[codex] Load user instructions through an injected provider (#27101)

## Why

We want to remove implicit use of `$CODEX_HOME` from `codex-core` and
make embedders responsible for supplying user-level instructions. This
also ensures user instructions load when no primary environment is
selected.

## What changed

Stacked on #27415, which makes `codex exec` surface thread-scoped
runtime warnings.

- Added `UserInstructionsProvider` to `codex-extension-api`, with
absolute source attribution and recoverable loading warnings.
- Added `codex-home` with the filesystem-backed provider for
`AGENTS.override.md` and `AGENTS.md`, preserving precedence, fallback,
trimming, lossy UTF-8 handling, and the existing uncapped global
instruction size.
- Removed global instruction loading from `Config` and require
`ThreadManager` callers to inject a provider.
- Load provider instructions once for each fresh root runtime, including
runtimes without a primary environment. Running sessions retain their
snapshot, while child agents inherit the parent snapshot without
invoking the provider.
- Keep provider instructions separate while loading project `AGENTS.md`,
then assemble the model-visible instructions with the existing ordering,
source attribution, warning, and turn-context behavior.
- Wired the Codex home provider through the CLI, app server, MCP server,
core facade, and thread-manager sample.

## Validation

- `just test -p codex-home -p codex-extension-api`
- `just test -p codex-core agents_md`
- `just test -p codex-core guardian`
- `just test -p codex-app-server
thread_start_without_selected_environment_includes_only_global_instruction_source`
- `just test -p codex-exec warning`
- `just bazel-lock-check`

codex-rs/app-server/Cargo.toml

@@ -36,6 +36,7 @@ codex-cloud-config = { workspace = true }
 codex-config = { workspace = true }
 codex-core = { workspace = true }
 codex-core-plugins = { workspace = true }
+codex-home = { workspace = true }
 codex-exec-server = { workspace = true }
 codex-extension-api = { workspace = true }
 codex-external-agent-migration = { workspace = true }

codex-rs/app-server/src/mcp_refresh.rs

@@ -113,6 +113,7 @@ mod tests {
     use codex_core::thread_store_from_config;
     use codex_exec_server::EnvironmentManager;
     use codex_extension_api::NoopExtensionEventSink;
+    use codex_home::CodexHomeUserInstructionsProvider;
     use codex_login::AuthManager;
     use codex_login::CodexAuth;
     use codex_protocol::protocol::SessionSource;
@@ -205,6 +206,9 @@ mod tests {
                         thread_store: Arc::clone(&thread_store),
                     },
                 ),
+                Arc::new(CodexHomeUserInstructionsProvider::new(
+                    good_config.codex_home.clone(),
+                )),
                 /*analytics_events_client*/ None,
                 Arc::clone(&thread_store),
                 Some(state_db.clone()),

codex-rs/app-server/src/message_processor.rs

@@ -72,6 +72,7 @@ use codex_core::config::Config;
 use codex_exec_server::EnvironmentManager;
 use codex_feedback::CodexFeedback;
 use codex_goal_extension::GoalService;
+use codex_home::CodexHomeUserInstructionsProvider;
 use codex_login::AuthManager;
 use codex_login::auth::ExternalAuth;
 use codex_login::auth::ExternalAuthRefreshContext;
@@ -339,6 +340,9 @@ impl MessageProcessor {
                         thread_store: Arc::clone(&thread_store),
                     },
                 ),
+                Arc::new(CodexHomeUserInstructionsProvider::new(
+                    config.codex_home.clone(),
+                )),
                 Some(analytics_events_client.clone()),
                 Arc::clone(&thread_store),
                 state_db.clone(),

codex-rs/app-server/tests/suite/v2/thread_start.rs

@@ -1,3 +1,4 @@
+use anyhow::Context;
 use anyhow::Result;
 use app_test_support::ChatGptAuthFixture;
 use app_test_support::PathBufExt;
@@ -21,6 +22,8 @@ use codex_app_server_protocol::ThreadStartedNotification;
 use codex_app_server_protocol::ThreadStatus;
 use codex_app_server_protocol::ThreadStatusChangedNotification;
 use codex_app_server_protocol::TurnEnvironmentParams;
+use codex_app_server_protocol::TurnStartParams;
+use codex_app_server_protocol::UserInput as V2UserInput;
 use codex_config::loader::project_trust_key;
 use codex_config::types::AuthCredentialsStoreMode;
 use codex_core::config::set_project_trust_level;
@@ -400,11 +403,13 @@ async fn thread_start_response_excludes_empty_project_instruction_source() -> Re
 }
 
 #[tokio::test]
-async fn thread_start_without_selected_environment_excludes_instruction_sources() -> Result<()> {
+async fn thread_start_without_selected_environment_includes_only_global_instruction_source()
+-> Result<()> {
     let server = create_mock_responses_server_repeating_assistant("Done").await;
     let codex_home = TempDir::new()?;
     create_config_toml_without_approval_policy(codex_home.path(), &server.uri())?;
-    std::fs::write(codex_home.path().join("AGENTS.md"), "global instructions")?;
+    let global_agents_path = codex_home.path().join("AGENTS.md");
+    std::fs::write(&global_agents_path, "global instructions")?;
     let workspace = TempDir::new()?;
     std::fs::write(workspace.path().join("AGENTS.md"), "project instructions")?;
 
@@ -424,11 +429,56 @@ async fn thread_start_without_selected_environment_excludes_instruction_sources(
     )
     .await??;
     let ThreadStartResponse {
+        thread,
         instruction_sources,
         ..
     } = to_response::<ThreadStartResponse>(response)?;
 
-    assert!(instruction_sources.is_empty());
+    assert_eq!(
+        instruction_sources
+            .into_iter()
+            .map(normalize_path_for_comparison)
+            .collect::<Vec<_>>(),
+        vec![normalize_path_for_comparison(std::fs::canonicalize(
+            global_agents_path,
+        )?)]
+    );
+
+    let turn_request_id = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id,
+            input: vec![V2UserInput::Text {
+                text: "inspect instructions".to_string(),
+                text_elements: Vec::new(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_request_id)),
+    )
+    .await??;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    let requests = server
+        .received_requests()
+        .await
+        .context("failed to fetch received requests")?;
+    let model_request = requests
+        .iter()
+        .find(|request| request.url.path().ends_with("/responses"))
+        .context("expected model request")?;
+    let model_request_body = model_request
+        .body_json::<Value>()
+        .context("model request body should be JSON")?
+        .to_string();
+    assert!(model_request_body.contains("global instructions"));
+    assert!(!model_request_body.contains("project instructions"));
 
     Ok(())
 }