Struct OrchestrationRuntime 

pub struct OrchestrationRuntime { /* private fields */ }

Session-scoped runtime enforcement shared by every subagent tool path.

Implementations

impl OrchestrationRuntime

pub fn trusted_catalog( foreground: &QualifiedModelId, _manifest_references: impl IntoIterator<Item = QualifiedModelId>, ) -> Result<CatalogSnapshot, &'static str>

Build a snapshot exclusively from runtime-owned routing/catalog descriptors. manifest_references are deliberately ignored here: manifests may narrow the snapshot, but can never create catalog membership. The foreground is admitted only after the credential-free runtime router resolves its exact identity.

pub fn from_limits( foreground: QualifiedModelId, limits: SubagentLimits, ) -> Result<OrchestrationRuntime, &'static str>

The manifestless baseline sized by the user’s subagent.* limits. Every production install without a prompt-manifest policy uses this.

pub fn baseline( foreground: QualifiedModelId, concurrent: usize, total: usize, ) -> Result<OrchestrationRuntime, &'static str>

Secure manifestless baseline: the exact foreground identity is the only worker choice in a deterministic runtime-controlled catalog.

pub fn new(policy: DelegationPolicy) -> OrchestrationRuntime

pub fn with_max_depth( self, max_depth: u16, ) -> Result<OrchestrationRuntime, &'static str>

How deep workers may nest (subagent.max_depth): 1 = only the foreground delegates. Zero would forbid all delegation and is refused.

pub fn reserve_delegation( &self, worker_id: &str, parent: Option<&str>, ) -> Result<u16, DelegationTreeDenied>

Reserve one tree edge before allocating channels/threads/provider runtimes. parent=None is a foreground-root child. Only nesting depth is bounded here: fan-out and totals are the delegation policy’s limits, checked right after in the same spawn path. Every denial is fail-closed and leaves the tree unchanged.

pub fn release_delegation(&self, worker_id: &str, _parent: Option<&str>)

Release one reserved tree edge. Descendants are recursively removed because a live child cannot outlive its parent lineage. The caller- supplied parent is intentionally ignored: ownership is taken from the reservation recorded under the tree lock, so a stale or hostile caller cannot decrement another parent’s fanout.

pub fn delegation_descendants(&self) -> usize

pub fn preflight(&self, model: &str) -> Result<(), String>

pub fn grant_worker_model(&self, model: &str) -> Result<(), String>

Honors an explicit mid-session user trust grant for one exact worker model. Trusted models were never meant to be pinned at session start: the live policy and catalog are extended in place while worker lifecycle state and concurrency limits are preserved. The grant takes effect for the next dispatch decision.

pub fn ultracode_readiness(&self, model: &str) -> Result<(usize, usize), String>

Read-only UltraCode authorization snapshot. The exact foreground identity, policy authorization, and configured limits are checked under one lock; no worker reservation or lifecycle state is created.

pub fn resolve_and_authorize( &self, runtime_handle: &str, requested: Option<&str>, ) -> Result<AuthorizedWorkerModel, DispatchDenial>

Single parse/resolve/catalog/policy/limits decision point used by every public spawn path. No credential or provider object is reachable here.

pub fn mark_starting(&self, runtime_handle: &str) -> Result<(), String>

Marks the persisted handle Starting immediately before the runtime/thread factory is invoked. Starting remains rollback-safe until mark_running.

pub fn mark_running(&self, runtime_handle: &str) -> Result<(), String>

pub fn rollback(&self, runtime_handle: &str)

pub fn effective_choices(&self) -> Vec<String>

pub fn foreground_model(&self) -> String

pub fn authorize_with_policy( &self, runtime_handle: &str, model: &str, role: WorkerRole, writes: WorkerWritePolicy, ) -> Result<(), String>

pub fn authorize(&self, runtime_handle: &str, model: &str) -> Result<(), String>

pub fn poll(&self, id: &str, fingerprint: &str) -> Result<(), String>

pub fn steer(&self, id: &str) -> Result<(), String>

pub fn terminal_and_collect( &self, id: &str, terminal: WorkerTerminal, ) -> Result<(), String>

pub fn finish_one_shot( &self, id: &str, terminal: WorkerTerminal, ) -> Result<(), String>

pub fn reconcile(&self, id: &str) -> Result<(), String>

pub fn retire_reconciled(&self, removed_handles: &HashSet<String>)

Reaper handoff: use exact IDs removed under the runtime registry lock, never a complement snapshot that could include newly registered workers.

pub fn unreconciled_runtime_handles(&self) -> HashSet<String>

Snapshot runtime handle IDs still named by the completion gate. Taking this snapshot before locking the subagent registry avoids cross-registry lock order.

pub fn is_unreconciled(&self, runtime_handle: &str) -> bool

Whether a runtime handle is still named by the policy completion gate.

pub fn completion_gate(&self) -> CompletionGate

pub fn check_foreground_write(&self, path: &str) -> ScopeDecision

pub fn telemetry_json(&self) -> String

pub fn enforcement_mode(&self) -> EnforcementMode