From 5d6086366844130b032f0761de2297c738281db8 Mon Sep 17 00:00:00 2001 From: rhan-oai Date: Tue, 21 Apr 2026 21:34:38 -0700 Subject: [PATCH] [codex-analytics] guardian review truncation --- .../core/src/guardian/approval_request.rs | 62 ++++++---- codex-rs/core/src/guardian/prompt.rs | 16 +-- codex-rs/core/src/guardian/review_session.rs | 85 ++++++++------ codex-rs/core/src/guardian/tests.rs | 106 ++++++++++++++++-- 4 files changed, 195 insertions(+), 74 deletions(-) diff --git a/codex-rs/core/src/guardian/approval_request.rs b/codex-rs/core/src/guardian/approval_request.rs index 471e054429..989c6ee119 100644 --- a/codex-rs/core/src/guardian/approval_request.rs +++ b/codex-rs/core/src/guardian/approval_request.rs @@ -184,32 +184,49 @@ fn guardian_command_source_tool_name(source: GuardianCommandSource) -> &'static } } -fn truncate_guardian_action_value(value: Value) -> Value { +fn truncate_guardian_action_value(value: Value) -> (Value, bool) { match value { - Value::String(text) => Value::String(guardian_truncate_text( - &text, - GUARDIAN_MAX_ACTION_STRING_TOKENS, - )), - Value::Array(values) => Value::Array( - values + Value::String(text) => { + let (text, truncated) = + guardian_truncate_text(&text, GUARDIAN_MAX_ACTION_STRING_TOKENS); + (Value::String(text), truncated) + } + Value::Array(values) => { + let mut truncated = false; + let values = values .into_iter() - .map(truncate_guardian_action_value) - .collect::>(), - ), + .map(|value| { + let (value, value_truncated) = truncate_guardian_action_value(value); + truncated |= value_truncated; + value + }) + .collect::>(); + (Value::Array(values), truncated) + } Value::Object(values) => { let mut entries = values.into_iter().collect::>(); entries.sort_by(|(left, _), (right, _)| left.cmp(right)); - Value::Object( - entries - .into_iter() - .map(|(key, value)| (key, truncate_guardian_action_value(value))) - .collect(), - ) + let mut truncated = false; + let values = entries + .into_iter() + .map(|(key, value)| { + let (value, value_truncated) = truncate_guardian_action_value(value); + truncated |= value_truncated; + (key, value) + }) + .collect(); + (Value::Object(values), truncated) } - other => other, + other => (other, false), } } +#[derive(Debug, Clone, PartialEq, Eq)] +pub(crate) struct FormattedGuardianAction { + pub(crate) text: String, + pub(crate) truncated: bool, +} + pub(crate) fn guardian_approval_request_to_json( action: &GuardianApprovalRequest, ) -> serde_json::Result { @@ -482,8 +499,11 @@ pub(crate) fn guardian_request_turn_id<'a>( pub(crate) fn format_guardian_action_pretty( action: &GuardianApprovalRequest, -) -> serde_json::Result { - let mut value = guardian_approval_request_to_json(action)?; - value = truncate_guardian_action_value(value); - serde_json::to_string_pretty(&value) +) -> serde_json::Result { + let value = guardian_approval_request_to_json(action)?; + let (value, truncated) = truncate_guardian_action_value(value); + Ok(FormattedGuardianAction { + text: serde_json::to_string_pretty(&value)?, + truncated, + }) } diff --git a/codex-rs/core/src/guardian/prompt.rs b/codex-rs/core/src/guardian/prompt.rs index 1a716f25b9..adf9dd8d70 100644 --- a/codex-rs/core/src/guardian/prompt.rs +++ b/codex-rs/core/src/guardian/prompt.rs @@ -59,6 +59,7 @@ impl GuardianTranscriptEntryKind { pub(crate) struct GuardianPromptItems { pub(crate) items: Vec, pub(crate) transcript_cursor: GuardianTranscriptCursor, + pub(crate) reviewed_action_truncated: bool, } /// Points to the end of the transcript that the guardian has already reviewed. @@ -179,11 +180,12 @@ pub(crate) async fn build_guardian_prompt_items( .to_string(), ); push_text("Planned action JSON:\n".to_string()); - push_text(format!("{planned_action_json}\n")); + push_text(format!("{}\n", planned_action_json.text)); push_text(">>> APPROVAL REQUEST END\n".to_string()); Ok(GuardianPromptItems { items, transcript_cursor, + reviewed_action_truncated: planned_action_json.truncated, }) } @@ -243,7 +245,7 @@ fn render_guardian_transcript_entries_with_offset( } else { GUARDIAN_MAX_MESSAGE_ENTRY_TOKENS }; - let text = guardian_truncate_text(&entry.text, token_cap); + let (text, _) = guardian_truncate_text(&entry.text, token_cap); let rendered = format!( "[{}] {}: {}", index + entry_number_offset + 1, @@ -423,20 +425,20 @@ pub(crate) fn collect_guardian_transcript_entries( entries } -pub(crate) fn guardian_truncate_text(content: &str, token_cap: usize) -> String { +pub(crate) fn guardian_truncate_text(content: &str, token_cap: usize) -> (String, bool) { if content.is_empty() { - return String::new(); + return (String::new(), false); } let max_bytes = approx_bytes_for_tokens(token_cap); if content.len() <= max_bytes { - return content.to_string(); + return (content.to_string(), false); } let omitted_tokens = approx_tokens_from_byte_count(content.len().saturating_sub(max_bytes)); let marker = format!("<{TRUNCATION_TAG} omitted_approx_tokens=\"{omitted_tokens}\" />"); if max_bytes <= marker.len() { - return marker; + return (marker, true); } let available_bytes = max_bytes.saturating_sub(marker.len()); @@ -444,7 +446,7 @@ pub(crate) fn guardian_truncate_text(content: &str, token_cap: usize) -> String let suffix_budget = available_bytes.saturating_sub(prefix_budget); let (prefix, suffix) = split_guardian_truncation_bounds(content, prefix_budget, suffix_budget); - format!("{prefix}{marker}{suffix}") + (format!("{prefix}{marker}{suffix}"), true) } fn split_guardian_truncation_bounds( diff --git a/codex-rs/core/src/guardian/review_session.rs b/codex-rs/core/src/guardian/review_session.rs index 4368b38da6..087872be65 100644 --- a/codex-rs/core/src/guardian/review_session.rs +++ b/codex-rs/core/src/guardian/review_session.rs @@ -635,7 +635,7 @@ async fn run_review_on_session( append_guardian_followup_reminder(review_session).await; } - let submit_result = run_before_review_deadline( + let prompt_items = run_before_review_deadline( deadline, params.external_cancel.as_ref(), Box::pin(async { @@ -648,59 +648,70 @@ async fn run_review_on_session( ) .await; - let prompt_items = build_guardian_prompt_items( + build_guardian_prompt_items( params.parent_session.as_ref(), params.retry_reason.clone(), params.request.clone(), prompt_mode, ) - .await?; - let token_usage_at_review_start = review_session - .codex - .session - .total_token_usage() - .await - .unwrap_or_default(); - - review_session - .codex - .submit(Op::UserTurn { - items: prompt_items.items, - cwd: params.parent_turn.cwd.to_path_buf(), - approval_policy: AskForApproval::Never, - approvals_reviewer: None, - sandbox_policy: SandboxPolicy::new_read_only_policy(), - model: params.model.clone(), - effort: params.reasoning_effort, - summary: Some(params.reasoning_summary), - service_tier: None, - final_output_json_schema: Some(params.schema.clone()), - collaboration_mode: None, - personality: params.personality, - }) - .await?; - - Ok::<(GuardianTranscriptCursor, TokenUsage), anyhow::Error>(( - prompt_items.transcript_cursor, - token_usage_at_review_start, - )) + .await }), ) .await; - let submit_result = match submit_result { - Ok(submit_result) => submit_result, + let prompt_items = match prompt_items { + Ok(prompt_items) => prompt_items, Err(outcome) => return (outcome, false, analytics_result), }; - let (transcript_cursor, token_usage_at_review_start) = match submit_result { - Ok(submit_result) => submit_result, + let prompt_items = match prompt_items { + Ok(prompt_items) => prompt_items, Err(err) => { return ( - GuardianReviewSessionOutcome::PromptBuildFailed(err), + GuardianReviewSessionOutcome::PromptBuildFailed(err.into()), false, analytics_result, ); } }; + let reviewed_action_truncated = prompt_items.reviewed_action_truncated; + let transcript_cursor = prompt_items.transcript_cursor; + let token_usage_at_review_start = review_session + .codex + .session + .total_token_usage() + .await + .unwrap_or_default(); + + let submit_result = run_before_review_deadline( + deadline, + params.external_cancel.as_ref(), + Box::pin(review_session.codex.submit(Op::UserTurn { + items: prompt_items.items, + cwd: params.parent_turn.cwd.to_path_buf(), + approval_policy: AskForApproval::Never, + approvals_reviewer: None, + sandbox_policy: SandboxPolicy::new_read_only_policy(), + model: params.model.clone(), + effort: params.reasoning_effort, + summary: Some(params.reasoning_summary), + service_tier: None, + final_output_json_schema: Some(params.schema.clone()), + collaboration_mode: None, + personality: params.personality, + })), + ) + .await; + match submit_result { + Ok(Ok(_)) => {} + Ok(Err(err)) => { + return ( + GuardianReviewSessionOutcome::SessionFailed(err.into()), + false, + analytics_result, + ); + } + Err(outcome) => return (outcome, false, analytics_result), + } + analytics_result.reviewed_action_truncated = reviewed_action_truncated; let outcome = wait_for_guardian_review(review_session, deadline, params.external_cancel.as_ref()).await; diff --git a/codex-rs/core/src/guardian/tests.rs b/codex-rs/core/src/guardian/tests.rs index 67bccb4c5b..17e70aa579 100644 --- a/codex-rs/core/src/guardian/tests.rs +++ b/codex-rs/core/src/guardian/tests.rs @@ -574,11 +574,12 @@ fn collect_guardian_transcript_entries_includes_recent_tool_calls_and_output() { fn guardian_truncate_text_keeps_prefix_suffix_and_xml_marker() { let content = "prefix ".repeat(200) + &" suffix".repeat(200); - let truncated = guardian_truncate_text(&content, /*token_cap*/ 20); + let (truncated, was_truncated) = guardian_truncate_text(&content, /*token_cap*/ 20); assert!(truncated.starts_with("prefix")); assert!(truncated.contains(" serde_json:: let rendered = format_guardian_action_pretty(&action)?; - assert!(rendered.contains("\"tool\": \"apply_patch\"")); - assert!(rendered.contains(" anyhow: ) .await; - let (GuardianReviewOutcome::Completed(first_assessment), _) = first_outcome else { + let (GuardianReviewOutcome::Completed(first_assessment), first_metadata) = first_outcome else { panic!("expected first guardian assessment"); }; - let (GuardianReviewOutcome::Completed(second_assessment), _) = second_outcome else { + let (GuardianReviewOutcome::Completed(second_assessment), second_metadata) = second_outcome + else { panic!("expected second guardian assessment"); }; - let (GuardianReviewOutcome::Completed(third_assessment), _) = third_outcome else { + let (GuardianReviewOutcome::Completed(third_assessment), third_metadata) = third_outcome else { panic!("expected third guardian assessment"); }; assert_eq!(first_assessment.outcome, GuardianAssessmentOutcome::Allow); assert_eq!(second_assessment.outcome, GuardianAssessmentOutcome::Allow); assert_eq!(third_assessment.outcome, GuardianAssessmentOutcome::Allow); + assert!(matches!( + first_metadata.guardian_session_kind, + Some(codex_analytics::GuardianReviewSessionKind::TrunkNew) + )); + assert!(matches!( + second_metadata.guardian_session_kind, + Some(codex_analytics::GuardianReviewSessionKind::TrunkReused) + )); + assert!(matches!( + third_metadata.guardian_session_kind, + Some(codex_analytics::GuardianReviewSessionKind::TrunkReused) + )); + ThreadId::from_string( + first_metadata + .guardian_thread_id + .as_deref() + .expect("first guardian thread id"), + ) + .expect("first guardian thread id should be a valid UUID"); + ThreadId::from_string( + second_metadata + .guardian_thread_id + .as_deref() + .expect("second guardian thread id"), + ) + .expect("second guardian thread id should be a valid UUID"); + ThreadId::from_string( + third_metadata + .guardian_thread_id + .as_deref() + .expect("third guardian thread id"), + ) + .expect("third guardian thread id should be a valid UUID"); + assert_eq!(first_metadata.had_prior_review_context, Some(false)); + assert_eq!(second_metadata.had_prior_review_context, Some(true)); + assert_eq!(third_metadata.had_prior_review_context, Some(true)); + assert_eq!( + first_metadata.guardian_thread_id, + second_metadata.guardian_thread_id + ); + assert_eq!( + second_metadata.guardian_thread_id, + third_metadata.guardian_thread_id + ); let requests = request_log.requests(); assert_eq!(requests.len(), 3);