Fix reasoning model directive response identity

apps/aether-gateway/src/ai_serving/planner/candidate_materialization.rs

@@ -267,6 +267,9 @@ where
             self.state,
             self.trace_id,
             self.persistence_policy.available,
+            self.persistence_policy
+                .skipped
+                .record_runtime_miss_diagnostic,
             candidates,
             self.sticky_session_token,
             self.requested_model,
@@ -528,6 +531,8 @@ where
         state,
         candidates,
         0,
+        Some(trace_id),
+        persistence_policy.skipped.record_runtime_miss_diagnostic,
         sticky_session_token,
         requested_model,
         request_auth_channel,
@@ -543,6 +548,8 @@ fn build_logical_candidate_items<'a>(
     state: PlannerAppState<'a>,
     candidates: Vec<EligibleLocalExecutionCandidate>,
     starting_candidate_index: u32,
+    trace_id: Option<&str>,
+    record_runtime_miss_diagnostic: bool,
     sticky_session_token: Option<&str>,
     requested_model: Option<&str>,
     request_auth_channel: Option<&str>,
@@ -563,14 +570,20 @@ fn build_logical_candidate_items<'a>(
                 }
             }
             LocalExecutionCandidateKind::PoolGroup => {
+                let cursor = PoolKeyCursor::new(
+                    state,
+                    candidate,
+                    sticky_session_token,
+                    requested_model,
+                    request_auth_channel,
+                );
+                let cursor = if let Some(trace_id) = trace_id {
+                    cursor.with_runtime_miss_diagnostic(trace_id, record_runtime_miss_diagnostic)
+                } else {
+                    cursor
+                };
                 items.push_back(LocalExecutionCandidateAttemptSourceItem::Pool {
-                    cursor: PoolKeyCursor::new(
-                        state,
-                        candidate,
-                        sticky_session_token,
-                        requested_model,
-                        request_auth_channel,
-                    ),
+                    cursor,
                     candidate_index,
                     pending_attempts: VecDeque::new(),
                 });
@@ -757,6 +770,8 @@ impl<'a> RequestedModelAttemptPageCursor<'a> {
                 self.state,
                 candidates,
                 self.next_candidate_index,
+                Some(&self.trace_id),
+                self.record_runtime_miss_diagnostic,
                 self.sticky_session_token.as_deref(),
                 Some(&self.requested_model),
                 self.request_auth_channel.as_deref(),
@@ -922,6 +937,7 @@ async fn materialize_logical_local_execution_candidate_attempts<F>(
     state: PlannerAppState<'_>,
     trace_id: &str,
     context: LocalAvailableCandidatePersistenceContext<'_>,
+    record_runtime_miss_diagnostic: bool,
     candidates: Vec<EligibleLocalExecutionCandidate>,
     sticky_session_token: Option<&str>,
     requested_model: Option<&str>,
@@ -956,7 +972,8 @@ where
                     sticky_session_token,
                     requested_model,
                     request_auth_channel,
-                );
+                )
+                .with_runtime_miss_diagnostic(trace_id, record_runtime_miss_diagnostic);
                 let attempt_count_before_pool = attempts.len();
                 while let Some(candidate) = cursor.next_key().await {
                     attempts.extend(build_unpersisted_local_execution_candidate_attempts(
@@ -1525,6 +1542,7 @@ mod tests {
                 required_capabilities: None,
                 error_context: "persist should not fail",
             },
+            false,
             vec![pool_group, sample_eligible("normal-key", None)],
             None,
             Some("gpt-5"),

apps/aether-gateway/src/ai_serving/planner/pool_scheduler.rs

@@ -19,6 +19,7 @@ use crate::ai_serving::planner::candidate_resolution::{
     candidate_auth_channel_skip_reason, read_candidate_transport_snapshot,
     EligibleLocalExecutionCandidate, LocalExecutionCandidateKind, SkippedLocalExecutionCandidate,
 };
+use crate::ai_serving::planner::runtime_miss::record_local_runtime_candidate_skip_reason;
 use crate::ai_serving::{
     candidate_common_transport_skip_reason, CandidateTransportPolicyFacts, PlannerAppState,
 };
@@ -171,6 +172,8 @@ pub(crate) struct PoolKeyCursor<'a> {
     sticky_session_token: Option<String>,
     requested_model: Option<String>,
     request_auth_channel: Option<String>,
+    runtime_miss_trace_id: Option<String>,
+    record_runtime_miss_diagnostic: bool,
     pool_key_order: StoredPoolKeyCandidateOrder,
     next_offset: u32,
     scanned_keys: u32,
@@ -183,6 +186,8 @@ pub(crate) struct PoolKeyCursor<'a> {
     queued_candidates: VecDeque<EligibleLocalExecutionCandidate>,
     skipped_candidates: Vec<SkippedLocalExecutionCandidate>,
     exhausted_logged: bool,
+    returned_key_count: u32,
+    exhaustion_skip_recorded: bool,
 }
 
 impl<'a> PoolKeyCursor<'a> {
@@ -200,6 +205,8 @@ impl<'a> PoolKeyCursor<'a> {
             sticky_session_token: sticky_session_token.map(str::to_string),
             requested_model: requested_model.map(str::to_string),
             request_auth_channel: request_auth_channel.map(str::to_string),
+            runtime_miss_trace_id: None,
+            record_runtime_miss_diagnostic: false,
             pool_key_order,
             next_offset: 0,
             scanned_keys: 0,
@@ -212,12 +219,27 @@ impl<'a> PoolKeyCursor<'a> {
             queued_candidates: VecDeque::new(),
             skipped_candidates: Vec::new(),
             exhausted_logged: false,
+            returned_key_count: 0,
+            exhaustion_skip_recorded: false,
         }
     }
 
+    pub(crate) fn with_runtime_miss_diagnostic(
+        mut self,
+        trace_id: &str,
+        record_runtime_miss_diagnostic: bool,
+    ) -> Self {
+        if record_runtime_miss_diagnostic {
+            self.runtime_miss_trace_id = Some(trace_id.to_string());
+            self.record_runtime_miss_diagnostic = true;
+        }
+        self
+    }
+
     pub(crate) async fn next_key(&mut self) -> Option<EligibleLocalExecutionCandidate> {
         loop {
             if let Some(candidate) = self.next_queued_candidate().await {
+                self.returned_key_count = self.returned_key_count.saturating_add(1);
                 return Some(candidate);
             }
 
@@ -254,6 +276,37 @@ impl<'a> PoolKeyCursor<'a> {
             skip_reason_counts = ?self.skip_reason_counts,
             "gateway pool scheduler exhausted pool group without a schedulable key"
         );
+        self.record_runtime_miss_pool_exhaustion_skip_reason();
+    }
+
+    fn record_runtime_miss_pool_exhaustion_skip_reason(&mut self) {
+        if self.exhaustion_skip_recorded
+            || !self.record_runtime_miss_diagnostic
+            || self.returned_key_count > 0
+        {
+            return;
+        }
+        let Some(trace_id) = self.runtime_miss_trace_id.as_deref() else {
+            return;
+        };
+        self.exhaustion_skip_recorded = true;
+        record_local_runtime_candidate_skip_reason(
+            self.state.app(),
+            trace_id,
+            self.runtime_miss_pool_exhaustion_skip_reason(),
+        );
+    }
+
+    fn runtime_miss_pool_exhaustion_skip_reason(&self) -> &'static str {
+        let mut selected_reason = "pool_group_exhausted";
+        let mut selected_count = 0;
+        for (reason, count) in &self.skip_reason_counts {
+            if *count > selected_count {
+                selected_reason = *reason;
+                selected_count = *count;
+            }
+        }
+        selected_reason
     }
 
     async fn next_page_candidates(&mut self) -> Option<Vec<EligibleLocalExecutionCandidate>> {
@@ -963,6 +1016,7 @@ mod tests {
     use crate::ai_serving::planner::candidate_resolution::{
         EligibleLocalExecutionCandidate, LocalExecutionCandidateKind,
     };
+    use crate::ai_serving::planner::runtime_miss::apply_local_runtime_candidate_terminal_reason;
     use crate::ai_serving::PlannerAppState;
     use crate::data::GatewayDataState;
     use crate::handlers::shared::provider_pool::{
@@ -970,7 +1024,7 @@ mod tests {
         try_claim_admin_provider_pool_key, AdminProviderPoolRuntimeState,
     };
     use crate::orchestration::LocalExecutionCandidateMetadata;
-    use crate::AppState;
+    use crate::{AppState, LocalExecutionRuntimeMissDiagnostic};
     use aether_ai_serving::{normalize_enabled_ai_pool_presets, AiPoolSchedulingPreset};
     use aether_data::repository::candidate_selection::InMemoryMinimalCandidateSelectionReadRepository;
     use aether_data::repository::provider_catalog::InMemoryProviderCatalogReadRepository;
@@ -1739,6 +1793,43 @@ mod tests {
         ));
     }
 
+    #[test]
+    fn pool_key_cursor_records_runtime_miss_when_exhausted_without_returning_key() {
+        let app = AppState::new().expect("state should build");
+        let trace_id = "trace-pool-exhausted-runtime-miss";
+        app.set_local_execution_runtime_miss_diagnostic(
+            trace_id,
+            LocalExecutionRuntimeMissDiagnostic {
+                reason: "candidate_evaluation_incomplete".to_string(),
+                requested_model: Some("gpt-5".to_string()),
+                candidate_count: Some(1),
+                ..LocalExecutionRuntimeMissDiagnostic::default()
+            },
+        );
+        let group = sample_eligible_candidate(
+            "provider-pool",
+            "endpoint-1",
+            "pool-group",
+            10,
+            Some(json!({ "pool_advanced": { "lru_enabled": true } })),
+        );
+        let mut cursor = PoolKeyCursor::new(PlannerAppState::new(&app), group, None, None, None)
+            .with_runtime_miss_diagnostic(trace_id, true);
+        cursor.record_skip_reason("pool_key_lease_busy");
+        cursor.record_skip_reason("pool_key_lease_busy");
+        cursor.record_skip_reason("transport_snapshot_missing");
+
+        cursor.log_exhausted();
+        apply_local_runtime_candidate_terminal_reason(&app, trace_id, "no_local_sync_plans");
+
+        let diagnostic = app
+            .take_local_execution_runtime_miss_diagnostic(trace_id)
+            .expect("runtime miss diagnostic should exist");
+        assert_eq!(diagnostic.reason, "all_candidates_skipped");
+        assert_eq!(diagnostic.skipped_candidate_count, Some(1));
+        assert_eq!(diagnostic.skip_reasons.get("pool_key_lease_busy"), Some(&1));
+    }
+
     #[tokio::test]
     async fn pool_key_cursor_revalidates_queued_candidates_before_return() {
         let app = AppState::new().expect("state should build");

apps/aether-gateway/src/handlers/proxy/mod.rs

@@ -1656,6 +1656,7 @@ fn local_execution_runtime_miss_skip_reasons_summary(
 fn local_execution_runtime_miss_skip_reason_label(reason: &str) -> &str {
     match reason {
         "api_key_concurrency_limit_reached" => "API Key 并发已达上限",
+        "auth_channel_mismatch" => "认证通道不匹配",
         "auth_snapshot_missing" => "API Key 本地执行配置缺失",
         "endpoint_api_format_changed" => "端点 API 格式已变更",
         "endpoint_inactive" => "端点未启用",
@@ -1664,6 +1665,10 @@ fn local_execution_runtime_miss_skip_reason_label(reason: &str) -> &str {
         "key_inactive" => "API Key 未启用",
         "key_model_disabled" => "API Key 未允许该模型",
         "mapped_model_missing" => "模型映射缺失",
+        "pool_cooldown" => "池内账号处于冷却中",
+        "pool_cost_limit_reached" => "池内账号成本额度已用尽",
+        "pool_group_exhausted" => "池化提供商没有可调度账号",
+        "pool_key_lease_busy" => "池内账号正被其他请求占用",
         "provider_inactive" => "提供商未启用",
         "provider_request_body_missing" => "无法构建上游请求体",
         "provider_request_body_build_failed" => "上游请求体转换失败",

crates/aether-ai-formats/src/formats/shared/model_directives.rs

@@ -98,6 +98,21 @@ pub fn model_directive_base_model(model: &str) -> Option<String> {
     parse_model_directive(model).map(|directive| directive.base_model)
 }
 
+pub(crate) fn model_directive_display_model(model: &str) -> Option<String> {
+    let model = model.trim();
+    parse_model_directive(model)?;
+    Some(model.to_string())
+}
+
+pub(crate) fn model_directive_display_model_from_report_context(
+    report_context: &Value,
+) -> Option<String> {
+    report_context
+        .get("model")
+        .and_then(Value::as_str)
+        .and_then(model_directive_display_model)
+}
+
 pub fn normalize_model_directive_model(model: &str) -> String {
     parse_model_directive(model)
         .map(|directive| directive.base_model)

crates/aether-ai-formats/src/formats/shared/stream_core/common.rs

@@ -1,5 +1,7 @@
 use serde_json::{json, Map, Value};
 
+use crate::formats::shared::model_directives::model_directive_display_model_from_report_context;
+
 pub use aether_ai_formats::protocol::stream::{
     CanonicalContentPart, CanonicalStreamEvent, CanonicalStreamFrame, CanonicalUsage,
 };
@@ -27,6 +29,9 @@ pub fn resolve_identity(
         .filter(|value| !value.is_empty())
         .unwrap_or(default_id)
         .to_string();
+    if let Some(display_model) = model_directive_display_model_from_report_context(report_context) {
+        return (id, display_model);
+    }
     let model = model
         .filter(|value| !value.is_empty())
         .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))

crates/aether-ai-formats/src/formats/shared/stream_rewrite.rs

@@ -1,6 +1,7 @@
 use serde_json::Value;
 
 use crate::formats::openai::image::stream::OpenAiImageStreamState;
+use crate::formats::shared::model_directives::model_directive_display_model_from_report_context;
 use crate::formats::shared::stream_core::StreamingStandardFormatMatrix;
 use crate::formats::shared::AiSurfaceFinalizeError;
 use crate::provider_compat::kiro_stream::KiroToClaudeCliStreamState;
@@ -12,6 +13,7 @@ use crate::provider_compat::surfaces::{
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum FinalizeStreamRewriteMode {
     EnvelopeUnwrap,
+    ModelDirectiveDisplay,
     OpenAiImage,
     Standard,
     KiroToClaudeCli,
@@ -73,6 +75,17 @@ pub fn resolve_finalize_stream_rewrite_mode(
             .then_some(FinalizeStreamRewriteMode::KiroToClaudeCli);
     }
 
+    if model_directive_display_model_from_report_context(report_context).is_some()
+        && provider_api_format == client_api_format
+        && is_standard_provider_api_format(provider_api_format.as_str())
+        && !provider_adaptation_should_unwrap_stream_envelope(
+            envelope_name.as_str(),
+            provider_api_format.as_str(),
+        )
+    {
+        return Some(FinalizeStreamRewriteMode::ModelDirectiveDisplay);
+    }
+
     (provider_api_format == client_api_format
         && provider_adaptation_should_unwrap_stream_envelope(
             envelope_name.as_str(),
@@ -83,6 +96,7 @@ pub fn resolve_finalize_stream_rewrite_mode(
 
 enum AiSurfaceStreamRewriteState {
     EnvelopeUnwrap,
+    ModelDirectiveDisplay,
     OpenAiImage(Box<OpenAiImageStreamState>),
     Standard(Box<StreamingStandardFormatMatrix>),
     KiroToClaudeCli(Box<KiroToClaudeCliStreamState>),
@@ -104,6 +118,9 @@ pub fn maybe_build_ai_surface_stream_rewriter<'a>(
     let report_context = report_context?;
     let state = match resolve_finalize_stream_rewrite_mode(report_context)? {
         FinalizeStreamRewriteMode::EnvelopeUnwrap => AiSurfaceStreamRewriteState::EnvelopeUnwrap,
+        FinalizeStreamRewriteMode::ModelDirectiveDisplay => {
+            AiSurfaceStreamRewriteState::ModelDirectiveDisplay
+        }
         FinalizeStreamRewriteMode::OpenAiImage => {
             AiSurfaceStreamRewriteState::OpenAiImage(Box::<OpenAiImageStreamState>::default())
         }
@@ -142,6 +159,7 @@ impl AiSurfaceStreamRewriter<'_> {
                 transform_standard_bytes(standard, self.report_context, claude_bytes)
             }
             AiSurfaceStreamRewriteState::EnvelopeUnwrap
+            | AiSurfaceStreamRewriteState::ModelDirectiveDisplay
             | AiSurfaceStreamRewriteState::Standard(_) => {
                 self.buffered.extend_from_slice(chunk);
                 let mut output = Vec::new();
@@ -170,6 +188,7 @@ impl AiSurfaceStreamRewriter<'_> {
                 Ok(output)
             }
             AiSurfaceStreamRewriteState::EnvelopeUnwrap
+            | AiSurfaceStreamRewriteState::ModelDirectiveDisplay
             | AiSurfaceStreamRewriteState::Standard(_) => {
                 if self.buffered.is_empty() {
                     if let AiSurfaceStreamRewriteState::Standard(state) = &mut self.state {
@@ -190,8 +209,12 @@ impl AiSurfaceStreamRewriter<'_> {
     fn transform_line(&mut self, line: Vec<u8>) -> Result<Vec<u8>, AiSurfaceFinalizeError> {
         match &mut self.state {
             AiSurfaceStreamRewriteState::EnvelopeUnwrap => {
-                transform_provider_private_stream_line(self.report_context, line)
-                    .map_err(AiSurfaceFinalizeError::from)
+                let output = transform_provider_private_stream_line(self.report_context, line)
+                    .map_err(AiSurfaceFinalizeError::from)?;
+                rewrite_model_directive_stream_line(self.report_context, output)
+            }
+            AiSurfaceStreamRewriteState::ModelDirectiveDisplay => {
+                rewrite_model_directive_stream_line(self.report_context, line)
             }
             AiSurfaceStreamRewriteState::Standard(state) => {
                 transform_standard_line(state, self.report_context, line)
@@ -203,6 +226,63 @@ impl AiSurfaceStreamRewriter<'_> {
     }
 }
 
+fn rewrite_model_directive_stream_line(
+    report_context: &Value,
+    line: Vec<u8>,
+) -> Result<Vec<u8>, AiSurfaceFinalizeError> {
+    let Some(display_model) = model_directive_display_model_from_report_context(report_context)
+    else {
+        return Ok(line);
+    };
+    let text = match std::str::from_utf8(&line) {
+        Ok(text) => text,
+        Err(_) => return Ok(line),
+    };
+    let trimmed_line_end = text.trim_end_matches(['\r', '\n']);
+    let trailing = &text[trimmed_line_end.len()..];
+    let Some((prefix, payload)) = trimmed_line_end.split_once(':') else {
+        return Ok(line);
+    };
+    if prefix.trim() != "data" {
+        return Ok(line);
+    }
+    let payload = payload.trim_start();
+    if payload.is_empty() || payload == "[DONE]" {
+        return Ok(line);
+    }
+    let mut value = match serde_json::from_str::<Value>(payload) {
+        Ok(value) => value,
+        Err(_) => return Ok(line),
+    };
+    if !rewrite_stream_payload_model(&mut value, &display_model) {
+        return Ok(line);
+    }
+    let mut output = Vec::new();
+    output.extend_from_slice(b"data: ");
+    output.extend(serde_json::to_vec(&value)?);
+    output.extend_from_slice(trailing.as_bytes());
+    Ok(output)
+}
+
+fn rewrite_stream_payload_model(value: &mut Value, display_model: &str) -> bool {
+    let Some(object) = value.as_object_mut() else {
+        return false;
+    };
+    let mut changed = false;
+    for key in ["model", "modelVersion"] {
+        if object.get(key).and_then(Value::as_str).is_some() {
+            object.insert(key.to_string(), Value::String(display_model.to_string()));
+            changed = true;
+        }
+    }
+    for key in ["response", "message"] {
+        if let Some(nested) = object.get_mut(key) {
+            changed |= rewrite_stream_payload_model(nested, display_model);
+        }
+    }
+    changed
+}
+
 fn transform_standard_bytes(
     standard: &mut StreamingStandardFormatMatrix,
     report_context: &Value,
@@ -288,7 +368,10 @@ fn is_standard_cli_client_api_format(api_format: &str) -> bool {
 mod tests {
     use serde_json::json;
 
-    use super::{resolve_finalize_stream_rewrite_mode, FinalizeStreamRewriteMode};
+    use super::{
+        maybe_build_ai_surface_stream_rewriter, resolve_finalize_stream_rewrite_mode,
+        FinalizeStreamRewriteMode,
+    };
 
     #[test]
     fn resolves_standard_mode_for_cross_format_standard_streams() {
@@ -341,6 +424,84 @@ mod tests {
         assert_eq!(resolve_finalize_stream_rewrite_mode(&report_context), None);
     }
 
+    #[test]
+    fn resolves_model_directive_display_mode_for_same_format_standard_streams() {
+        let report_context = json!({
+            "provider_api_format": "openai:responses",
+            "client_api_format": "openai:responses",
+            "model": "gpt-5.5-xhigh",
+            "mapped_model": "gpt-5.5",
+            "needs_conversion": false,
+        });
+        assert_eq!(
+            resolve_finalize_stream_rewrite_mode(&report_context),
+            Some(FinalizeStreamRewriteMode::ModelDirectiveDisplay)
+        );
+    }
+
+    #[test]
+    fn model_directive_display_mode_does_not_displace_kiro_stream_bridge() {
+        let report_context = json!({
+            "provider_api_format": "claude:messages",
+            "client_api_format": "claude:messages",
+            "envelope_name": "kiro:generateAssistantResponse",
+            "model": "claude-sonnet-4.5-high",
+            "mapped_model": "claude-sonnet-4.5",
+            "needs_conversion": false,
+        });
+        assert_eq!(
+            resolve_finalize_stream_rewrite_mode(&report_context),
+            Some(FinalizeStreamRewriteMode::KiroToClaudeCli)
+        );
+    }
+
+    #[test]
+    fn envelope_unwrap_rewriter_restores_model_directive_display_model() {
+        let report_context = json!({
+            "provider_api_format": "gemini:generate_content",
+            "client_api_format": "gemini:generate_content",
+            "envelope_name": "gemini_cli:v1internal",
+            "model": "gemini-2.5-pro-high",
+            "mapped_model": "gemini-2.5-pro",
+            "needs_conversion": false,
+        });
+        let mut rewriter = maybe_build_ai_surface_stream_rewriter(Some(&report_context))
+            .expect("rewriter should exist");
+        let output = rewriter
+            .push_chunk(
+                b"data: {\"response\":{\"modelVersion\":\"gemini-2.5-pro\",\"candidates\":[]}}\n\n",
+            )
+            .expect("rewrite should succeed");
+        let output = String::from_utf8(output).expect("output should be utf8");
+
+        assert!(output.contains("\"modelVersion\":\"gemini-2.5-pro-high\""));
+        assert!(!output.contains("\"modelVersion\":\"gemini-2.5-pro\""));
+    }
+
+    #[test]
+    fn model_directive_display_rewriter_restores_response_model() {
+        let report_context = json!({
+            "provider_api_format": "openai:responses",
+            "client_api_format": "openai:responses",
+            "model": "gpt-5.5-xhigh",
+            "mapped_model": "gpt-5.5",
+            "needs_conversion": false,
+        });
+        let mut rewriter = maybe_build_ai_surface_stream_rewriter(Some(&report_context))
+            .expect("rewriter should exist");
+        let output = rewriter
+            .push_chunk(
+                b"event: response.created\n\
+data: {\"type\":\"response.created\",\"response\":{\"id\":\"resp_123\",\"object\":\"response\",\"model\":\"gpt-5.5\",\"status\":\"in_progress\"}}\n\n",
+            )
+            .expect("rewrite should succeed");
+        let output = String::from_utf8(output).expect("output should be utf8");
+
+        assert!(output.contains("event: response.created"));
+        assert!(output.contains("\"model\":\"gpt-5.5-xhigh\""));
+        assert!(!output.contains("\"model\":\"gpt-5.5\""));
+    }
+
     #[test]
     fn resolves_openai_image_mode_for_same_format_image_streams() {
         let report_context = json!({

crates/aether-ai-formats/src/formats/shared/sync_products.rs

@@ -19,6 +19,7 @@ use serde_json::{json, Map, Value};
 
 use super::AiSurfaceFinalizeError;
 use crate::formats::gemini::generate_content::stream::GeminiProviderState;
+use crate::formats::shared::model_directives::model_directive_display_model_from_report_context;
 use crate::formats::shared::response::remove_empty_pages_from_tool_arguments;
 use crate::formats::shared::stream_core::common::{
     map_openai_finish_reason_to_gemini, parse_json_arguments_value, CanonicalContentPart,
@@ -382,7 +383,11 @@ fn maybe_build_standard_same_format_sync_body(
         return None;
     }
 
-    Some(body_json.clone())
+    Some(client_body_with_report_context_model(
+        body_json.clone(),
+        report_context,
+        &client_api_format,
+    ))
 }
 
 fn maybe_build_standard_same_format_stream_sync_body(
@@ -431,10 +436,11 @@ fn maybe_build_standard_same_format_stream_sync_body(
         return Ok(None);
     };
     let body_bytes = base64::engine::general_purpose::STANDARD.decode(body_base64)?;
-    Ok(aggregate_same_format_stream_sync_response(
-        expected_api_format,
-        &body_bytes,
-    ))
+    Ok(
+        aggregate_same_format_stream_sync_response(expected_api_format, &body_bytes).map(|body| {
+            client_body_with_report_context_model(body, report_context, &client_api_format)
+        }),
+    )
 }
 
 fn maybe_build_openai_responses_same_family_sync_body(
@@ -480,7 +486,11 @@ fn maybe_build_openai_responses_same_family_sync_body(
         return None;
     }
 
-    Some(body_json.clone())
+    Some(client_body_with_report_context_model(
+        body_json.clone(),
+        report_context,
+        &client_api_format,
+    ))
 }
 
 fn maybe_build_openai_responses_same_family_stream_sync_body(
@@ -528,7 +538,11 @@ fn maybe_build_openai_responses_same_family_stream_sync_body(
         return Ok(None);
     };
     let body_bytes = base64::engine::general_purpose::STANDARD.decode(body_base64)?;
-    Ok(aggregate_openai_responses_stream_sync_response(&body_bytes))
+    Ok(
+        aggregate_openai_responses_stream_sync_response(&body_bytes).map(|body| {
+            client_body_with_report_context_model(body, report_context, &client_api_format)
+        }),
+    )
 }
 
 fn maybe_build_openai_cross_format_provider_body_from_normalized_payload(
@@ -625,6 +639,8 @@ pub fn maybe_build_standard_cross_format_sync_product(
     } else {
         return None;
     };
+    let client_body_json =
+        client_body_with_report_context_model(client_body_json, report_context, &client_api_format);
 
     Some(StandardCrossFormatSyncProduct {
         client_body_json,
@@ -797,6 +813,30 @@ fn format_context_from_report_context(report_context: &Value) -> FormatContext {
     context
 }
 
+fn client_body_with_report_context_model(
+    mut body_json: Value,
+    report_context: &Value,
+    client_api_format: &str,
+) -> Value {
+    let Some(display_model) = model_directive_display_model_from_report_context(report_context)
+    else {
+        return body_json;
+    };
+    let Some(object) = body_json.as_object_mut() else {
+        return body_json;
+    };
+    match normalize_openai_responses_family_api_format(client_api_format).as_str() {
+        "openai:chat" | "openai:responses" | "openai:responses:compact" | "claude:messages" => {
+            object.insert("model".to_string(), Value::String(display_model));
+        }
+        "gemini:generate_content" => {
+            object.insert("modelVersion".to_string(), Value::String(display_model));
+        }
+        _ => {}
+    }
+    body_json
+}
+
 fn convert_openai_chat_canonical_chat_response(
     body_json: &Value,
     client_api_format: &str,
@@ -1195,6 +1235,10 @@ fn gemini_response_can_use_single_response_canonical(body_json: &Value) -> bool
 }
 
 fn apply_report_context_model_fallback(model: &mut String, report_context: &Value) {
+    if let Some(display_model) = model_directive_display_model_from_report_context(report_context) {
+        *model = display_model;
+        return;
+    }
     if model != "unknown" && !model.trim().is_empty() {
         return;
     }
@@ -2881,6 +2925,7 @@ mod tests {
         maybe_build_openai_chat_cross_format_sync_product_from_normalized_payload,
         maybe_build_openai_responses_cross_format_sync_product_from_normalized_payload,
         maybe_build_openai_responses_same_family_sync_body_from_normalized_payload,
+        maybe_build_standard_cross_format_sync_product,
         maybe_build_standard_cross_format_sync_product_from_normalized_payload,
         maybe_build_standard_same_format_sync_body_from_normalized_payload,
         maybe_build_standard_sync_finalize_product_from_normalized_payload,
@@ -3222,6 +3267,75 @@ mod tests {
         assert_eq!(body_json, provider_body_json);
     }
 
+    #[test]
+    fn same_format_sync_response_restores_model_directive_display_model() {
+        let report_context = json!({
+            "provider_api_format": "openai:responses",
+            "client_api_format": "openai:responses",
+            "model": "gpt-5.5-xhigh",
+            "mapped_model": "gpt-5.5",
+            "needs_conversion": false,
+        });
+        let provider_body_json = json!({
+            "id": "resp_123",
+            "object": "response",
+            "model": "gpt-5.5",
+            "status": "completed",
+            "output": []
+        });
+
+        let product = maybe_build_standard_sync_finalize_product_from_normalized_payload(
+            "openai_responses_sync_finalize",
+            200,
+            Some(&report_context),
+            Some(&provider_body_json),
+            None,
+        )
+        .expect("same-format sync body should succeed")
+        .expect("body should exist");
+        let StandardSyncFinalizeNormalizedProduct::SuccessBody(body_json) = product else {
+            panic!("same-format response should be a success body");
+        };
+
+        assert_eq!(body_json["model"], "gpt-5.5-xhigh");
+    }
+
+    #[test]
+    fn cross_format_sync_response_restores_model_directive_display_model() {
+        let report_context = json!({
+            "provider_api_format": "openai:responses",
+            "client_api_format": "claude:messages",
+            "model": "gpt-5.5-xhigh",
+            "mapped_model": "gpt-5.5",
+            "needs_conversion": true,
+        });
+        let provider_body_json = json!({
+            "id": "resp_123",
+            "object": "response",
+            "model": "gpt-5.5",
+            "status": "completed",
+            "output": [{
+                "type": "message",
+                "id": "msg_123",
+                "role": "assistant",
+                "status": "completed",
+                "content": [{ "type": "output_text", "text": "done" }]
+            }]
+        });
+
+        let product = maybe_build_standard_cross_format_sync_product(
+            "claude_cli_sync_finalize",
+            "openai:responses",
+            "claude:messages",
+            &report_context,
+            provider_body_json,
+        )
+        .expect("cross-format product should exist");
+
+        assert_eq!(product.client_body_json["model"], "gpt-5.5-xhigh");
+        assert_eq!(product.provider_body_json["model"], "gpt-5.5");
+    }
+
     #[test]
     fn rejects_standard_same_format_when_needs_conversion_is_true() {
         let report_context = json!({

crates/aether-ai-formats/src/provider_compat/kiro_stream/stream/state/lifecycle.rs

@@ -1,6 +1,7 @@
 use serde_json::Value;
 use uuid::Uuid;
 
+use crate::formats::shared::model_directives::model_directive_display_model_from_report_context;
 use crate::formats::shared::AiSurfaceFinalizeError;
 use crate::provider_compat::kiro_stream::{
     build_kiro_initial_sse_events, build_kiro_stream_error_sse_events, encode_kiro_sse_events,
@@ -63,18 +64,22 @@ impl KiroToClaudeCliStreamState {
 
 impl KiroClaudeStreamState {
     pub(super) fn new(report_context: &Value) -> Self {
-        let model = report_context
-            .get("mapped_model")
-            .and_then(Value::as_str)
-            .filter(|value| !value.is_empty())
+        let model = model_directive_display_model_from_report_context(report_context)
+            .or_else(|| {
+                report_context
+                    .get("mapped_model")
+                    .and_then(Value::as_str)
+                    .filter(|value| !value.is_empty())
+                    .map(ToOwned::to_owned)
+            })
             .or_else(|| {
                 report_context
                     .get("model")
                     .and_then(Value::as_str)
                     .filter(|value| !value.is_empty())
+                    .map(ToOwned::to_owned)
             })
-            .unwrap_or("unknown")
-            .to_string();
+            .unwrap_or_else(|| "unknown".to_string());
         let thinking_enabled = report_context
             .get("original_request_body")
             .and_then(Value::as_object)

crates/aether-ai-formats/src/provider_compat/kiro_stream/stream/tests.rs

@@ -86,6 +86,32 @@ fn kiro_stream_rewriter_converts_text_events_to_claude_sse() {
     assert!(text.contains("\"input_tokens\":2000"));
 }
 
+#[test]
+fn kiro_stream_rewriter_restores_model_directive_display_model() {
+    let report_context = json!({
+        "provider_api_format": "claude:messages",
+        "client_api_format": "claude:messages",
+        "envelope_name": "kiro:generateAssistantResponse",
+        "model": "claude-sonnet-4.5-high",
+        "mapped_model": "claude-sonnet-4.5"
+    });
+    let mut rewriter = KiroToClaudeCliStreamState::new(&report_context);
+    let first = rewriter
+        .push_chunk(
+            &report_context,
+            &encode_event_frame(
+                "event",
+                Some("assistantResponseEvent"),
+                &json!({"content": "Hello"}),
+            ),
+        )
+        .expect("rewrite should succeed");
+    let text = String::from_utf8(first).expect("utf8 should decode");
+
+    assert!(text.contains("\"model\":\"claude-sonnet-4.5-high\""));
+    assert!(!text.contains("\"model\":\"claude-sonnet-4.5\""));
+}
+
 #[test]
 fn kiro_stream_rewriter_converts_tool_use_to_claude_events() {
     let report_context = kiro_report_context(false);