fix(core) Preserve base_instructions in SessionMeta (#9427)

## Summary
This PR consolidates base_instructions onto SessionMeta /
SessionConfiguration, so we ensure `base_instructions` is set once per
session and should be (mostly) immutable, unless:
- overridden by config on resume / fork
- sub-agent tasks, like review or collab


In a future PR, we should convert all references to `base_instructions`
to consistently used the typed struct, so it's less likely that we put
other strings there. See #9423. However, this PR is already quite
complex, so I'm deferring that to a follow-up.

## Testing
- [x] Added a resume test to assert that instructions are preserved. In
particular, `resume_switches_models_preserves_base_instructions` fails
against main.

Existing test coverage thats assert base instructions are preserved
across multiple requests in a session:
- Manual compact keeps baseline instructions:
core/tests/suite/compact.rs:199
- Auto-compact keeps baseline instructions:
core/tests/suite/compact.rs:1142
- Prompt caching reuses the same instructions across two requests:
core/tests/suite/prompt_caching.rs:150 and
core/tests/suite/prompt_caching.rs:157
- Prompt caching with explicit expected string across two requests:
core/tests/suite/prompt_caching.rs:213 and
core/tests/suite/prompt_caching.rs:222
- Resume with model switch keeps original instructions:
core/tests/suite/resume.rs:136
- Compact/resume/fork uses request 0 instructions for later expected
payloads: core/tests/suite/compact_resume_fork.rs:215

codex-rs/core/src/rollout/recorder.rs

@@ -9,6 +9,7 @@ use std::path::PathBuf;
 use std::time::SystemTime;
 
 use codex_protocol::ThreadId;
+use codex_protocol::models::BaseInstructions;
 use serde_json::Value;
 use time::OffsetDateTime;
 use time::format_description::FormatItem;
@@ -59,6 +60,7 @@ pub enum RolloutRecorderParams {
         conversation_id: ThreadId,
         forked_from_id: Option<ThreadId>,
         source: SessionSource,
+        base_instructions: BaseInstructions,
     },
     Resume {
         path: PathBuf,
@@ -81,11 +83,13 @@ impl RolloutRecorderParams {
         conversation_id: ThreadId,
         forked_from_id: Option<ThreadId>,
         source: SessionSource,
+        base_instructions: BaseInstructions,
     ) -> Self {
         Self::Create {
             conversation_id,
             forked_from_id,
             source,
+            base_instructions,
         }
     }
 
@@ -160,6 +164,7 @@ impl RolloutRecorder {
                 conversation_id,
                 forked_from_id,
                 source,
+                base_instructions,
             } => {
                 let LogFileInfo {
                     file,
@@ -188,6 +193,7 @@ impl RolloutRecorder {
                         cli_version: env!("CARGO_PKG_VERSION").to_string(),
                         source,
                         model_provider: Some(config.model_provider_id.clone()),
+                        base_instructions: Some(base_instructions),
                     }),
                 )
             }

codex-rs/core/src/rollout/tests.rs

@@ -6,6 +6,7 @@ use std::fs::{self};
 use std::io::Write;
 use std::path::Path;
 
+use pretty_assertions::assert_eq;
 use tempfile::TempDir;
 use time::Duration;
 use time::OffsetDateTime;
@@ -89,6 +90,7 @@ fn write_session_file_with_provider(
         "cwd": ".",
         "originator": "test_originator",
         "cli_version": "test_version",
+        "base_instructions": null,
     });
 
     if let Some(source) = source {
@@ -129,6 +131,48 @@ fn write_session_file_with_provider(
     Ok((dt, uuid))
 }
 
+fn write_session_file_with_meta_payload(
+    root: &Path,
+    ts_str: &str,
+    uuid: Uuid,
+    payload: serde_json::Value,
+) -> std::io::Result<()> {
+    let format: &[FormatItem] =
+        format_description!("[year]-[month]-[day]T[hour]-[minute]-[second]");
+    let dt = PrimitiveDateTime::parse(ts_str, format)
+        .unwrap()
+        .assume_utc();
+    let dir = root
+        .join("sessions")
+        .join(format!("{:04}", dt.year()))
+        .join(format!("{:02}", u8::from(dt.month())))
+        .join(format!("{:02}", dt.day()));
+    fs::create_dir_all(&dir)?;
+
+    let filename = format!("rollout-{ts_str}-{uuid}.jsonl");
+    let file_path = dir.join(filename);
+    let mut file = File::create(file_path)?;
+
+    let meta = serde_json::json!({
+        "timestamp": ts_str,
+        "type": "session_meta",
+        "payload": payload,
+    });
+    writeln!(file, "{meta}")?;
+
+    let user_event = serde_json::json!({
+        "timestamp": ts_str,
+        "type": "event_msg",
+        "payload": {"type": "user_message", "message": "Hello from user", "kind": "plain"}
+    });
+    writeln!(file, "{user_event}")?;
+
+    let times = FileTimes::new().set_modified(dt.into());
+    file.set_times(times)?;
+
+    Ok(())
+}
+
 #[tokio::test]
 async fn test_list_conversations_latest_first() {
     let temp = TempDir::new().unwrap();
@@ -206,6 +250,7 @@ async fn test_list_conversations_latest_first() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let head_2 = vec![serde_json::json!({
         "id": u2,
@@ -215,6 +260,7 @@ async fn test_list_conversations_latest_first() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let head_1 = vec![serde_json::json!({
         "id": u1,
@@ -224,6 +270,7 @@ async fn test_list_conversations_latest_first() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
 
     let updated_times: Vec<Option<String>> =
@@ -344,6 +391,7 @@ async fn test_pagination_cursor() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let head_4 = vec![serde_json::json!({
         "id": u4,
@@ -353,6 +401,7 @@ async fn test_pagination_cursor() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let updated_page1: Vec<Option<String>> =
         page1.items.iter().map(|i| i.updated_at.clone()).collect();
@@ -410,6 +459,7 @@ async fn test_pagination_cursor() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let head_2 = vec![serde_json::json!({
         "id": u2,
@@ -419,6 +469,7 @@ async fn test_pagination_cursor() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let updated_page2: Vec<Option<String>> =
         page2.items.iter().map(|i| i.updated_at.clone()).collect();
@@ -470,6 +521,7 @@ async fn test_pagination_cursor() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let updated_page3: Vec<Option<String>> =
         page3.items.iter().map(|i| i.updated_at.clone()).collect();
@@ -527,6 +579,7 @@ async fn test_get_thread_contents() {
         "cli_version": "test_version",
         "source": "vscode",
         "model_provider": "test-provider",
+        "base_instructions": null,
     })];
     let expected_page = ThreadsPage {
         items: vec![ThreadItem {
@@ -551,6 +604,7 @@ async fn test_get_thread_contents() {
             "cwd": ".",
             "originator": "test_originator",
             "cli_version": "test_version",
+            "base_instructions": null,
             "source": "vscode",
             "model_provider": "test-provider",
         }
@@ -566,6 +620,93 @@ async fn test_get_thread_contents() {
     assert_eq!(content, expected_content);
 }
 
+#[tokio::test]
+async fn test_base_instructions_missing_in_meta_defaults_to_null() {
+    let temp = TempDir::new().unwrap();
+    let home = temp.path();
+
+    let ts = "2025-04-02T10-30-00";
+    let uuid = Uuid::from_u128(101);
+    let payload = serde_json::json!({
+        "id": uuid,
+        "timestamp": ts,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version",
+        "source": "vscode",
+        "model_provider": "test-provider",
+    });
+    write_session_file_with_meta_payload(home, ts, uuid, payload).unwrap();
+
+    let provider_filter = provider_vec(&[TEST_PROVIDER]);
+    let page = get_threads(
+        home,
+        1,
+        None,
+        ThreadSortKey::CreatedAt,
+        INTERACTIVE_SESSION_SOURCES,
+        Some(provider_filter.as_slice()),
+        TEST_PROVIDER,
+    )
+    .await
+    .unwrap();
+
+    let head = page
+        .items
+        .first()
+        .and_then(|item| item.head.first())
+        .expect("session meta head");
+    assert_eq!(
+        head.get("base_instructions"),
+        Some(&serde_json::Value::Null)
+    );
+}
+
+#[tokio::test]
+async fn test_base_instructions_present_in_meta_is_preserved() {
+    let temp = TempDir::new().unwrap();
+    let home = temp.path();
+
+    let ts = "2025-04-03T10-30-00";
+    let uuid = Uuid::from_u128(102);
+    let base_text = "Custom base instructions";
+    let payload = serde_json::json!({
+        "id": uuid,
+        "timestamp": ts,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version",
+        "source": "vscode",
+        "model_provider": "test-provider",
+        "base_instructions": {"text": base_text},
+    });
+    write_session_file_with_meta_payload(home, ts, uuid, payload).unwrap();
+
+    let provider_filter = provider_vec(&[TEST_PROVIDER]);
+    let page = get_threads(
+        home,
+        1,
+        None,
+        ThreadSortKey::CreatedAt,
+        INTERACTIVE_SESSION_SOURCES,
+        Some(provider_filter.as_slice()),
+        TEST_PROVIDER,
+    )
+    .await
+    .unwrap();
+
+    let head = page
+        .items
+        .first()
+        .and_then(|item| item.head.first())
+        .expect("session meta head");
+    let base = head
+        .get("base_instructions")
+        .and_then(|value| value.get("text"))
+        .and_then(serde_json::Value::as_str);
+    assert_eq!(base, Some(base_text));
+}
+
 #[tokio::test]
 async fn test_created_at_sort_uses_file_mtime_for_updated_at() -> Result<()> {
     let temp = TempDir::new().unwrap();
@@ -637,6 +778,7 @@ async fn test_updated_at_uses_file_mtime() -> Result<()> {
                 cli_version: "test_version".into(),
                 source: SessionSource::VSCode,
                 model_provider: Some("test-provider".into()),
+                base_instructions: None,
             },
             git: None,
         }),
@@ -744,6 +886,7 @@ async fn test_stable_ordering_same_second_pagination() {
             "cli_version": "test_version",
             "source": "vscode",
             "model_provider": "test-provider",
+            "base_instructions": null,
         })]
     };
     let updated_page1: Vec<Option<String>> =