisionMetadata; use codex_protocol::request_permissions::RequestPermissionsResponse; @@ -61,15 +62,20 @@ pub(crate) struct RunningTask { #[derive(Clone, Debug, PartialEq, Eq)] pub(crate) struct PendingApprovalMetadata { pub(crate) call_id: String, + pub(crate) approval_source: ApprovalSourceMetadata, } #[derive(Clone, Debug, PartialEq, Eq)] pub(crate) struct ApprovalOutcomeMetadata { pub(crate) review_decision: Option, + pub(crate) approval_source: ApprovalSourceMetadata, } impl ApprovalOutcomeMetadata { - pub(crate) fn reviewed(decision: &ReviewDecision) -> Self { + pub(crate) fn reviewed( + decision: &ReviewDecision, + approval_source: ApprovalSourceMetadata, + ) -> Self { let review_decision = match decision { ReviewDecision::Approved => ReviewDecisionMetadata::Approved, ReviewDecision::ApprovedExecpolicyAmendment { .. } => { @@ -91,6 +97,7 @@ impl ApprovalOutcomeMetadata { }; Self { review_decision: Some(review_decision), + approval_source, } } } diff --git a/codex-rs/core/src/tools/runtimes/apply_patch.rs b/codex-rs/core/src/tools/runtimes/apply_patch.rs index f1e9912bc5..7cceaccca8 100644 --- a/codex-rs/core/src/tools/runtimes/apply_patch.rs +++ b/codex-rs/core/src/tools/runtimes/apply_patch.rs @@ -12,6 +12,7 @@ use crate::guardian::routes_approval_to_guardian; use crate::sandboxing::CommandSpec; use crate::sandboxing::SandboxPermissions; use crate::sandboxing::execute_env; +use crate::state::ApprovalOutcomeMetadata; use crate::tools::sandboxing::Approvable; use crate::tools::sandboxing::ApprovalCtx; use crate::tools::sandboxing::ExecApprovalRequirement; @@ -24,6 +25,7 @@ use crate::tools::sandboxing::ToolRuntime; use crate::tools::sandboxing::with_cached_approval; use codex_apply_patch::ApplyPatchAction; use codex_apply_patch::CODEX_CORE_APPLY_PATCH_ARG1; +use codex_protocol::models::ApprovalSourceMetadata; use codex_protocol::models::PermissionProfile; use codex_protocol::protocol::AskForApproval; use codex_protocol::protocol::FileChange; @@ -142,7 +144,17 @@ impl Approvable for ApplyPatchRuntime { Box::pin(async move { if routes_approval_to_guardian(turn) { let action = ApplyPatchRuntime::build_guardian_review_request(req, ctx.call_id); - return review_approval_request(session, turn, action, retry_reason).await; + let decision = review_approval_request(session, turn, action, retry_reason).await; + session + .record_call_approval_outcome( + call_id.clone(), + ApprovalOutcomeMetadata::reviewed( + &decision, + ApprovalSourceMetadata::Guardian, + ), + ) + .await; + return decision; } if req.permissions_preapproved && retry_reason.is_none() { return ReviewDecision::Approved; diff --git a/codex-rs/core/src/tools/runtimes/shell.rs b/codex-rs/core/src/tools/runtimes/shell.rs index 18afb20bc5..00fa223870 100644 --- a/codex-rs/core/src/tools/runtimes/shell.rs +++ b/codex-rs/core/src/tools/runtimes/shell.rs @@ -17,6 +17,7 @@ use crate::powershell::prefix_powershell_script_with_utf8; use crate::sandboxing::SandboxPermissions; use crate::sandboxing::execute_env; use crate::shell::ShellType; +use crate::state::ApprovalOutcomeMetadata; use crate::tools::network_approval::NetworkApprovalMode; use crate::tools::network_approval::NetworkApprovalSpec; use crate::tools::runtimes::build_command_spec; @@ -35,6 +36,7 @@ use crate::tools::sandboxing::sandbox_override_for_first_attempt; use crate::tools::sandboxing::with_cached_approval; use codex_features::Feature; use codex_network_proxy::NetworkProxy; +use codex_protocol::models::ApprovalSourceMetadata; use codex_protocol::models::PermissionProfile; use codex_protocol::protocol::ReviewDecision; use futures::future::BoxFuture; @@ -153,11 +155,11 @@ impl Approvable for ShellRuntime { let call_id = ctx.call_id.to_string(); Box::pin(async move { if routes_approval_to_guardian(turn) { - return review_approval_request( + let decision = review_approval_request( session, turn, GuardianApprovalRequest::Shell { - id: call_id, + id: call_id.clone(), command, cwd, sandbox_permissions: req.sandbox_permissions, @@ -167,6 +169,16 @@ impl Approvable for ShellRuntime { retry_reason, ) .await; + session + .record_call_approval_outcome( + call_id.clone(), + ApprovalOutcomeMetadata::reviewed( + &decision, + ApprovalSourceMetadata::Guardian, + ), + ) + .await; + return decision; } with_cached_approval(&session.services, "shell", keys, move || async move { let available_decisions = None; diff --git a/codex-rs/core/src/tools/runtimes/shell/unix_escalation.rs b/codex-rs/core/src/tools/runtimes/shell/unix_escalation.rs index 948018dae6..eca9cb5629 100644 --- a/codex-rs/core/src/tools/runtimes/shell/unix_escalation.rs +++ b/codex-rs/core/src/tools/runtimes/shell/unix_escalation.rs @@ -13,6 +13,7 @@ use crate::sandboxing::ExecRequest; use crate::sandboxing::SandboxPermissions; use crate::shell::ShellType; use crate::skills::SkillMetadata; +use crate::state::ApprovalOutcomeMetadata; use crate::tools::runtimes::ExecveSessionApproval; use crate::tools::runtimes::build_command_spec; use crate::tools::sandboxing::SandboxAttempt; @@ -26,6 +27,7 @@ use codex_execpolicy::Policy; use codex_execpolicy::RuleMatch; use codex_features::Feature; use codex_protocol::config_types::WindowsSandboxLevel; +use codex_protocol::models::ApprovalSourceMetadata; use codex_protocol::models::MacOsSeatbeltProfileExtensions; use codex_protocol::models::PermissionProfile; use codex_protocol::permissions::FileSystemSandboxPolicy; @@ -426,7 +428,7 @@ impl CoreShellActionProvider { Ok(stopwatch .pause_for(async move { if routes_approval_to_guardian(&turn) { - return review_approval_request( + let decision = review_approval_request( &session, &turn, GuardianApprovalRequest::Execve { @@ -440,6 +442,16 @@ impl CoreShellActionProvider { /*retry_reason*/ None, ) .await; + session + .record_call_approval_outcome( + call_id.clone(), + ApprovalOutcomeMetadata::reviewed( + &decision, + ApprovalSourceMetadata::Guardian, + ), + ) + .await; + return decision; } let available_decisions = vec![ Some(ReviewDecision::Approved), @@ -522,12 +534,31 @@ impl CoreShellActionProvider { ) -> anyhow::Result { let action = match decision { Decision::Forbidden => { + self.session + .record_call_approval_outcome( + self.call_id.clone(), + ApprovalOutcomeMetadata { + review_decision: None, + approval_source: codex_protocol::models::ApprovalSourceMetadata::Policy, + }, + ) + .await; EscalationDecision::deny(Some("Execution forbidden by policy".to_string())) } Decision::Prompt => { if execve_prompt_is_rejected_by_policy(self.approval_policy, &decision_source) .is_some() { + self.session + .record_call_approval_outcome( + self.call_id.clone(), + ApprovalOutcomeMetadata { + review_decision: None, + approval_source: + codex_protocol::models::ApprovalSourceMetadata::Policy, + }, + ) + .await; EscalationDecision::deny(Some("Execution forbidden by policy".to_string())) } else { match self diff --git a/codex-rs/core/src/tools/runtimes/unified_exec.rs b/codex-rs/core/src/tools/runtimes/unified_exec.rs index 0b64092156..23356410be 100644 --- a/codex-rs/core/src/tools/runtimes/unified_exec.rs +++ b/codex-rs/core/src/tools/runtimes/unified_exec.rs @@ -14,6 +14,7 @@ use crate::guardian::routes_approval_to_guardian; use crate::powershell::prefix_powershell_script_with_utf8; use crate::sandboxing::SandboxPermissions; use crate::shell::ShellType; +use crate::state::ApprovalOutcomeMetadata; use crate::tools::network_approval::NetworkApprovalMode; use crate::tools::network_approval::NetworkApprovalSpec; use crate::tools::runtimes::build_command_spec; @@ -38,6 +39,7 @@ use crate::unified_exec::UnifiedExecProcess; use crate::unified_exec::UnifiedExecProcessManager; use codex_features::Feature; use codex_network_proxy::NetworkProxy; +use codex_protocol::models::ApprovalSourceMetadata; use codex_protocol::models::PermissionProfile; use codex_protocol::protocol::ReviewDecision; use futures::future::BoxFuture; @@ -121,11 +123,11 @@ impl Approvable for UnifiedExecRuntime<'_> { let reason = retry_reason.clone().or_else(|| req.justification.clone()); Box::pin(async move { if routes_approval_to_guardian(turn) { - return review_approval_request( + let decision = review_approval_request( session, turn, GuardianApprovalRequest::ExecCommand { - id: call_id, + id: call_id.clone(), command, cwd, sandbox_permissions: req.sandbox_permissions, @@ -136,6 +138,16 @@ impl Approvable for UnifiedExecRuntime<'_> { retry_reason, ) .await; + session + .record_call_approval_outcome( + call_id.clone(), + ApprovalOutcomeMetadata::reviewed( + &decision, + ApprovalSourceMetadata::Guardian, + ), + ) + .await; + return decision; } with_cached_approval(&session.services, "unified_exec", keys, || async move { let available_decisions = None; diff --git a/codex-rs/protocol/src/models.rs b/codex-rs/protocol/src/models.rs index 6b15a4009c..d9fd95f9bf 100644 --- a/codex-rs/protocol/src/models.rs +++ b/codex-rs/protocol/src/models.rs @@ -291,6 +291,16 @@ pub enum SandboxPolicyMetadata { FullAccess, } +#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq, JsonSchema, TS)] +#[serde(rename_all = "snake_case")] +#[ts(rename_all = "snake_case")] +pub enum ApprovalSourceMetadata { + Guardian, + User, + Policy, + Unknown, +} + #[derive(Debug, Clone, Default, Serialize, Deserialize, PartialEq, Eq, JsonSchema, TS)] pub struct ResponseItemMetadata { #[serde(default, skip_serializing_if = "Option::is_none")] @@ -309,6 +319,9 @@ pub struct ResponseItemMetadata { #[serde(default, skip_serializing_if = "Option::is_none")] #[ts(optional)] pub review_decision: Option, + #[serde(default, skip_serializing_if = "Option::is_none")] + #[ts(optional)] + pub approval_source: Option, } impl ResponseItemMetadata { @@ -316,6 +329,7 @@ impl ResponseItemMetadata { self.sandbox_policy.is_none() && self.is_tool_call_escalated.is_none() && self.review_decision.is_none() + && self.approval_source.is_none() } pub fn merge_from(&mut self, other: Self) { @@ -328,6 +342,9 @@ impl ResponseItemMetadata { if other.review_decision.is_some() { self.review_decision = other.review_decision; } + if other.approval_source.is_some() { + self.approval_source = other.approval_source; + } } }