TypeScript agent orchestration entrance (createAgent, createTeam) — durable memory/HITL/stream replay when you need them. Production-oriented architecture, young ecosystem — not a graph DSL.

**Plugin load/activation contracts — extend a host without forking core packages.**

**Planning engine contracts — turn objectives into executable plans for Runtime.**

**Logs, metrics, traces, diagnostics, and health contracts for platform hosts.**

**Select which implementation fulfills a capability binding — provider-neutral resolution.**

**Evaluation framework — score and compare agent runs with replaceable evaluators.**

**Audit & compliance fact ingestion contracts — immutable audit records for hosts.**

**Persistence ports — repositories and transactions without locking you to a database vendor.**

**Configuration contracts — typed config loading for replaceable host settings.**

**Platform event bus contracts for operational events between components.**

**Assemble execution context packages (memory, knowledge, constraints) for a turn.**

**Prompt package construction contracts — assemble model-facing prompts without owning AI I/O.**

**Composition root ownership — instantiate and wire platform dependencies without scattering `new` across the app.**

**Human-in-the-loop interaction contracts — approvals and human tasks for hosts.**

**CLI framework contracts for operator and developer command-line surfaces.**

**Plugin marketplace / distribution contracts — discovery and integrity facts (not code execution).**

**Scheduler & background job contracts for deferred and recurring host work.**

**Platform governance & evolution contracts — versioning, compatibility, governance facts.**