vibe-kanban/crates/local-deployment/src/container.rs

use std::{
    collections::HashMap,
    io,
    path::{Path, PathBuf},
    sync::Arc,
    time::Duration,
};

use anyhow::anyhow;
use async_trait::async_trait;
use command_group::AsyncGroupChild;
use db::{
    DBService,
    models::{
        coding_agent_turn::CodingAgentTurn,
        execution_process::{
            ExecutionContext, ExecutionProcess, ExecutionProcessRunReason, ExecutionProcessStatus,
        },
        execution_process_repo_state::ExecutionProcessRepoState,
        project_repo::ProjectRepo,
        repo::Repo,
        scratch::{DraftFollowUpData, Scratch, ScratchType},
        task::{Task, TaskStatus},
        workspace::Workspace,
        workspace_repo::WorkspaceRepo,
    },
};
use deployment::{DeploymentError, RemoteClientNotConfigured};
use executors::{
    actions::{
        Executable, ExecutorAction, ExecutorActionType,
        coding_agent_follow_up::CodingAgentFollowUpRequest,
        coding_agent_initial::CodingAgentInitialRequest,
    },
    approvals::{ExecutorApprovalService, NoopExecutorApprovalService},
    env::ExecutionEnv,
    executors::{BaseCodingAgent, ExecutorExitResult, ExecutorExitSignal, InterruptSender},
    logs::{NormalizedEntryType, utils::patch::extract_normalized_entry_from_patch},
    profile::ExecutorProfileId,
};
use futures::{FutureExt, TryStreamExt, stream::select};
use serde_json::json;
use services::services::{
    analytics::AnalyticsContext,
    approvals::{Approvals, executor_approvals::ExecutorApprovalBridge},
    config::Config,
    container::{ContainerError, ContainerRef, ContainerService},
    diff_stream::{self, DiffStreamHandle},
    git::{Commit, GitCli, GitService},
    image::ImageService,
    notification::NotificationService,
    queued_message::QueuedMessageService,
    share::SharePublisher,
    workspace_manager::{RepoWorkspaceInput, WorkspaceManager},
};
use tokio::{sync::RwLock, task::JoinHandle};
use tokio_util::io::ReaderStream;
use utils::{
    log_msg::LogMsg,
    msg_store::MsgStore,
    text::{git_branch_id, short_uuid, truncate_to_char_boundary},
};
use uuid::Uuid;

use crate::{command, copy};

#[derive(Clone)]
pub struct LocalContainerService {
    db: DBService,
    child_store: Arc<RwLock<HashMap<Uuid, Arc<RwLock<AsyncGroupChild>>>>>,
    interrupt_senders: Arc<RwLock<HashMap<Uuid, InterruptSender>>>,
    msg_stores: Arc<RwLock<HashMap<Uuid, Arc<MsgStore>>>>,
    config: Arc<RwLock<Config>>,
    git: GitService,
    image_service: ImageService,
    analytics: Option<AnalyticsContext>,
    approvals: Approvals,
    queued_message_service: QueuedMessageService,
    publisher: Result<SharePublisher, RemoteClientNotConfigured>,
    notification_service: NotificationService,
}

impl LocalContainerService {
    #[allow(clippy::too_many_arguments)]
    pub async fn new(
        db: DBService,
        msg_stores: Arc<RwLock<HashMap<Uuid, Arc<MsgStore>>>>,
        config: Arc<RwLock<Config>>,
        git: GitService,
        image_service: ImageService,
        analytics: Option<AnalyticsContext>,
        approvals: Approvals,
        queued_message_service: QueuedMessageService,
        publisher: Result<SharePublisher, RemoteClientNotConfigured>,
    ) -> Self {
        let child_store = Arc::new(RwLock::new(HashMap::new()));
        let interrupt_senders = Arc::new(RwLock::new(HashMap::new()));
        let notification_service = NotificationService::new(config.clone());

        let container = LocalContainerService {
            db,
            child_store,
            interrupt_senders,
            msg_stores,
            config,
            git,
            image_service,
            analytics,
            approvals,
            queued_message_service,
            publisher,
            notification_service,
        };

        container.spawn_workspace_cleanup().await;

        container
    }

    pub async fn get_child_from_store(&self, id: &Uuid) -> Option<Arc<RwLock<AsyncGroupChild>>> {
        let map = self.child_store.read().await;
        map.get(id).cloned()
    }

    pub async fn add_child_to_store(&self, id: Uuid, exec: AsyncGroupChild) {
        let mut map = self.child_store.write().await;
        map.insert(id, Arc::new(RwLock::new(exec)));
    }

    pub async fn remove_child_from_store(&self, id: &Uuid) {
        let mut map = self.child_store.write().await;
        map.remove(id);
    }

    async fn add_interrupt_sender(&self, id: Uuid, sender: InterruptSender) {
        let mut map = self.interrupt_senders.write().await;
        map.insert(id, sender);
    }

    async fn take_interrupt_sender(&self, id: &Uuid) -> Option<InterruptSender> {
        let mut map = self.interrupt_senders.write().await;
        map.remove(id)
    }

    pub async fn cleanup_workspace(db: &DBService, workspace: &Workspace) {
        let Some(container_ref) = &workspace.container_ref else {
            return;
        };
        let workspace_dir = PathBuf::from(container_ref);

        let repositories = WorkspaceRepo::find_repos_for_workspace(&db.pool, workspace.id)
            .await
            .unwrap_or_default();

        if repositories.is_empty() {
            tracing::warn!(
                "No repositories found for workspace {}, cleaning up workspace directory only",
                workspace.id
            );
            if workspace_dir.exists()
                && let Err(e) = tokio::fs::remove_dir_all(&workspace_dir).await
            {
                tracing::warn!("Failed to remove workspace directory: {}", e);
            }
        } else {
            WorkspaceManager::cleanup_workspace(&workspace_dir, &repositories)
                .await
                .unwrap_or_else(|e| {
                    tracing::warn!(
                        "Failed to clean up workspace for workspace {}: {}",
                        workspace.id,
                        e
                    );
                });
        }

        // Clear container_ref so this workspace won't be picked up again
        let _ = Workspace::clear_container_ref(&db.pool, workspace.id).await;
    }

    pub async fn cleanup_expired_workspaces(db: &DBService) -> Result<(), DeploymentError> {
        let expired_workspaces = Workspace::find_expired_for_cleanup(&db.pool).await?;
        if expired_workspaces.is_empty() {
            tracing::debug!("No expired workspaces found");
            return Ok(());
        }
        tracing::info!(
            "Found {} expired workspaces to clean up",
            expired_workspaces.len()
        );
        for workspace in &expired_workspaces {
            Self::cleanup_workspace(db, workspace).await;
        }
        Ok(())
    }

    pub async fn spawn_workspace_cleanup(&self) {
        let db = self.db.clone();
        let mut cleanup_interval = tokio::time::interval(tokio::time::Duration::from_secs(1800)); // 30 minutes
        WorkspaceManager::cleanup_orphan_workspaces(&self.db.pool).await;
        tokio::spawn(async move {
            loop {
                cleanup_interval.tick().await;
                tracing::info!("Starting periodic workspace cleanup...");
                Self::cleanup_expired_workspaces(&db)
                    .await
                    .unwrap_or_else(|e| {
                        tracing::error!("Failed to clean up expired workspaces: {}", e)
                    });
            }
        });
    }

    /// Record the current HEAD commit for each repository as the "after" state.
    /// Errors are silently ignored since this runs after the main execution completes
    /// and failure should not block process finalization.
    async fn update_after_head_commits(&self, exec_id: Uuid) {
        if let Ok(ctx) = ExecutionProcess::load_context(&self.db.pool, exec_id).await {
            let workspace_root = self.workspace_to_current_dir(&ctx.workspace);
            for repo in &ctx.repos {
                let repo_path = workspace_root.join(&repo.name);
                if let Ok(head) = self.git().get_head_info(&repo_path) {
                    let _ = ExecutionProcessRepoState::update_after_head_commit(
                        &self.db.pool,
                        exec_id,
                        repo.id,
                        &head.oid,
                    )
                    .await;
                }
            }
        }
    }

    /// Get the commit message based on the execution run reason.
    async fn get_commit_message(&self, ctx: &ExecutionContext) -> String {
        match ctx.execution_process.run_reason {
            ExecutionProcessRunReason::CodingAgent => {
                // Try to retrieve the task summary from the coding agent turn
                // otherwise fallback to default message
                match CodingAgentTurn::find_by_execution_process_id(
                    &self.db().pool,
                    ctx.execution_process.id,
                )
                .await
                {
                    Ok(Some(turn)) if turn.summary.is_some() => turn.summary.unwrap(),
                    Ok(_) => {
                        tracing::debug!(
                            "No summary found for execution process {}, using default message",
                            ctx.execution_process.id
                        );
                        format!(
                            "Commit changes from coding agent for workspace {}",
                            ctx.workspace.id
                        )
                    }
                    Err(e) => {
                        tracing::debug!(
                            "Failed to retrieve summary for execution process {}: {}",
                            ctx.execution_process.id,
                            e
                        );
                        format!(
                            "Commit changes from coding agent for workspace {}",
                            ctx.workspace.id
                        )
                    }
                }
            }
            ExecutionProcessRunReason::CleanupScript => {
                format!("Cleanup script changes for workspace {}", ctx.workspace.id)
            }
            _ => format!(
                "Changes from execution process {}",
                ctx.execution_process.id
            ),
        }
    }

    /// Check which repos have uncommitted changes. Fails if any repo is inaccessible.
    fn check_repos_for_changes(
        &self,
        workspace_root: &Path,
        repos: &[Repo],
    ) -> Result<Vec<(Repo, PathBuf)>, ContainerError> {
        let git = GitCli::new();
        let mut repos_with_changes = Vec::new();

        for repo in repos {
            let worktree_path = workspace_root.join(&repo.name);

            match git.has_changes(&worktree_path) {
                Ok(true) => {
                    repos_with_changes.push((repo.clone(), worktree_path));
                }
                Ok(false) => {
                    tracing::debug!("No changes in repo '{}'", repo.name);
                }
                Err(e) => {
                    return Err(ContainerError::Other(anyhow!(
                        "Pre-flight check failed for repo '{}': {}",
                        repo.name,
                        e
                    )));
                }
            }
        }

        Ok(repos_with_changes)
    }

    /// Commit changes to each repo. Logs failures but continues with other repos.
    fn commit_repos(&self, repos_with_changes: Vec<(Repo, PathBuf)>, message: &str) -> bool {
        let mut any_committed = false;

        for (repo, worktree_path) in repos_with_changes {
            tracing::debug!(
                "Committing changes for repo '{}' at {:?}",
                repo.name,
                &worktree_path
            );

            match self.git().commit(&worktree_path, message) {
                Ok(true) => {
                    any_committed = true;
                    tracing::info!("Committed changes in repo '{}'", repo.name);
                }
                Ok(false) => {
                    tracing::warn!("No changes committed in repo '{}' (unexpected)", repo.name);
                }
                Err(e) => {
                    tracing::warn!("Failed to commit in repo '{}': {}", repo.name, e);
                }
            }
        }

        any_committed
    }

    /// Spawn a background task that polls the child process for completion and
    /// cleans up the execution entry when it exits.
    pub fn spawn_exit_monitor(
        &self,
        exec_id: &Uuid,
        exit_signal: Option<ExecutorExitSignal>,
    ) -> JoinHandle<()> {
        let exec_id = *exec_id;
        let child_store = self.child_store.clone();
        let msg_stores = self.msg_stores.clone();
        let db = self.db.clone();
        let config = self.config.clone();
        let container = self.clone();
        let analytics = self.analytics.clone();
        let publisher = self.publisher.clone();

        let mut process_exit_rx = self.spawn_os_exit_watcher(exec_id);

        tokio::spawn(async move {
            let mut exit_signal_future = exit_signal
                .map(|rx| rx.boxed()) // wait for result
                .unwrap_or_else(|| std::future::pending().boxed()); // no signal, stall forever

            let status_result: std::io::Result<std::process::ExitStatus>;

            // Wait for process to exit, or exit signal from executor
            tokio::select! {
                // Exit signal with result.
                // Some coding agent processes do not automatically exit after processing the user request; instead the executor
                // signals when processing has finished to gracefully kill the process.
                exit_result = &mut exit_signal_future => {
                    // Executor signaled completion: kill group and use the provided result
                    if let Some(child_lock) = child_store.read().await.get(&exec_id).cloned() {
                        let mut child = child_lock.write().await ;
                        if let Err(err) = command::kill_process_group(&mut child).await {
                            tracing::error!("Failed to kill process group after exit signal: {} {}", exec_id, err);
                        }
                    }

                    // Map the exit result to appropriate exit status
                    status_result = match exit_result {
                        Ok(ExecutorExitResult::Success) => Ok(success_exit_status()),
                        Ok(ExecutorExitResult::Failure) => Ok(failure_exit_status()),
                        Err(_) => Ok(success_exit_status()), // Channel closed, assume success
                    };
                }
                // Process exit
                exit_status_result = &mut process_exit_rx => {
                    status_result = exit_status_result.unwrap_or_else(|e| Err(std::io::Error::other(e)));
                }
            }

            let (exit_code, status) = match status_result {
                Ok(exit_status) => {
                    let code = exit_status.code().unwrap_or(-1) as i64;
                    let status = if exit_status.success() {
                        ExecutionProcessStatus::Completed
                    } else {
                        ExecutionProcessStatus::Failed
                    };
                    (Some(code), status)
                }
                Err(_) => (None, ExecutionProcessStatus::Failed),
            };

            if !ExecutionProcess::was_stopped(&db.pool, exec_id).await
                && let Err(e) =
                    ExecutionProcess::update_completion(&db.pool, exec_id, status, exit_code).await
            {
                tracing::error!("Failed to update execution process completion: {}", e);
            }

            if let Ok(ctx) = ExecutionProcess::load_context(&db.pool, exec_id).await {
                // Update executor session summary if available
                if let Err(e) = container.update_executor_session_summary(&exec_id).await {
                    tracing::warn!("Failed to update executor session summary: {}", e);
                }

                let success = matches!(
                    ctx.execution_process.status,
                    ExecutionProcessStatus::Completed
                ) && exit_code == Some(0);

                let cleanup_done = matches!(
                    ctx.execution_process.run_reason,
                    ExecutionProcessRunReason::CleanupScript
                ) && !matches!(
                    ctx.execution_process.status,
                    ExecutionProcessStatus::Running
                );

                if success || cleanup_done {
                    // Commit changes (if any) and get feedback about whether changes were made
                    let changes_committed = match container.try_commit_changes(&ctx).await {
                        Ok(committed) => committed,
                        Err(e) => {
                            tracing::error!("Failed to commit changes after execution: {}", e);
                            // Treat commit failures as if changes were made to be safe
                            true
                        }
                    };

                    let should_start_next = if matches!(
                        ctx.execution_process.run_reason,
                        ExecutionProcessRunReason::CodingAgent
                    ) {
                        changes_committed
                    } else {
                        true
                    };

                    if should_start_next {
                        // If the process exited successfully, start the next action
                        if let Err(e) = container.try_start_next_action(&ctx).await {
                            tracing::error!("Failed to start next action after completion: {}", e);
                        }
                    } else {
                        tracing::info!(
                            "Skipping cleanup script for workspace {} - no changes made by coding agent",
                            ctx.workspace.id
                        );

                        // Manually finalize task since we're bypassing normal execution flow
                        container.finalize_task(publisher.as_ref().ok(), &ctx).await;
                    }
                }

                if container.should_finalize(&ctx) {
                    // Only execute queued messages if the execution succeeded
                    // If it failed or was killed, just clear the queue and finalize
                    let should_execute_queued = !matches!(
                        ctx.execution_process.status,
                        ExecutionProcessStatus::Failed | ExecutionProcessStatus::Killed
                    );

                    if let Some(queued_msg) =
                        container.queued_message_service.take_queued(ctx.session.id)
                    {
                        if should_execute_queued {
                            tracing::info!(
                                "Found queued message for session {}, starting follow-up execution",
                                ctx.session.id
                            );

                            // Delete the scratch since we're consuming the queued message
                            if let Err(e) = Scratch::delete(
                                &db.pool,
                                ctx.session.id,
                                &ScratchType::DraftFollowUp,
                            )
                            .await
                            {
                                tracing::warn!(
                                    "Failed to delete scratch after consuming queued message: {}",
                                    e
                                );
                            }

                            // Execute the queued follow-up
                            if let Err(e) = container
                                .start_queued_follow_up(&ctx, &queued_msg.data)
                                .await
                            {
                                tracing::error!("Failed to start queued follow-up: {}", e);
                                // Fall back to finalization if follow-up fails
                                container.finalize_task(publisher.as_ref().ok(), &ctx).await;
                            }
                        } else {
                            // Execution failed or was killed - discard the queued message and finalize
                            tracing::info!(
                                "Discarding queued message for session {} due to execution status {:?}",
                                ctx.session.id,
                                ctx.execution_process.status
                            );
                            container.finalize_task(publisher.as_ref().ok(), &ctx).await;
                        }
                    } else {
                        container.finalize_task(publisher.as_ref().ok(), &ctx).await;
                    }
                }

                // Fire analytics event when CodingAgent execution has finished
                if config.read().await.analytics_enabled
                    && matches!(
                        &ctx.execution_process.run_reason,
                        ExecutionProcessRunReason::CodingAgent
                    )
                    && let Some(analytics) = &analytics
                {
                    analytics.analytics_service.track_event(&analytics.user_id, "task_attempt_finished", Some(json!({
                        "task_id": ctx.task.id.to_string(),
                        "project_id": ctx.task.project_id.to_string(),
                        "workspace_id": ctx.workspace.id.to_string(),
                        "session_id": ctx.session.id.to_string(),
                        "execution_success": matches!(ctx.execution_process.status, ExecutionProcessStatus::Completed),
                        "exit_code": ctx.execution_process.exit_code,
                    })));
                }
            }

            // Now that commit/next-action/finalization steps for this process are complete,
            // capture the HEAD OID as the definitive "after" state (best-effort).
            container.update_after_head_commits(exec_id).await;

            // Cleanup msg store
            if let Some(msg_arc) = msg_stores.write().await.remove(&exec_id) {
                msg_arc.push_finished();
                tokio::time::sleep(Duration::from_millis(50)).await; // Wait for the finish message to propogate
                match Arc::try_unwrap(msg_arc) {
                    Ok(inner) => drop(inner),
                    Err(arc) => tracing::error!(
                        "There are still {} strong Arcs to MsgStore for {}",
                        Arc::strong_count(&arc),
                        exec_id
                    ),
                }
            }

            // Cleanup child handle
            child_store.write().await.remove(&exec_id);
        })
    }

    pub fn spawn_os_exit_watcher(
        &self,
        exec_id: Uuid,
    ) -> tokio::sync::oneshot::Receiver<std::io::Result<std::process::ExitStatus>> {
        let (tx, rx) = tokio::sync::oneshot::channel::<std::io::Result<std::process::ExitStatus>>();
        let child_store = self.child_store.clone();
        tokio::spawn(async move {
            loop {
                let child_lock = {
                    let map = child_store.read().await;
                    map.get(&exec_id).cloned()
                };
                if let Some(child_lock) = child_lock {
                    let mut child_handler = child_lock.write().await;
                    match child_handler.try_wait() {
                        Ok(Some(status)) => {
                            let _ = tx.send(Ok(status));
                            break;
                        }
                        Ok(None) => {}
                        Err(e) => {
                            let _ = tx.send(Err(e));
                            break;
                        }
                    }
                } else {
                    let _ = tx.send(Err(io::Error::other(format!(
                        "Child handle missing for {exec_id}"
                    ))));
                    break;
                }
                tokio::time::sleep(Duration::from_millis(250)).await;
            }
        });
        rx
    }

    pub fn dir_name_from_workspace(workspace_id: &Uuid, task_title: &str) -> String {
        let task_title_id = git_branch_id(task_title);
        format!("{}-{}", short_uuid(workspace_id), task_title_id)
    }

    async fn track_child_msgs_in_store(&self, id: Uuid, child: &mut AsyncGroupChild) {
        let store = Arc::new(MsgStore::new());

        let out = child.inner().stdout.take().expect("no stdout");
        let err = child.inner().stderr.take().expect("no stderr");

        // Map stdout bytes -> LogMsg::Stdout
        let out = ReaderStream::new(out)
            .map_ok(|chunk| LogMsg::Stdout(String::from_utf8_lossy(&chunk).into_owned()));

        // Map stderr bytes -> LogMsg::Stderr
        let err = ReaderStream::new(err)
            .map_ok(|chunk| LogMsg::Stderr(String::from_utf8_lossy(&chunk).into_owned()));

        // If you have a JSON Patch source, map it to LogMsg::JsonPatch too, then select all three.

        // Merge and forward into the store
        let merged = select(out, err); // Stream<Item = Result<LogMsg, io::Error>>
        store.clone().spawn_forwarder(merged);

        let mut map = self.msg_stores().write().await;
        map.insert(id, store);
    }

    /// Create a live diff log stream for ongoing attempts for WebSocket
    /// Returns a stream that owns the filesystem watcher - when dropped, watcher is cleaned up
    async fn create_live_diff_stream(
        &self,
        worktree_path: &Path,
        base_commit: &Commit,
        stats_only: bool,
        path_prefix: Option<String>,
    ) -> Result<DiffStreamHandle, ContainerError> {
        diff_stream::create(
            self.git().clone(),
            worktree_path.to_path_buf(),
            base_commit.clone(),
            stats_only,
            path_prefix,
        )
        .await
        .map_err(|e| ContainerError::Other(anyhow!("{e}")))
    }

    /// Extract the last assistant message from the MsgStore history
    fn extract_last_assistant_message(&self, exec_id: &Uuid) -> Option<String> {
        // Get the MsgStore for this execution
        let msg_stores = self.msg_stores.try_read().ok()?;
        let msg_store = msg_stores.get(exec_id)?;

        // Get the history and scan in reverse for the last assistant message
        let history = msg_store.get_history();

        for msg in history.iter().rev() {
            if let LogMsg::JsonPatch(patch) = msg {
                // Try to extract a NormalizedEntry from the patch
                if let Some((_, entry)) = extract_normalized_entry_from_patch(patch)
                    && matches!(entry.entry_type, NormalizedEntryType::AssistantMessage)
                {
                    let content = entry.content.trim();
                    if !content.is_empty() {
                        const MAX_SUMMARY_LENGTH: usize = 4096;
                        if content.len() > MAX_SUMMARY_LENGTH {
                            let truncated = truncate_to_char_boundary(content, MAX_SUMMARY_LENGTH);
                            return Some(format!("{truncated}..."));
                        }
                        return Some(content.to_string());
                    }
                }
            }
        }

        None
    }

    /// Update the coding agent turn summary with the final assistant message
    async fn update_executor_session_summary(&self, exec_id: &Uuid) -> Result<(), anyhow::Error> {
        // Check if there's a coding agent turn for this execution process
        let turn = CodingAgentTurn::find_by_execution_process_id(&self.db.pool, *exec_id).await?;

        if let Some(turn) = turn {
            // Only update if summary is not already set
            if turn.summary.is_none() {
                if let Some(summary) = self.extract_last_assistant_message(exec_id) {
                    CodingAgentTurn::update_summary(&self.db.pool, *exec_id, &summary).await?;
                } else {
                    tracing::debug!("No assistant message found for execution {}", exec_id);
                }
            }
        }

        Ok(())
    }

    /// Copy project files and images to the workspace.
    /// Skips files/images that already exist (fast no-op if all exist).
    async fn copy_files_and_images(
        &self,
        workspace_dir: &Path,
        workspace: &Workspace,
    ) -> Result<(), ContainerError> {
        let repos = WorkspaceRepo::find_repos_with_copy_files(&self.db.pool, workspace.id).await?;

        for repo in &repos {
            if let Some(copy_files) = &repo.copy_files
                && !copy_files.trim().is_empty()
            {
                let worktree_path = workspace_dir.join(&repo.name);
                self.copy_project_files(&repo.path, &worktree_path, copy_files)
                    .await
                    .unwrap_or_else(|e| {
                        tracing::warn!(
                            "Failed to copy project files for repo '{}': {}",
                            repo.name,
                            e
                        );
                    });
            }
        }

        if let Err(e) = self
            .image_service
            .copy_images_by_task_to_worktree(workspace_dir, workspace.task_id)
            .await
        {
            tracing::warn!("Failed to copy task images to workspace: {}", e);
        }

        Ok(())
    }

    /// Create workspace-level CLAUDE.md and AGENTS.md files that import from each repo.
    /// Uses the @import syntax to reference each repo's config files.
    /// Skips creating files if they already exist or if no repos have the source file.
    async fn create_workspace_config_files(
        workspace_dir: &Path,
        repos: &[Repo],
    ) -> Result<(), ContainerError> {
        const CONFIG_FILES: [&str; 2] = ["CLAUDE.md", "AGENTS.md"];

        for config_file in CONFIG_FILES {
            let workspace_config_path = workspace_dir.join(config_file);

            if workspace_config_path.exists() {
                tracing::debug!(
                    "Workspace config file {} already exists, skipping",
                    config_file
                );
                continue;
            }

            let mut import_lines = Vec::new();
            for repo in repos {
                let repo_config_path = workspace_dir.join(&repo.name).join(config_file);
                if repo_config_path.exists() {
                    import_lines.push(format!("@{}/{}", repo.name, config_file));
                }
            }

            if import_lines.is_empty() {
                tracing::debug!(
                    "No repos have {}, skipping workspace config creation",
                    config_file
                );
                continue;
            }

            let content = import_lines.join("\n") + "\n";
            if let Err(e) = tokio::fs::write(&workspace_config_path, &content).await {
                tracing::warn!(
                    "Failed to create workspace config file {}: {}",
                    config_file,
                    e
                );
                continue;
            }

            tracing::info!(
                "Created workspace {} with {} import(s)",
                config_file,
                import_lines.len()
            );
        }

        Ok(())
    }

    /// Start a follow-up execution from a queued message
    async fn start_queued_follow_up(
        &self,
        ctx: &ExecutionContext,
        queued_data: &DraftFollowUpData,
    ) -> Result<ExecutionProcess, ContainerError> {
        // Get executor profile from the latest CodingAgent process in this session
        let initial_executor_profile_id =
            ExecutionProcess::latest_executor_profile_for_session(&self.db.pool, ctx.session.id)
                .await
                .map_err(|e| {
                    ContainerError::Other(anyhow!("Failed to get executor profile: {e}"))
                })?;

        let executor_profile_id = ExecutorProfileId {
            executor: initial_executor_profile_id.executor,
            variant: queued_data.variant.clone(),
        };

        // Get latest agent session ID for session continuity (from coding agent turns)
        let latest_agent_session_id = ExecutionProcess::find_latest_coding_agent_turn_session_id(
            &self.db.pool,
            ctx.session.id,
        )
        .await?;

        let project_repos =
            ProjectRepo::find_by_project_id_with_names(&self.db.pool, ctx.project.id).await?;
        let cleanup_action = self.cleanup_actions_for_repos(&project_repos);

        let working_dir = ctx
            .workspace
            .agent_working_dir
            .as_ref()
            .filter(|dir| !dir.is_empty())
            .cloned();

        let action_type = if let Some(agent_session_id) = latest_agent_session_id {
            ExecutorActionType::CodingAgentFollowUpRequest(CodingAgentFollowUpRequest {
                prompt: queued_data.message.clone(),
                session_id: agent_session_id,
                executor_profile_id: executor_profile_id.clone(),
                working_dir: working_dir.clone(),
            })
        } else {
            ExecutorActionType::CodingAgentInitialRequest(CodingAgentInitialRequest {
                prompt: queued_data.message.clone(),
                executor_profile_id: executor_profile_id.clone(),
                working_dir,
            })
        };

        let action = ExecutorAction::new(action_type, cleanup_action.map(Box::new));

        self.start_execution(
            &ctx.workspace,
            &ctx.session,
            &action,
            &ExecutionProcessRunReason::CodingAgent,
        )
        .await
    }
}

fn failure_exit_status() -> std::process::ExitStatus {
    #[cfg(unix)]
    {
        use std::os::unix::process::ExitStatusExt;
        ExitStatusExt::from_raw(256) // Exit code 1 (shifted by 8 bits)
    }
    #[cfg(windows)]
    {
        use std::os::windows::process::ExitStatusExt;
        ExitStatusExt::from_raw(1)
    }
}

#[async_trait]
impl ContainerService for LocalContainerService {
    fn msg_stores(&self) -> &Arc<RwLock<HashMap<Uuid, Arc<MsgStore>>>> {
        &self.msg_stores
    }

    fn db(&self) -> &DBService {
        &self.db
    }

    fn git(&self) -> &GitService {
        &self.git
    }

    fn share_publisher(&self) -> Option<&SharePublisher> {
        self.publisher.as_ref().ok()
    }

    fn notification_service(&self) -> &NotificationService {
        &self.notification_service
    }

    async fn git_branch_prefix(&self) -> String {
        self.config.read().await.git_branch_prefix.clone()
    }

    fn workspace_to_current_dir(&self, workspace: &Workspace) -> PathBuf {
        PathBuf::from(workspace.container_ref.clone().unwrap_or_default())
    }

    async fn create(&self, workspace: &Workspace) -> Result<ContainerRef, ContainerError> {
        let task = workspace
            .parent_task(&self.db.pool)
            .await?
            .ok_or(sqlx::Error::RowNotFound)?;

        let workspace_dir_name =
            LocalContainerService::dir_name_from_workspace(&workspace.id, &task.title);
        let workspace_dir = WorkspaceManager::get_workspace_base_dir().join(&workspace_dir_name);

        let workspace_repos =
            WorkspaceRepo::find_by_workspace_id(&self.db.pool, workspace.id).await?;
        if workspace_repos.is_empty() {
            return Err(ContainerError::Other(anyhow!(
                "Workspace has no repositories configured"
            )));
        }

        let repositories =
            WorkspaceRepo::find_repos_for_workspace(&self.db.pool, workspace.id).await?;

        let target_branches: HashMap<_, _> = workspace_repos
            .iter()
            .map(|wr| (wr.repo_id, wr.target_branch.clone()))
            .collect();

        let workspace_inputs: Vec<RepoWorkspaceInput> = repositories
            .iter()
            .map(|repo| {
                let target_branch = target_branches.get(&repo.id).cloned().unwrap_or_default();
                RepoWorkspaceInput::new(repo.clone(), target_branch)
            })
            .collect();

        let created_workspace = WorkspaceManager::create_workspace(
            &workspace_dir,
            &workspace_inputs,
            &workspace.branch,
        )
        .await?;

        // Copy project files and images to workspace
        self.copy_files_and_images(&created_workspace.workspace_dir, workspace)
            .await?;

        Self::create_workspace_config_files(&created_workspace.workspace_dir, &repositories)
            .await?;

        Workspace::update_container_ref(
            &self.db.pool,
            workspace.id,
            &created_workspace.workspace_dir.to_string_lossy(),
        )
        .await?;

        Ok(created_workspace
            .workspace_dir
            .to_string_lossy()
            .to_string())
    }

    async fn delete(&self, workspace: &Workspace) -> Result<(), ContainerError> {
        self.try_stop(workspace, true).await;
        Self::cleanup_workspace(&self.db, workspace).await;
        Ok(())
    }

    async fn ensure_container_exists(
        &self,
        workspace: &Workspace,
    ) -> Result<ContainerRef, ContainerError> {
        let repositories =
            WorkspaceRepo::find_repos_for_workspace(&self.db.pool, workspace.id).await?;

        if repositories.is_empty() {
            return Err(ContainerError::Other(anyhow!(
                "Workspace has no repositories configured"
            )));
        }

        let workspace_dir = if let Some(container_ref) = &workspace.container_ref {
            PathBuf::from(container_ref)
        } else {
            let task = workspace
                .parent_task(&self.db.pool)
                .await?
                .ok_or(sqlx::Error::RowNotFound)?;
            let workspace_dir_name =
                LocalContainerService::dir_name_from_workspace(&workspace.id, &task.title);
            WorkspaceManager::get_workspace_base_dir().join(&workspace_dir_name)
        };

        WorkspaceManager::ensure_workspace_exists(&workspace_dir, &repositories, &workspace.branch)
            .await?;

        if workspace.container_ref.is_none() {
            Workspace::update_container_ref(
                &self.db.pool,
                workspace.id,
                &workspace_dir.to_string_lossy(),
            )
            .await?;
        }

        // Copy project files and images (fast no-op if already exist)
        self.copy_files_and_images(&workspace_dir, workspace)
            .await?;

        Self::create_workspace_config_files(&workspace_dir, &repositories).await?;

        Ok(workspace_dir.to_string_lossy().to_string())
    }

    async fn is_container_clean(&self, workspace: &Workspace) -> Result<bool, ContainerError> {
        let Some(container_ref) = &workspace.container_ref else {
            return Ok(true);
        };

        let workspace_dir = PathBuf::from(container_ref);
        if !workspace_dir.exists() {
            return Ok(true);
        }

        let repositories =
            WorkspaceRepo::find_repos_for_workspace(&self.db.pool, workspace.id).await?;

        for repo in &repositories {
            let worktree_path = workspace_dir.join(&repo.name);
            if worktree_path.exists() && !self.git().is_worktree_clean(&worktree_path)? {
                return Ok(false);
            }
        }

        Ok(true)
    }

    async fn start_execution_inner(
        &self,
        workspace: &Workspace,
        execution_process: &ExecutionProcess,
        executor_action: &ExecutorAction,
    ) -> Result<(), ContainerError> {
        // Get the worktree path
        let container_ref = workspace
            .container_ref
            .as_ref()
            .ok_or(ContainerError::Other(anyhow!(
                "Container ref not found for workspace"
            )))?;
        let current_dir = PathBuf::from(container_ref);

        let approvals_service: Arc<dyn ExecutorApprovalService> =
            match executor_action.base_executor() {
                Some(
                    BaseCodingAgent::Codex
                    | BaseCodingAgent::ClaudeCode
                    | BaseCodingAgent::Gemini
                    | BaseCodingAgent::QwenCode
                    | BaseCodingAgent::Opencode,
                ) => ExecutorApprovalBridge::new(
                    self.approvals.clone(),
                    self.db.clone(),
                    self.notification_service.clone(),
                    execution_process.id,
                ),
                _ => Arc::new(NoopExecutorApprovalService {}),
            };

        // Build ExecutionEnv with VK_* variables
        let mut env = ExecutionEnv::new();

        // Load task and project context for environment variables
        let task = workspace
            .parent_task(&self.db.pool)
            .await?
            .ok_or(ContainerError::Other(anyhow!(
                "Task not found for workspace"
            )))?;
        let project = task
            .parent_project(&self.db.pool)
            .await?
            .ok_or(ContainerError::Other(anyhow!("Project not found for task")))?;

        env.insert("VK_PROJECT_NAME", &project.name);
        env.insert("VK_PROJECT_ID", project.id.to_string());
        env.insert("VK_TASK_ID", task.id.to_string());
        env.insert("VK_WORKSPACE_ID", workspace.id.to_string());
        env.insert("VK_WORKSPACE_BRANCH", &workspace.branch);

        // Create the child and stream, add to execution tracker with timeout
        let mut spawned = tokio::time::timeout(
            Duration::from_secs(30),
            executor_action.spawn(&current_dir, approvals_service, &env),
        )
        .await
        .map_err(|_| {
            ContainerError::Other(anyhow!(
                "Timeout: process took more than 30 seconds to start"
            ))
        })??;

        self.track_child_msgs_in_store(execution_process.id, &mut spawned.child)
            .await;

        self.add_child_to_store(execution_process.id, spawned.child)
            .await;

        // Store interrupt sender for graceful shutdown
        if let Some(interrupt_sender) = spawned.interrupt_sender {
            self.add_interrupt_sender(execution_process.id, interrupt_sender)
                .await;
        }

        // Spawn unified exit monitor: watches OS exit and optional executor signal
        let _hn = self.spawn_exit_monitor(&execution_process.id, spawned.exit_signal);

        Ok(())
    }

    async fn stop_execution(
        &self,
        execution_process: &ExecutionProcess,
        status: ExecutionProcessStatus,
    ) -> Result<(), ContainerError> {
        let child = self
            .get_child_from_store(&execution_process.id)
            .await
            .ok_or_else(|| {
                ContainerError::Other(anyhow!("Child process not found for execution"))
            })?;
        let exit_code = if status == ExecutionProcessStatus::Completed {
            Some(0)
        } else {
            None
        };

        ExecutionProcess::update_completion(&self.db.pool, execution_process.id, status, exit_code)
            .await?;

        // Try graceful interrupt first, then force kill
        if let Some(interrupt_sender) = self.take_interrupt_sender(&execution_process.id).await {
            // Send interrupt signal (ignore error if receiver dropped)
            let _ = interrupt_sender.send(());

            // Wait for graceful exit with timeout
            let graceful_exit = {
                let mut child_guard = child.write().await;
                tokio::time::timeout(Duration::from_secs(5), child_guard.wait()).await
            };

            match graceful_exit {
                Ok(Ok(_)) => {
                    tracing::debug!(
                        "Process {} exited gracefully after interrupt",
                        execution_process.id
                    );
                }
                Ok(Err(e)) => {
                    tracing::info!("Error waiting for process {}: {}", execution_process.id, e);
                }
                Err(_) => {
                    tracing::debug!(
                        "Graceful shutdown timed out for process {}, force killing",
                        execution_process.id
                    );
                }
            }
        }

        // Kill the child process and remove from the store
        {
            let mut child_guard = child.write().await;
            if let Err(e) = command::kill_process_group(&mut child_guard).await {
                tracing::error!(
                    "Failed to stop execution process {}: {}",
                    execution_process.id,
                    e
                );
                return Err(e);
            }
        }
        self.remove_child_from_store(&execution_process.id).await;

        // Mark the process finished in the MsgStore
        if let Some(msg) = self.msg_stores.write().await.remove(&execution_process.id) {
            msg.push_finished();
        }

        // Update task status to InReview when execution is stopped
        if let Ok(ctx) = ExecutionProcess::load_context(&self.db.pool, execution_process.id).await
            && !matches!(
                ctx.execution_process.run_reason,
                ExecutionProcessRunReason::DevServer
            )
        {
            match Task::update_status(&self.db.pool, ctx.task.id, TaskStatus::InReview).await {
                Ok(_) => {
                    if let Some(publisher) = self.share_publisher()
                        && let Err(err) = publisher.update_shared_task_by_id(ctx.task.id).await
                    {
                        tracing::warn!(
                            ?err,
                            "Failed to propagate shared task update for {}",
                            ctx.task.id
                        );
                    }
                }
                Err(e) => {
                    tracing::error!("Failed to update task status to InReview: {e}");
                }
            }
        }

        tracing::debug!(
            "Execution process {} stopped successfully",
            execution_process.id
        );

        // Record after-head commit OID (best-effort)
        self.update_after_head_commits(execution_process.id).await;

        Ok(())
    }

    async fn stream_diff(
        &self,
        workspace: &Workspace,
        stats_only: bool,
    ) -> Result<futures::stream::BoxStream<'static, Result<LogMsg, std::io::Error>>, ContainerError>
    {
        let workspace_repos =
            WorkspaceRepo::find_by_workspace_id(&self.db.pool, workspace.id).await?;
        let target_branches: HashMap<_, _> = workspace_repos
            .iter()
            .map(|wr| (wr.repo_id, wr.target_branch.clone()))
            .collect();

        let repositories =
            WorkspaceRepo::find_repos_for_workspace(&self.db.pool, workspace.id).await?;

        let mut streams = Vec::new();

        let container_ref = self.ensure_container_exists(workspace).await?;
        let workspace_root = PathBuf::from(container_ref);

        for repo in repositories {
            let worktree_path = workspace_root.join(&repo.name);
            let branch = &workspace.branch;

            let Some(target_branch) = target_branches.get(&repo.id) else {
                tracing::warn!(
                    "Skipping diff stream for repo {}: no target branch configured",
                    repo.name
                );
                continue;
            };

            let base_commit = match self
                .git()
                .get_base_commit(&repo.path, branch, target_branch)
            {
                Ok(c) => c,
                Err(e) => {
                    tracing::warn!(
                        "Skipping diff stream for repo {}: failed to get base commit: {}",
                        repo.name,
                        e
                    );
                    continue;
                }
            };

            let stream = self
                .create_live_diff_stream(
                    &worktree_path,
                    &base_commit,
                    stats_only,
                    Some(repo.name.clone()),
                )
                .await?;

            streams.push(Box::pin(stream));
        }

        if streams.is_empty() {
            return Ok(Box::pin(futures::stream::empty()));
        }

        // Merge all streams into one
        Ok(Box::pin(futures::stream::select_all(streams)))
    }

    async fn try_commit_changes(&self, ctx: &ExecutionContext) -> Result<bool, ContainerError> {
        if !matches!(
            ctx.execution_process.run_reason,
            ExecutionProcessRunReason::CodingAgent | ExecutionProcessRunReason::CleanupScript,
        ) {
            return Ok(false);
        }

        let message = self.get_commit_message(ctx).await;

        let container_ref = ctx
            .workspace
            .container_ref
            .as_ref()
            .ok_or_else(|| ContainerError::Other(anyhow!("Container reference not found")))?;
        let workspace_root = PathBuf::from(container_ref);

        let repos_with_changes = self.check_repos_for_changes(&workspace_root, &ctx.repos)?;
        if repos_with_changes.is_empty() {
            tracing::debug!("No changes to commit in any repository");
            return Ok(false);
        }

        Ok(self.commit_repos(repos_with_changes, &message))
    }

    /// Copy files from the original project directory to the worktree.
    /// Skips files that already exist at target with same size.
    async fn copy_project_files(
        &self,
        source_dir: &Path,
        target_dir: &Path,
        copy_files: &str,
    ) -> Result<(), ContainerError> {
        let source_dir = source_dir.to_path_buf();
        let target_dir = target_dir.to_path_buf();
        let copy_files = copy_files.to_string();

        tokio::time::timeout(
            std::time::Duration::from_secs(30),
            tokio::task::spawn_blocking(move || {
                copy::copy_project_files_impl(&source_dir, &target_dir, &copy_files)
            }),
        )
        .await
        .map_err(|_| ContainerError::Other(anyhow!("Copy project files timed out after 30s")))?
        .map_err(|e| ContainerError::Other(anyhow!("Copy files task failed: {e}")))?
    }

    async fn kill_all_running_processes(&self) -> Result<(), ContainerError> {
        tracing::info!("Killing all running processes");
        let running_processes = ExecutionProcess::find_running(&self.db.pool).await?;

        for process in running_processes {
            if let Err(error) = self
                .stop_execution(&process, ExecutionProcessStatus::Killed)
                .await
            {
                tracing::error!(
                    "Failed to cleanly kill running execution process {:?}: {:?}",
                    process,
                    error
                );
            }
        }

        Ok(())
    }
}
fn success_exit_status() -> std::process::ExitStatus {
    #[cfg(unix)]
    {
        use std::os::unix::process::ExitStatusExt;
        ExitStatusExt::from_raw(0)
    }
    #[cfg(windows)]
    {
        use std::os::windows::process::ExitStatusExt;
        ExitStatusExt::from_raw(0)
    }
}