Struct ProdWorkerFacade 

pub struct ProdWorkerFacade {
    workspace: Weak<Workspace>,
}

Production impl. Holds a Weak<Workspace> so construction doesn’t close a strong cycle through the Workspace -> bridge -> MCP -> Tool -> facade -> Workspace path. Every method starts with upgrade() and short-circuits when the workspace has been dropped (only possible during shutdown).

Fields

workspace: Weak<Workspace>

Implementations

impl ProdWorkerFacade

pub fn from_arc(workspace: &Arc<Workspace>) -> Arc<dyn WorkerFacade>

Trait Implementations

impl WorkerFacade for ProdWorkerFacade

fn caller_project(&self, caller: &SessionSlot) -> Option<CallerProject>

Resolve the caller’s project key + lead/worker flag. Returns None when caller matches no known session.

fn spawn_worker<'life0, 'life1, 'async_trait>( &'life0 self, caller: &'life1 SessionSlot, label: String, charter: String, kick: Option<String>, resume_kick: Option<String>, interactive: bool, resume_session: bool, mcp_families: Option<Vec<String>>, cwd: Option<String>, ) -> Pin<Box<dyn Future<Output = Result<WorkerSpawnReply, WorkerSpawnError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Dispatch a Command::SpawnWorker and await its synchronous reply. Gating (lead-only, non-empty label, non-empty charter) happens before dispatch. resume_kick is persisted rather than delivered now; it replaces the generic restart note when this worker resumes after a forge restart with a turn that was cut off. A worker that was idle when forge stopped resumes with nothing delivered. interactive keeps the built-in AskUserQuestion tool, which every other worker is denied. resume_session resumes the label’s most recent prior session instead of starting fresh. A label with none resolves to ResumeTarget::None, which starts a fresh session and says so in the reply’s SessionChoice; only a lookup that failed is an error.

fn update_worker( &self, caller: &SessionSlot, label: &str, charter: Option<String>, kick: Option<String>, resume_kick: Option<String>, mcp_families: Option<Vec<String>>, ) -> Result<(), WorkerUpdateError>

Merge the supplied fields onto the stored dynamic-worker row for label in the caller’s project, leaving a None field at its stored value. Errors with NoSuchWorker when no row exists: update revises an existing worker, it never creates one. Takes effect on that worker’s next respawn, since a session’s system prompt and tool list are fixed when the session spawns. mcp_families follows the same contract: None leaves the stored selection alone, Some(names) replaces it - canonicalised by the tool - and Some(empty) resets it to every family.

fn despawn_worker<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, caller: &'life1 SessionSlot, label: &'life2 str, force: bool, ) -> Pin<Box<dyn Future<Output = Result<DespawnOutcome, WorkerDespawnError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Dispatch a Command::DespawnWorker and await its result. Gating (lead-only, non-empty label) happens before dispatch. force removes a dirty worktree; without it a dirty worktree blocks the despawn (DespawnOutcome::Blocked).

fn list_workers(&self, caller: &SessionSlot) -> Vec<WorkerStatus>

Snapshot of every worker in the caller’s project. Returns an empty Vec when the caller resolves to no project.

fn list_all_workers(&self) -> Vec<WorkerStatus>

Snapshot of every project’s workers: the real workspace walks list_projects in config order, the mock sorts its project keys.

fn capacity(&self, caller: &SessionSlot) -> Option<WorkerCapacity>

The caller’s project worker capacity: the cap the spawn path enforces (max_workers in the project’s [[orgs.projects]] entry, else DEFAULT_MAX_WORKERS_PER_PROJECT) paired with the live count. A plain data read - the same one the spawn path’s cap check performs. Returns None when the caller resolves to no project.

fn deliver_worker_prompt<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, caller: &'life1 SessionSlot, target_label: &'life2 str, wrapped: WrappedPrompt, ) -> Pin<Box<dyn Future<Output = Result<WorkerTargetStatus, WorkerDeliverError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Dispatch a worker-bound wrapped prompt. Delivered means the prompt was accepted - handed to the target’s session queue, held for its turn end, or parked for its connect - which is the whole claim the sender’s ack makes; anything that reached no session is an error. UnknownLabel when no live worker matches.

fn deliver_worker_prompt_to_project<'life0, 'life1, 'life2, 'life3, 'life4, 'async_trait>( &'life0 self, caller: &'life1 SessionSlot, org: &'life2 str, project: &'life3 str, target_label: &'life4 str, wrapped: WrappedPrompt, ) -> Pin<Box<dyn Future<Output = Result<WorkerTargetStatus, WorkerDeliverError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait, 'life4: 'async_trait,

Dispatch a worker-bound wrapped prompt to a project named by its (org, project) pair rather than the caller’s own. The caller lives in another project, so the target cannot be resolved from it. Mirrors deliver_worker_prompt otherwise: an unknown label is an error and a sleeping target is not auto-spawned.

fn deliver_prompt_to_lead<'life0, 'life1, 'async_trait>( &'life0 self, caller: &'life1 SessionSlot, wrapped: WrappedPrompt, ) -> Pin<Box<dyn Future<Output = Result<WorkerTargetStatus, WorkerLeadDeliverError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Dispatch a wrapped prompt from a worker back to its lead. Caller MUST be a worker. The target lead is resolved from the caller’s WorkerEntry::spawned_by_session_id. Same wire-shape contract as deliver_worker_prompt: Delivered means the lead’s session accepted it, and one the session could not take is an error.