structured_llm | StructuredLlmClient, DefaultStructuredLlmClient | Yes | Type-safe JSON schema LLM output | structured-llm.md |
decomposer | TaskDecomposer<S> | Yes | Break objective into validated Subtasks | orchestration-pipeline.md |
designer | AgentDesigner<S>, CapabilityMatcher | Yes* | Match capabilities, spawn/inline decision, novel specs | orchestration-pipeline.md |
composer | PlanComposer<S>, ComposedPlan | Yes | Wire subtasks into validated plan DAG | orchestration-pipeline.md |
diagnoser | FailureDiagnoser<S> | Yes | Classify failures into Diagnosis categories | orchestration-pipeline.md |
recomposer | Recomposer<S>, RecompositionAction | Yes | Generate plan modification actions from diagnosis | orchestration-pipeline.md |
monitor | PlanMonitor<S>, MonitorConfig, MonitorEvent | Yes | Full lifecycle orchestration loop | orchestration-pipeline.md |
gradient | classify_failure, GradientAction | No | Deterministic G1-G9 gradient classification rules | orchestration-pipeline.md |
envelope_allocator | allocate_equal, allocate_weighted | No | Budget division across children | budget-and-context.md |
budget_warnings | evaluate_budget, BudgetWarning | No | Threshold-based budget warning emission | budget-and-context.md |
context_injector | inject_deterministic, inject_semantic | No | Context key selection for child scopes | budget-and-context.md |
context_summarizer | ContextSummarizer<S> | Yes | Compress large context values via LLM | budget-and-context.md |
delegation | DelegationProtocol<S>, DelegationMessage | Yes | Parent-to-child task delegation | protocols.md |
protocols | ClarificationProtocol<S>, EscalationProtocol<S> | Yes | Inter-agent Q&A and escalation | protocols.md |
conversions | Subtask, Diagnosis, SpawnDecision | No | LLM JSON -> validated SDK types at boundary | orchestration-pipeline.md |
reasoning | TraceEmitter, OrchestrationDecision, ReasoningStore | No | EATP-aligned reasoning trace emission for LLM decisions | reasoning-and-history.md |
history | ConversationHistory, HistoryConfig, TurnRole | Yes** | Sliding-window conversation compaction | reasoning-and-history.md |
error | OrchestrationError | No | Unified error type for all modules | (inline below) |
| Governance | | | | |
governance/accountability | AccountabilityTracker, AccountabilityRecord | No | D/T/R address → agent assignment tracking | governance.md |
governance/clearance | ClearanceEnforcer, ClearanceLevel | No | 5-level classification enforcement (monotonic raise) | governance.md |
governance/cascade | CascadeManager, CascadeEvent | No | Multi-level agent termination with budget reclamation | governance.md |
governance/bypass | BypassManager, BypassRecord, HeldAction | No | Emergency bypass (human approval required) | governance.md |
governance/vacancy | VacancyManager, VacancyEvent | No | Orphaned resource detection (idempotent scan) | governance.md |
governance/dereliction | DerelictionDetector, DerelictionWarning | No | Duty monitoring with severity escalation | governance.md |
governance/budget | GovernanceBudgetTracker, BudgetSnapshot | No | Unified budget view (CAS-protected consumption) | governance.md |
| Audit | | | | |
audit/trail | AuditTrail, AuditRecord, AuditEventType | No | Append-only audit chain with reasoning trace links | governance.md |
| Bridges | | | | |
scope_bridge | GovernanceSnapshot, project_for_child | No | Context projection + anti-amnesia injection | governance.md |
message_transport | MessageTransport, TransportEnvelope | No | Protocol → L3 MessageRouter bridge | governance.md |
agent_lifecycle | AgentLifecycleManager, AgentRecord | No | PlanMonitor → L3 Factory coordinator | governance.md |
| Tool Hydration | | | | |
hydration | ToolHydrator, DefaultToolHydrator, ToolHydratorConfig | No | Progressive tool disclosure via TF-IDF search | hydration-streaming.md |
hydration/meta_tool | create_search_tools_meta_tool() | No | Auto-registered search_tools meta-tool | hydration-streaming.md |
| Caller Event Streaming | | | | |
streaming/caller_event | CallerEvent, CallerEventWire, TaodResultWire | No | Event stream + serializable wire types (all 7 bindings) | hydration-streaming.md |
streaming/agent | StreamingAgent::run_stream() | No | Single-shot token streaming via CallerEvent | hydration-streaming.md |
agent_engine/concrete | Agent::chat_stream(), push_assistant_turn() | No | Conversational streaming via CallerEvent | hydration-streaming.md |
agent_engine/taod | TaodRunner::run_stream() | No | TAOD loop with tool lifecycle events | hydration-streaming.md |
| Entry Point | | | | |
supervisor | GovernedSupervisor, SupervisorResult | No | Progressive-disclosure governance entry point | governance.md |