feat(backend,application,frontend,domain,web-server,app-tauri): bouton Redémarrer l'agent dans la barre supérieure des cellules — #317 (QA verte)

La cellule expose un bouton Redémarrer l'agent valable pour les trois
modes de présentation (headless, TUI native, CLI custom), sans jamais
prendre la présentation pour vérité runtime.

- Application: use case RestartCellAgentRuntime validant le binding
  exact cellule/runtime via les snapshots live (RestartRuntimeNotFound,
  ConflictAgentVisibleElsewhere, AgentNotBoundToCell) puis déléguant au
  scheduler self-restart existant (schedule_cell); CellAgentPresentation
  purement informationnelle.
- Backend/app-tauri/web-server: commande restart_cell_agent sur contrat
  DTO identique Tauri/HTTP, stop borné par timeout typé
  (RestartStopTimeout), relance préservant routage TUI natif, géométrie
  de cellule (node/rows/cols) et conversation; évenements domaine
  enrichis.
- Frontend: bouton dans la barre de cellule avec confirmation,
  overlay d'état Redémarrage…/échec (libellés par code d'erreur, Fermer/
  Réessayer), gardes d'identité de cellule et de restarts externes,
  gateway/mock/http adapters et ports alignés.
- Tests: dto_agents, workstate_actions, contrats web-server,
  LayoutGrid.cellControlsLayering, adapters agent/mock/http.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-09-11 15:00:43 +02:00
parent 46449bdf46
commit 7fb9d8bb86
26 changed files with 1404 additions and 58 deletions

View File

@ -26,11 +26,12 @@ use application::{
ReadAgentContextInput, ReadConversationPageInput, ReadMcpToolPermissionsInput,
ReadMemoryIndexInput, ReadProjectContextInput, RecallMemoryInput, RenameDeviceInput,
RenameLayoutInput, ResolveAgentPermissionsInput, ResolveAgentSystemPermissionsInput,
ResolveMemoryLinksInput, RevokeDeviceInput, RotateConversationLogInput, SaveChatDraftInput,
SetActiveLayoutInput, SetAgentMcpToolAllowedInput, SlashCommandEffect,
SnapshotRunningAgentsInput, StopLiveAgentInput, SyncAgentWithTemplateInput,
UnassignSkillFromAgentInput, UpdateAgentContactableInput, UpdateAgentContextInput,
UpdateAgentEffortInput, UpdateAgentMcpToolPermissionsInput, UpdateAgentPermissionsInput,
ResolveMemoryLinksInput, RestartCellAgentRuntimeInput, RevokeDeviceInput,
RotateConversationLogInput, SaveChatDraftInput, SetActiveLayoutInput,
SetAgentMcpToolAllowedInput, SlashCommandEffect, SnapshotRunningAgentsInput,
StopLiveAgentInput, SyncAgentWithTemplateInput, UnassignSkillFromAgentInput,
UpdateAgentContactableInput, UpdateAgentContextInput, UpdateAgentEffortInput,
UpdateAgentMcpToolPermissionsInput, UpdateAgentPermissionsInput,
UpdateAgentSystemPermissionsInput, UpdateMemoryInput, UpdateProjectContextInput,
UpdateProjectMcpToolPermissionsInput, UpdateProjectOrchestratorInput,
UpdateProjectPermissionsInput, UpdateProjectSystemPermissionsInput, UpdateSkillInput,
@ -73,6 +74,7 @@ use crate::dto::{
RemoveProjectResponseDto, RenameLayoutRequestDto, ReplyChunk, ResizeTerminalRequestDto,
ResolveAgentPermissionsRequestDto, ResolveAgentPermissionsResponseDto,
ResolveAgentSystemPermissionsRequestDto, ResolvedAgentSystemPermissionsDto,
RestartCellAgentRuntimeRequestDto, RestartCellAgentRuntimeResponseDto,
RestoreOpenProjectsOutputDto, ResumableAgentListDto, RetryOpenProjectRestoreRequestDto,
SaveChatDraftRequestDto, SaveEmbedderProfileRequestDto, SaveModelServerRequestDto,
SaveOpenCodeProviderProfileRequestDto, SaveOpenProjectRestoreOrderRequestDto,
@ -2268,6 +2270,32 @@ pub async fn stop_live_agent(
.map_err(ErrorDto::from)
}
/// Atomically restarts the runtime bound to an exact visible agent cell.
///
/// The presentation hint is retained for correlation only. Runtime selection,
/// stop ordering and relaunch are owned by the backend restart coordinator.
#[tauri::command]
pub async fn restart_cell_agent_runtime(
request: RestartCellAgentRuntimeRequestDto,
state: State<'_, AppState>,
) -> Result<RestartCellAgentRuntimeResponseDto, ErrorDto> {
let project = resolve_project(&request.project_id, &state).await?;
let output = state
.restart_cell_agent_runtime
.execute(RestartCellAgentRuntimeInput {
project,
node_id: parse_node_id(&request.node_id)?,
agent_id: parse_agent_id(&request.agent_id)?,
presentation: request.presentation.into(),
layout_id: request
.layout_id
.as_deref()
.map(parse_layout_id)
.transpose()?,
})?;
Ok(output.into())
}
/// `read_agent_context` — read an agent's Markdown context.
///
/// # Errors

View File

@ -351,6 +351,7 @@ pub fn run() {
commands::list_live_agents,
commands::attach_live_agent,
commands::stop_live_agent,
commands::restart_cell_agent_runtime,
commands::read_agent_context,
commands::update_agent_context,
commands::delete_agent,

View File

@ -741,3 +741,53 @@ fn stop_live_agent_response_serialises_camelcase_with_kind() {
assert_eq!(v["alreadyStopped"], false);
assert!(v.get("session_id").is_none());
}
#[test]
fn restart_cell_agent_request_deserialises_all_presentations() {
use app_tauri_lib::dto::{CellAgentPresentationDto, RestartCellAgentRuntimeRequestDto};
for (wire, expected) in [
("headless", CellAgentPresentationDto::Headless),
("tui", CellAgentPresentationDto::Tui),
("customCli", CellAgentPresentationDto::CustomCli),
("unknown", CellAgentPresentationDto::Unknown),
] {
let raw = json!({
"projectId": Uuid::from_u128(1).to_string(),
"nodeId": Uuid::from_u128(2).to_string(),
"agentId": Uuid::from_u128(3).to_string(),
"presentation": wire,
"layoutId": Uuid::from_u128(4).to_string()
});
let dto: RestartCellAgentRuntimeRequestDto = serde_json::from_value(raw).unwrap();
assert_eq!(dto.presentation, expected);
assert_eq!(dto.layout_id, Some(Uuid::from_u128(4).to_string()));
}
}
#[test]
fn restart_cell_agent_response_uses_restart_event_correlation_contract() {
use app_tauri_lib::dto::RestartCellAgentRuntimeResponseDto;
use application::{CellAgentPresentation, LiveSessionKind, RestartCellAgentRuntimeOutput};
let agent = AgentId::from_uuid(Uuid::from_u128(11));
let node = NodeId::from_uuid(Uuid::from_u128(12));
let layout = domain::LayoutId::from_uuid(Uuid::from_u128(13));
let dto = RestartCellAgentRuntimeResponseDto::from(RestartCellAgentRuntimeOutput {
restart_id: "restart-317".to_owned(),
agent_id: agent,
node_id: node,
kind: LiveSessionKind::Structured,
presentation: CellAgentPresentation::CustomCli,
layout_id: Some(layout),
});
let value = serde_json::to_value(dto).unwrap();
assert_eq!(value["restartId"], "restart-317");
assert_eq!(value["status"], "scheduled");
assert_eq!(value["agentId"], agent.to_string());
assert_eq!(value["nodeId"], node.to_string());
assert_eq!(value["kind"], "structured");
assert_eq!(value["presentation"], "customCli");
assert_eq!(value["layoutId"], layout.to_string());
}

View File

@ -73,6 +73,36 @@ pub enum AppError {
node_id: NodeId,
},
/// No live runtime can be restarted for the requested agent.
#[error("runtime not found for agent {agent_id}")]
RestartRuntimeNotFound {
/// Agent whose runtime was requested.
agent_id: AgentId,
},
/// The requested cell does not host the requested live agent runtime.
#[error("agent {agent_id} is not bound to cell {node_id}")]
AgentNotBoundToCell {
/// Agent supplied by the caller.
agent_id: AgentId,
/// Cell supplied by the caller.
node_id: NodeId,
},
/// More than one live runtime/cell currently represents the same agent.
#[error("agent {agent_id} is visible in more than one live cell")]
ConflictAgentVisibleElsewhere {
/// Ambiguous agent.
agent_id: AgentId,
},
/// Runtime teardown did not reach its terminal boundary in time.
#[error("timed out while stopping agent {agent_id} for restart")]
RestartStopTimeout {
/// Agent whose old runtime could not be stopped conclusively.
agent_id: AgentId,
},
/// The delegation target finished its structured/headless turn without producing
/// a [`ReplyEvent::Final`](domain::ports::ReplyEvent::Final), so the rendezvous
/// could not capture an inline answer. Surfaced promptly instead of blocking the
@ -120,6 +150,10 @@ impl AppError {
Self::Git(_) => "GIT",
Self::Remote(_) => "REMOTE",
Self::AgentAlreadyRunning { .. } => "AGENT_ALREADY_RUNNING",
Self::RestartRuntimeNotFound { .. } => "RUNTIME_NOT_FOUND",
Self::AgentNotBoundToCell { .. } => "AGENT_NOT_BOUND_TO_CELL",
Self::ConflictAgentVisibleElsewhere { .. } => "CONFLICT_AGENT_VISIBLE_ELSEWHERE",
Self::RestartStopTimeout { .. } => "STOP_TIMEOUT",
Self::TargetReturnedNoReply(_) => "TARGET_RETURNED_NO_REPLY",
Self::TargetCeilingActive(_) => "RENDEZVOUS_CEILING_ACTIVE",
Self::Internal(_) => "INTERNAL",

View File

@ -280,11 +280,13 @@ pub use window::{
pub use workstate::{
AgentBackgroundTaskState, AgentTicketState, AgentWorkState, AppExitWorkGuardDetail,
AppExitWorkGuardState, AttachLiveAgent, AttachLiveAgentInput, AttachLiveAgentOutput,
BackgroundTaskKindLabel, ConversationLogProvider, ConversationPreviewStatus,
ConversationTurnWorkPreview, ConversationWorkSummary, GetAppExitWorkGuardState,
GetAppExitWorkGuardStateInput, GetLiveStateLean, GetProjectWorkState, GetProjectWorkStateInput,
LeanLiveEntry, LeanLiveState, LiveWorkSession, ProjectWorkState, ReconcileLiveState,
ReconcileLiveStateInput, StopAgentExecutionCascade, StopAgentRuntime, StopLiveAgent,
StopLiveAgentInput, StopLiveAgentOutput, TicketWorkSource, TicketWorkStatus, UpdateLiveState,
UpdateLiveStateInput, LIVE_STATE_MAX_ENTRIES, LIVE_STATE_TTL_MS,
BackgroundTaskKindLabel, CellAgentPresentation, ConversationLogProvider,
ConversationPreviewStatus, ConversationTurnWorkPreview, ConversationWorkSummary,
GetAppExitWorkGuardState, GetAppExitWorkGuardStateInput, GetLiveStateLean, GetProjectWorkState,
GetProjectWorkStateInput, LeanLiveEntry, LeanLiveState, LiveWorkSession, ProjectWorkState,
ReconcileLiveState, ReconcileLiveStateInput, RestartCellAgentRuntime,
RestartCellAgentRuntimeInput, RestartCellAgentRuntimeOutput, StopAgentExecutionCascade,
StopAgentRuntime, StopLiveAgent, StopLiveAgentInput, StopLiveAgentOutput, TicketWorkSource,
TicketWorkStatus, UpdateLiveState, UpdateLiveStateInput, LIVE_STATE_MAX_ENTRIES,
LIVE_STATE_TTL_MS,
};

View File

@ -32,7 +32,7 @@ use domain::ports::{
use domain::project::ProjectPath;
use domain::{
AgentId, AgentProfile, BackgroundTask, BackgroundTaskKind, BackgroundTaskRendezvousLink,
BackgroundTaskResult, BackgroundTaskState, BackgroundTaskWakePolicy, DomainEvent,
BackgroundTaskResult, BackgroundTaskState, BackgroundTaskWakePolicy, DomainEvent, NodeId,
OrchestratorCommand, OrchestratorVisibility, ProfileId, Project, RendezvousId, RuntimeAgentKey,
TaskId,
};
@ -83,6 +83,20 @@ pub trait AgentSelfRestartScheduler: Send + Sync {
agent: AgentId,
reason: Option<String>,
) -> Result<AgentRestartScheduled, AppError>;
/// Schedules the same restart while pinning it to a caller-validated cell.
/// Implementations must revalidate the binding atomically with their
/// concurrency guard before accepting the request.
fn schedule_cell(
&self,
project: &Project,
agent: AgentId,
node: NodeId,
reason: Option<String>,
) -> Result<AgentRestartScheduled, AppError> {
let _ = node;
self.schedule(project, agent, reason)
}
}
/// Internal continuation prompt used when a self-restart is handed back to the

View File

@ -12,10 +12,10 @@ use std::sync::Arc;
use async_trait::async_trait;
use domain::input::InputMediator;
use domain::live_state::WorkStatus;
use domain::{AgentId, NodeId, Project, RuntimeAgentKey, SessionId};
use domain::{AgentId, LayoutId, NodeId, Project, RuntimeAgentKey, SessionId};
use crate::error::AppError;
use crate::orchestrator::{LiveStateProvider, OrchestratorService};
use crate::orchestrator::{AgentSelfRestartScheduler, LiveStateProvider, OrchestratorService};
use crate::terminal::{
stop_live_agent_runtime_in_project, CloseTerminal, LiveSessionKind, LiveSessions,
};
@ -122,6 +122,113 @@ pub struct StopLiveAgentOutput {
pub already_stopped: bool,
}
/// Caller-side indication of how the hosting cell presents the agent.
///
/// This value is never trusted to select the runtime: the live registries and
/// the launch context remain authoritative.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum CellAgentPresentation {
/// Agent driven without an interactive human surface.
Headless,
/// Native provider terminal/TUI.
Tui,
/// IdeA structured/custom CLI chat.
CustomCli,
/// Older caller that cannot identify its presentation.
Unknown,
}
/// Input for [`RestartCellAgentRuntime::execute`].
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct RestartCellAgentRuntimeInput {
/// Owning project resolved by the driving adapter.
pub project: Project,
/// Exact leaf whose runtime is being restarted.
pub node_id: NodeId,
/// Agent expected to be bound to `node_id`.
pub agent_id: AgentId,
/// Informational UI presentation; never used as runtime truth.
pub presentation: CellAgentPresentation,
/// Optional layout identity supplied by clients that already have it.
pub layout_id: Option<LayoutId>,
}
/// Immediate acknowledgement for an accepted atomic cell restart.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct RestartCellAgentRuntimeOutput {
/// Restart correlation id used by the existing restart event stream.
pub restart_id: String,
/// Validated agent identity.
pub agent_id: AgentId,
/// Validated hosting node.
pub node_id: NodeId,
/// Actual runtime family resolved from the live registry.
pub kind: LiveSessionKind,
/// Informational presentation echoed for caller correlation.
pub presentation: CellAgentPresentation,
/// Optional layout identity echoed for caller correlation.
pub layout_id: Option<LayoutId>,
}
/// Atomically validates a cell/runtime binding and schedules its replacement
/// through IdeA's existing restart coordinator.
pub struct RestartCellAgentRuntime {
live: Arc<LiveSessions>,
scheduler: Arc<dyn AgentSelfRestartScheduler>,
}
impl RestartCellAgentRuntime {
/// Builds the use case over the authoritative runtime registry and restart seam.
#[must_use]
pub fn new(live: Arc<LiveSessions>, scheduler: Arc<dyn AgentSelfRestartScheduler>) -> Self {
Self { live, scheduler }
}
/// Validates the exact live binding before the scheduler repeats that check
/// under its per-agent concurrency guard and accepts the detached restart.
pub fn execute(
&self,
input: RestartCellAgentRuntimeInput,
) -> Result<RestartCellAgentRuntimeOutput, AppError> {
let snapshots = self
.live
.live_agent_snapshots_for(input.project.id, &input.agent_id);
let snapshot = match snapshots.as_slice() {
[] => {
return Err(AppError::RestartRuntimeNotFound {
agent_id: input.agent_id,
});
}
[snapshot] => *snapshot,
_ => {
return Err(AppError::ConflictAgentVisibleElsewhere {
agent_id: input.agent_id,
});
}
};
if snapshot.node_id != input.node_id {
return Err(AppError::AgentNotBoundToCell {
agent_id: input.agent_id,
node_id: input.node_id,
});
}
let scheduled = self.scheduler.schedule_cell(
&input.project,
input.agent_id,
input.node_id,
Some("redémarrage demandé depuis la cellule".to_owned()),
)?;
Ok(RestartCellAgentRuntimeOutput {
restart_id: scheduled.restart_id,
agent_id: input.agent_id,
node_id: input.node_id,
kind: snapshot.kind,
presentation: input.presentation,
layout_id: input.layout_id,
})
}
}
/// Best-effort stopper for execution resources owned by one live agent turn.
///
/// Implementations must be idempotent: a missing, already-finished or already-stopped

View File

@ -10,8 +10,10 @@ mod live;
mod reconcile;
pub use actions::{
AttachLiveAgent, AttachLiveAgentInput, AttachLiveAgentOutput, StopAgentExecutionCascade,
StopAgentRuntime, StopLiveAgent, StopLiveAgentInput, StopLiveAgentOutput,
AttachLiveAgent, AttachLiveAgentInput, AttachLiveAgentOutput, CellAgentPresentation,
RestartCellAgentRuntime, RestartCellAgentRuntimeInput, RestartCellAgentRuntimeOutput,
StopAgentExecutionCascade, StopAgentRuntime, StopLiveAgent, StopLiveAgentInput,
StopLiveAgentOutput,
};
pub use live::{
GetLiveStateLean, LeanLiveEntry, LeanLiveState, UpdateLiveState, UpdateLiveStateInput,

View File

@ -11,9 +11,10 @@ use std::sync::{Arc, Mutex};
use async_trait::async_trait;
use application::{
AttachLiveAgent, AttachLiveAgentInput, CloseTerminal, LiveRuntimeState, LiveSessionKind,
LiveSessions, StopAgentExecutionCascade, StopLiveAgent, StopLiveAgentInput, StructuredSessions,
TerminalSessions,
AgentRestartScheduled, AgentSelfRestartScheduler, AppError, AttachLiveAgent,
AttachLiveAgentInput, CellAgentPresentation, CloseTerminal, LiveRuntimeState, LiveSessionKind,
LiveSessions, RestartCellAgentRuntime, RestartCellAgentRuntimeInput, StopAgentExecutionCascade,
StopLiveAgent, StopLiveAgentInput, StructuredSessions, TerminalSessions,
};
use domain::ports::{
AgentSession, AgentSessionError, ExitStatus, OutputStream, PtyError, PtyHandle, PtyPort,
@ -21,7 +22,7 @@ use domain::ports::{
};
use domain::Project;
use domain::{
AgentId, NodeId, ProjectId, ProjectPath, PtySize, RemoteRef, SessionId, SessionKind,
AgentId, LayoutId, NodeId, ProjectId, ProjectPath, PtySize, RemoteRef, SessionId, SessionKind,
TerminalSession,
};
use uuid::Uuid;
@ -131,6 +132,35 @@ struct FakeCascade {
calls: Mutex<Vec<(ProjectId, AgentId)>>,
}
#[derive(Default)]
struct FakeRestartScheduler {
calls: Mutex<Vec<(ProjectId, AgentId, NodeId)>>,
}
impl AgentSelfRestartScheduler for FakeRestartScheduler {
fn schedule(
&self,
_project: &Project,
_agent: AgentId,
_reason: Option<String>,
) -> Result<AgentRestartScheduled, AppError> {
unreachable!("cell restart must use the cell-aware scheduler seam")
}
fn schedule_cell(
&self,
project: &Project,
agent: AgentId,
node: NodeId,
_reason: Option<String>,
) -> Result<AgentRestartScheduled, AppError> {
self.calls.lock().unwrap().push((project.id, agent, node));
Ok(AgentRestartScheduled {
restart_id: "restart-cell-test".to_owned(),
})
}
}
#[async_trait]
impl StopAgentExecutionCascade for FakeCascade {
async fn stop_agent_execution_children(&self, project: &Project, agent_id: AgentId) {
@ -148,6 +178,8 @@ struct Fixture {
pty: Arc<TerminalSessions>,
structured: Arc<StructuredSessions>,
pty_port: Arc<FakePty>,
restart: RestartCellAgentRuntime,
restart_scheduler: Arc<FakeRestartScheduler>,
}
fn fixture() -> Fixture {
@ -159,12 +191,32 @@ fn fixture() -> Fixture {
Arc::clone(&pty_port) as Arc<dyn PtyPort>,
Arc::clone(&pty),
));
let restart_scheduler = Arc::new(FakeRestartScheduler::default());
Fixture {
attach: AttachLiveAgent::new(Arc::clone(&live)),
stop: StopLiveAgent::new(Arc::clone(&live), close),
pty,
structured,
pty_port,
restart: RestartCellAgentRuntime::new(
Arc::clone(&live),
Arc::clone(&restart_scheduler) as Arc<dyn AgentSelfRestartScheduler>,
),
restart_scheduler,
}
}
fn restart_input(
agent_id: AgentId,
node_id: NodeId,
presentation: CellAgentPresentation,
) -> RestartCellAgentRuntimeInput {
RestartCellAgentRuntimeInput {
project: project(),
node_id,
agent_id,
presentation,
layout_id: Some(LayoutId::from_uuid(Uuid::from_u128(500))),
}
}
@ -315,6 +367,106 @@ fn attach_same_node_is_idempotent() {
assert_eq!(f.pty.len(), 1, "no extra session created");
}
// ---------------------------------------------------------------------------
// RestartCellAgentRuntime
// ---------------------------------------------------------------------------
#[test]
fn restart_cell_accepts_tui_and_preserves_exact_binding() {
let f = fixture();
let agent = aid(10);
let node = nid(100);
insert_pty(&f.pty, sid(1), agent, node);
let out = f
.restart
.execute(restart_input(agent, node, CellAgentPresentation::Tui))
.unwrap();
assert_eq!(out.restart_id, "restart-cell-test");
assert_eq!(out.agent_id, agent);
assert_eq!(out.node_id, node);
assert_eq!(out.kind, LiveSessionKind::Pty);
assert_eq!(out.presentation, CellAgentPresentation::Tui);
assert_eq!(
*f.restart_scheduler.calls.lock().unwrap(),
vec![(project().id, agent, node)]
);
}
#[test]
fn restart_cell_accepts_structured_custom_cli_and_headless_hints() {
for presentation in [
CellAgentPresentation::CustomCli,
CellAgentPresentation::Headless,
] {
let f = fixture();
let agent = aid(10);
let node = nid(100);
insert_structured(&f.structured, sid(2), agent, node);
let out = f
.restart
.execute(restart_input(agent, node, presentation))
.unwrap();
assert_eq!(out.kind, LiveSessionKind::Structured);
assert_eq!(out.presentation, presentation);
assert_eq!(f.restart_scheduler.calls.lock().unwrap().len(), 1);
}
}
#[test]
fn restart_cell_rejects_stale_agent_or_node_without_scheduling() {
let f = fixture();
let agent = aid(10);
insert_pty(&f.pty, sid(1), agent, nid(100));
let error = f
.restart
.execute(restart_input(
agent,
nid(999),
CellAgentPresentation::Unknown,
))
.unwrap_err();
assert_eq!(error.code(), "AGENT_NOT_BOUND_TO_CELL");
assert!(f.restart_scheduler.calls.lock().unwrap().is_empty());
}
#[test]
fn restart_cell_rejects_missing_runtime_without_scheduling() {
let f = fixture();
let error = f
.restart
.execute(restart_input(
aid(999),
nid(100),
CellAgentPresentation::Unknown,
))
.unwrap_err();
assert_eq!(error.code(), "RUNTIME_NOT_FOUND");
assert!(f.restart_scheduler.calls.lock().unwrap().is_empty());
}
#[test]
fn restart_cell_rejects_agent_visible_in_multiple_live_cells() {
let f = fixture();
let agent = aid(10);
insert_pty(&f.pty, sid(1), agent, nid(100));
insert_structured(&f.structured, sid(2), agent, nid(200));
let error = f
.restart
.execute(restart_input(agent, nid(100), CellAgentPresentation::Tui))
.unwrap_err();
assert_eq!(error.code(), "CONFLICT_AGENT_VISIBLE_ELSEWHERE");
assert!(f.restart_scheduler.calls.lock().unwrap().is_empty());
}
// ---------------------------------------------------------------------------
// StopLiveAgent
// ---------------------------------------------------------------------------

View File

@ -12,13 +12,14 @@ use serde_json::Value;
use application::{
AgentBackgroundTaskState, AgentTicketState, AppError, AppExitWorkGuardDetail,
AppExitWorkGuardState, AttachLiveAgentOutput, BackgroundTaskKindLabel,
AppExitWorkGuardState, AttachLiveAgentOutput, BackgroundTaskKindLabel, CellAgentPresentation,
ConversationPreviewStatus, ConversationTurnWorkPreview, ConversationWorkSummary,
CreateProjectInput, CreateProjectOutput, ExecuteSlashCommandOutput, GitGraphOutput,
HealthInput, HealthReport, LayoutKind, ListProjectsOutput, ListSlashCommandsOutput,
LiveRuntimeState, LiveSessionKind, LiveSessionSnapshot, OpenProjectOutput, ProjectWorkState,
RestoreOpenProjectFailure, RestoreOpenProjectsOutput, SlashCommandEffect, StopLiveAgentOutput,
TicketWorkSource, TicketWorkStatus, TurnPage, TurnSource, TurnView,
RestartCellAgentRuntimeOutput, RestoreOpenProjectFailure, RestoreOpenProjectsOutput,
SlashCommandEffect, StopLiveAgentOutput, TicketWorkSource, TicketWorkStatus, TurnPage,
TurnSource, TurnView,
};
use domain::ports::{ReplyProgress, ReplyProgressKind, ReplyProgressSource, ReplyProgressStage};
use domain::{
@ -4611,6 +4612,98 @@ pub struct StopLiveAgentRequestDto {
pub reason: Option<String>,
}
/// Informational presentation supplied by the cell that requests a restart.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub enum CellAgentPresentationDto {
/// Background/headless orchestration surface.
Headless,
/// Native interactive provider terminal.
Tui,
/// IdeA structured/custom CLI chat.
CustomCli,
/// Presentation cannot be determined by the caller.
Unknown,
}
impl From<CellAgentPresentationDto> for CellAgentPresentation {
fn from(value: CellAgentPresentationDto) -> Self {
match value {
CellAgentPresentationDto::Headless => Self::Headless,
CellAgentPresentationDto::Tui => Self::Tui,
CellAgentPresentationDto::CustomCli => Self::CustomCli,
CellAgentPresentationDto::Unknown => Self::Unknown,
}
}
}
impl From<CellAgentPresentation> for CellAgentPresentationDto {
fn from(value: CellAgentPresentation) -> Self {
match value {
CellAgentPresentation::Headless => Self::Headless,
CellAgentPresentation::Tui => Self::Tui,
CellAgentPresentation::CustomCli => Self::CustomCli,
CellAgentPresentation::Unknown => Self::Unknown,
}
}
}
/// Request for an atomic restart of the runtime bound to one agent cell.
#[derive(Debug, Clone, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct RestartCellAgentRuntimeRequestDto {
/// Owning project registry id.
pub project_id: String,
/// Exact hosting layout leaf.
pub node_id: String,
/// Agent expected to be bound to the leaf.
pub agent_id: String,
/// Informational presentation; the backend resolves the real runtime.
#[serde(default = "unknown_cell_agent_presentation")]
pub presentation: CellAgentPresentationDto,
/// Optional named layout id when already known by the caller.
#[serde(default)]
pub layout_id: Option<String>,
}
fn unknown_cell_agent_presentation() -> CellAgentPresentationDto {
CellAgentPresentationDto::Unknown
}
/// Immediate acknowledgement for an accepted cell restart.
#[derive(Debug, Clone, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct RestartCellAgentRuntimeResponseDto {
/// Correlation id used by `agentRestart*` events.
pub restart_id: String,
/// Stable acknowledgement state.
pub status: &'static str,
/// Validated agent id.
pub agent_id: String,
/// Validated node id.
pub node_id: String,
/// Actual live runtime family (`pty` or `structured`).
pub kind: LiveWorkSessionKindDto,
/// Informational presentation echoed to the caller.
pub presentation: CellAgentPresentationDto,
/// Optional layout id echoed to the caller.
pub layout_id: Option<String>,
}
impl From<RestartCellAgentRuntimeOutput> for RestartCellAgentRuntimeResponseDto {
fn from(out: RestartCellAgentRuntimeOutput) -> Self {
Self {
restart_id: out.restart_id,
status: "scheduled",
agent_id: out.agent_id.to_string(),
node_id: out.node_id.to_string(),
kind: out.kind.into(),
presentation: out.presentation.into(),
layout_id: out.layout_id.map(|id| id.to_string()),
}
}
}
/// Response DTO for `stop_live_agent`: the torn-down session's coordinates.
#[derive(Debug, Clone, Serialize)]
#[serde(rename_all = "camelCase")]

View File

@ -139,6 +139,7 @@ pub enum DomainEventDto {
agent_id: String,
restart_id: String,
phase: String,
code: String,
message: String,
},
/// A project was created.
@ -844,11 +845,13 @@ impl From<&DomainEvent> for DomainEventDto {
agent_id,
restart_id,
phase,
code,
message,
} => Self::AgentRestartFailed {
agent_id: agent_id.to_string(),
restart_id: restart_id.clone(),
phase: restart_phase_label(*phase).to_owned(),
code: code.clone(),
message: message.clone(),
},
DomainEvent::ProjectCreated { project_id } => Self::ProjectCreated {
@ -1427,11 +1430,13 @@ mod tests {
agent_id: id,
restart_id: "restart-288".to_owned(),
phase: domain::events::AgentRestartPhase::Launching,
code: "LAUNCH_FAILED".to_owned(),
message: "launch failed".to_owned(),
});
let failed = serde_json::to_value(failed).expect("serialisable");
assert_eq!(failed["type"], "agentRestartFailed");
assert_eq!(failed["phase"], "launching");
assert_eq!(failed["code"], "LAUNCH_FAILED");
assert_eq!(failed["message"], "launch failed");
}

View File

@ -51,21 +51,22 @@ use application::{
ReferenceProfiles, ReleaseManagedTemp, ReloadPlugin, ReloadPluginInput,
RemoveProjectFromRegistry, RenameDevice, RenameLayout, RenameSprint, ReorderSprints,
ResizeTerminal, ResolveAgentCapabilities, ResolveAgentPermissions,
ResolveAgentSystemPermissions, ResolveManagedTemp, ResolveMemoryLinks, RestoreOpenProjects,
RestoreOpenProjectsOutput, RestoreOpenWindows, RetryBackgroundTask, RetryOpenProjectRestore,
ReviewPluginPackage, RevokeAllDevices, RevokeDevice, RotateConversationLog, SaveChatDraft,
SaveEmbedderProfile, SaveModelServer, SaveOpenCodeProviderProfile, SaveOpenProjectRestoreOrder,
SaveProfile, ScheduledProjectTaskService, SessionLimitService, SetActiveLayout,
SetAgentMcpToolAllowed, SetPluginEnabled, SetProjectPluginEnabled, SlashCommandRegistry,
SnapshotOpenWindows, SnapshotRunningAgents, SnapshotRunningAgentsInput, SpawnBackgroundCommand,
StopAgentRuntime, StructuredRoutingMode, StructuredSessions, SuggestedThisSession,
SyncAgentWithTemplate, TerminalSessions, TouchDevice, UnassignSkillFromAgent,
UnassignTicketFromSprint, UninstallPlugin, UnlinkIssues, UpdateAgentContactable,
UpdateAgentContext, UpdateAgentEffort, UpdateAgentMcpToolPermissions, UpdateAgentPermissions,
UpdateAgentSystemPermissions, UpdateIssue, UpdateIssueCarnet, UpdateLiveState, UpdateMemory,
UpdateProjectContext, UpdateProjectMcpToolPermissions, UpdateProjectOrchestrator,
UpdateProjectPermissions, UpdateProjectSettings, UpdateProjectSystemPermissions, UpdateSkill,
UpdateTemplate, UpdateUserSettings, WakeSessionProvider, WriteMemory, WriteToTerminal,
ResolveAgentSystemPermissions, ResolveManagedTemp, ResolveMemoryLinks, RestartCellAgentRuntime,
RestoreOpenProjects, RestoreOpenProjectsOutput, RestoreOpenWindows, RetryBackgroundTask,
RetryOpenProjectRestore, ReviewPluginPackage, RevokeAllDevices, RevokeDevice,
RotateConversationLog, SaveChatDraft, SaveEmbedderProfile, SaveModelServer,
SaveOpenCodeProviderProfile, SaveOpenProjectRestoreOrder, SaveProfile,
ScheduledProjectTaskService, SessionLimitService, SetActiveLayout, SetAgentMcpToolAllowed,
SetPluginEnabled, SetProjectPluginEnabled, SlashCommandRegistry, SnapshotOpenWindows,
SnapshotRunningAgents, SnapshotRunningAgentsInput, SpawnBackgroundCommand, StopAgentRuntime,
StructuredRoutingMode, StructuredSessions, SuggestedThisSession, SyncAgentWithTemplate,
TerminalSessions, TouchDevice, UnassignSkillFromAgent, UnassignTicketFromSprint,
UninstallPlugin, UnlinkIssues, UpdateAgentContactable, UpdateAgentContext, UpdateAgentEffort,
UpdateAgentMcpToolPermissions, UpdateAgentPermissions, UpdateAgentSystemPermissions,
UpdateIssue, UpdateIssueCarnet, UpdateLiveState, UpdateMemory, UpdateProjectContext,
UpdateProjectMcpToolPermissions, UpdateProjectOrchestrator, UpdateProjectPermissions,
UpdateProjectSettings, UpdateProjectSystemPermissions, UpdateSkill, UpdateTemplate,
UpdateUserSettings, WakeSessionProvider, WriteMemory, WriteToTerminal,
AGENT_MEMORY_RECALL_BUDGET,
};
use async_trait::async_trait;
@ -1090,6 +1091,19 @@ struct AppAgentSelfRestarter {
rendezvous: Arc<application::RestartRendezvousCoordinator>,
}
async fn await_cell_restart_stop<F, T>(
agent_id: AgentId,
timeout: std::time::Duration,
stop: F,
) -> Result<T, AppError>
where
F: std::future::Future<Output = Result<T, AppError>>,
{
tokio::time::timeout(timeout, stop)
.await
.map_err(|_| AppError::RestartStopTimeout { agent_id })?
}
fn self_restart_launch_input(
context: ResumeContext,
agent_id: AgentId,
@ -1215,6 +1229,41 @@ mod self_restart_contract_tests {
assert!(!input.fresh_start);
}
#[test]
fn relaunch_preserves_native_tui_routing_and_cell_geometry() {
let project = Project::new(
ProjectId::from_uuid(Uuid::from_u128(317)),
"restart-tui",
domain::project::ProjectPath::new("/tmp/idea-restart-tui").unwrap(),
domain::remote::RemoteRef::local(),
0,
)
.unwrap();
let agent = AgentId::from_uuid(Uuid::from_u128(1));
let node = domain::NodeId::from_uuid(Uuid::from_u128(2));
let input = self_restart_launch_input(
ResumeContext {
project,
rows: 42,
cols: 132,
node_id: Some(node),
conversation_id: Some("conversation-tui".to_owned()),
require_structured: false,
},
agent,
None,
);
assert_eq!(input.node_id, Some(node));
assert_eq!((input.rows, input.cols), (42, 132));
assert_eq!(
input.terminal_interface,
Some(domain::DefaultAgentTerminalInterface::NativeTui)
);
assert!(!input.require_structured);
assert!(!input.fresh_start);
}
#[tokio::test]
async fn human_turn_ignores_old_terminal_and_accepts_replacement() {
let mailbox = Arc::new(InMemoryMailbox::new());
@ -1287,30 +1336,67 @@ mod self_restart_contract_tests {
assert_eq!(input.busy_state(key), domain::input::AgentBusyState::Idle);
assert_eq!(mailbox.pending(&key), 0);
}
#[tokio::test]
async fn cell_restart_stop_timeout_is_typed_and_never_reaches_the_next_step() {
let agent = AgentId::from_uuid(Uuid::from_u128(317));
let reached_after_stop = Arc::new(std::sync::atomic::AtomicBool::new(false));
let result: Result<(), AppError> = await_cell_restart_stop(
agent,
std::time::Duration::from_millis(1),
std::future::pending(),
)
.await;
if result.is_ok() {
reached_after_stop.store(true, std::sync::atomic::Ordering::SeqCst);
}
assert_eq!(result.unwrap_err().code(), "STOP_TIMEOUT");
assert!(!reached_after_stop.load(std::sync::atomic::Ordering::SeqCst));
}
}
impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
fn schedule(
impl AppAgentSelfRestarter {
fn schedule_restart(
&self,
project: &Project,
agent: AgentId,
requested_node: Option<domain::NodeId>,
reason: Option<String>,
) -> Result<AgentRestartScheduled, AppError> {
let key = domain::RuntimeAgentKey::new(project.id, agent);
let snapshots = self.live.live_agent_snapshots_for(project.id, &agent);
if snapshots.is_empty() {
return Err(AppError::Invalid(format!(
"requester agent {agent} is not live"
)));
return if requested_node.is_some() {
Err(AppError::RestartRuntimeNotFound { agent_id: agent })
} else {
Err(AppError::Invalid(format!(
"requester agent {agent} is not live"
)))
};
}
if let Some(node_id) = requested_node {
if snapshots.len() != 1 {
return Err(AppError::ConflictAgentVisibleElsewhere { agent_id: agent });
}
if snapshots[0].node_id != node_id {
return Err(AppError::AgentNotBoundToCell {
agent_id: agent,
node_id,
});
}
}
{
let mut active = self.active.lock().map_err(|_| {
AppError::Internal("agent restart registry lock poisoned".to_owned())
})?;
if !active.insert(key) {
return Err(AppError::Invalid(format!(
"a restart is already scheduled for agent {agent}"
)));
let message = format!("a restart is already scheduled for agent {agent}");
return if requested_node.is_some() {
Err(AppError::Conflict(message))
} else {
Err(AppError::Invalid(message))
};
}
}
@ -1346,29 +1432,36 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
};
let context = contexts.lock().ok().and_then(|map| map.get(&key).cloned());
let snapshot = snapshots[0];
let context = context.unwrap_or(ResumeContext {
let mut context = context.unwrap_or(ResumeContext {
project: project.clone(),
rows: 24,
cols: 80,
node_id: None,
node_id: Some(snapshot.node_id),
conversation_id: live
.structured
.meta_for_session(&snapshot.session_id)
.and_then(|(_, _, _, conversation)| conversation),
require_structured: snapshot.kind == application::LiveSessionKind::Structured,
});
if let Some(node_id) = requested_node {
context.node_id = Some(node_id);
}
publish_phase(AgentRestartPhase::Stopping);
if !delegated {
input.preempt(key);
}
let stop_result = application::terminal::stop_live_agent_runtime_in_project(
let stop = application::terminal::stop_live_agent_runtime_in_project(
&project,
agent,
live.as_ref(),
close_terminal.as_ref(),
)
.await;
);
let stop_result = if requested_node.is_some() {
await_cell_restart_stop(agent, std::time::Duration::from_secs(10), stop).await
} else {
stop.await
};
publish_phase(AgentRestartPhase::ReleasingInput);
if !delegated {
@ -1382,10 +1475,19 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
if delegated {
rendezvous.fail(key, &restart_id_task, error.to_string());
}
let code = match (&error, phase) {
(AppError::RestartStopTimeout { .. }, _) => "STOP_TIMEOUT",
(_, AgentRestartPhase::Stopping | AgentRestartPhase::ReleasingInput) => {
"STOP_FAILED"
}
(_, AgentRestartPhase::Launching) => "LAUNCH_FAILED",
(_, AgentRestartPhase::Resuming) => "RESUME_FAILED",
};
events.publish(DomainEvent::AgentRestartFailed {
agent_id: agent,
restart_id: restart_id_task.clone(),
phase,
code: code.to_owned(),
message: error.to_string(),
});
};
@ -1395,7 +1497,7 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
}
};
match stop_result {
Ok(stopped) if !stopped.is_empty() => {}
Ok(stopped) if !stopped.is_empty() || requested_node.is_some() => {}
Ok(_) => {
fail(
AgentRestartPhase::Stopping,
@ -1541,6 +1643,27 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
}
}
impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
fn schedule(
&self,
project: &Project,
agent: AgentId,
reason: Option<String>,
) -> Result<AgentRestartScheduled, AppError> {
self.schedule_restart(project, agent, None, reason)
}
fn schedule_cell(
&self,
project: &Project,
agent: AgentId,
node: domain::NodeId,
reason: Option<String>,
) -> Result<AgentRestartScheduled, AppError> {
self.schedule_restart(project, agent, Some(node), reason)
}
}
/// Guard for a direct human turn sent to a structured agent session.
///
/// Direct chat adapters (`agent_send`, web `chat.send`) still drive the provider
@ -2111,6 +2234,8 @@ pub struct BackendCore {
pub attach_live_agent: Arc<AttachLiveAgent>,
/// Tears down an already-running agent's live session by agent id (Lot D).
pub stop_live_agent: Arc<StopAgentRuntime>,
/// Atomically restarts the runtime bound to an exact visible agent cell.
pub restart_cell_agent_runtime: Arc<RestartCellAgentRuntime>,
/// Best-effort inspection of a conversation (last topic + token indicator)
/// for the resume popup (T7). Optional/extensible: backed by a `Vec` of
/// [`domain::ports::SessionInspector`]s; an empty/missing match yields empty
@ -3976,6 +4101,10 @@ impl BackendCore {
active: Arc::new(Mutex::new(HashSet::new())),
rendezvous: Arc::clone(&restart_rendezvous),
}) as Arc<dyn AgentSelfRestartScheduler>;
let restart_cell_agent_runtime = Arc::new(RestartCellAgentRuntime::new(
Arc::clone(&live_sessions),
Arc::clone(&self_restart),
));
// Drain du scheduler (§21.5-b) : à chaque réveil tiré, `TokioScheduler` pousse une
// `ScheduledTask::ResumeAgent` ; on l'exécute via le service (relance + AgentResumed).
// Patron `sweep_stalled` : spawner detache (pas `tokio::spawn` — `build`
@ -4333,6 +4462,7 @@ impl BackendCore {
rotate_conversation_log,
attach_live_agent,
stop_live_agent,
restart_cell_agent_runtime,
inspect_conversation,
project_store,
read_user_settings,

View File

@ -82,6 +82,8 @@ pub enum DomainEvent {
restart_id: String,
/// Phase that failed.
phase: AgentRestartPhase,
/// Stable machine-readable failure code.
code: String,
/// Human-readable failure safe to expose in the UI.
message: String,
},

View File

@ -100,6 +100,7 @@ use backend::dto::{
RemoveOpenProjectRestoreEntryRequestDto, RemoveProjectRequestDto, RemoveProjectResponseDto,
RenameLayoutRequestDto, ResolveAgentPermissionsRequestDto, ResolveAgentPermissionsResponseDto,
ResolveAgentSystemPermissionsRequestDto, ResolvedAgentSystemPermissionsDto,
RestartCellAgentRuntimeRequestDto, RestartCellAgentRuntimeResponseDto,
RestoreOpenProjectsOutputDto, ResumableAgentListDto, RetryOpenProjectRestoreRequestDto,
SaveEmbedderProfileRequestDto, SaveOpenCodeProviderProfileRequestDto,
SaveOpenProjectRestoreOrderRequestDto, SaveProfileRequestDto, SetActiveLayoutRequestDto,
@ -3442,6 +3443,9 @@ async fn invoke(
"list_resumable_agents" => invoke_list_resumable_agents(&request.args, &state.app).await,
"attach_live_agent" => invoke_attach_live_agent(&request.args, &state.app).await,
"stop_live_agent" => invoke_stop_live_agent(&request.args, &state.app).await,
"restart_cell_agent_runtime" => {
invoke_restart_cell_agent_runtime(&request.args, &state.app).await
}
"create_agent" => invoke_create_agent(&request.args, &state.app).await,
"change_agent_profile" => invoke_change_agent_profile(&request.args, &state.app).await,
"read_agent_context" => invoke_read_agent_context(&request.args, &state.app).await,
@ -4128,6 +4132,31 @@ async fn invoke_stop_live_agent(args: &Value, state: &BackendCore) -> Result<Val
serde_json::to_value(output).map_err(serialization_error)
}
async fn invoke_restart_cell_agent_runtime(
args: &Value,
state: &BackendCore,
) -> Result<Value, ErrorDto> {
let request =
required_request::<RestartCellAgentRuntimeRequestDto>("restart_cell_agent_runtime", args)?;
let project = resolve_project_readonly(&request.project_id, state).await?;
let output = state
.restart_cell_agent_runtime
.execute(application::RestartCellAgentRuntimeInput {
project,
node_id: parse_node_id(&request.node_id)?,
agent_id: parse_agent_id(&request.agent_id)?,
presentation: request.presentation.into(),
layout_id: request
.layout_id
.as_deref()
.map(parse_layout_id)
.transpose()?,
})
.map(RestartCellAgentRuntimeResponseDto::from)
.map_err(ErrorDto::from)?;
serde_json::to_value(output).map_err(serialization_error)
}
async fn invoke_create_agent(args: &Value, state: &BackendCore) -> Result<Value, ErrorDto> {
let request = required_request::<CreateAgentRequestDto>("create_agent", args)?;
let project = resolve_project_readonly(&request.project_id, state).await?;
@ -10748,6 +10777,7 @@ mod tests {
"list_resumable_agents",
"attach_live_agent",
"stop_live_agent",
"restart_cell_agent_runtime",
"create_agent",
"change_agent_profile",
"read_agent_context",