What problem does it solve?
It provides the standing runtime policy that turns loaded natural-language harness skills into an explicit executable graph of agent calls, ensuring the orchestrator launches delegated child work instead of doing substantive task work itself.
Core Features & Use Cases
- Delegated agent execution boundary: Keeps the parent runtime as an interpreter that schedules, routes, and closes runs while child agents perform concrete task work.
- Deterministic call-graph reconstruction: Rebuilds intended roles, stages, termination conditions, and context semantics from the loaded skill text before executing.
- Context management discipline: Controls when children inherit shared dialogue via
fork_context versus starting fresh with minimal self-contained task packets.
- Staged evidence and artifact discipline: Establishes rules for where intermediate state must live (e.g., under a runtime STATE_ROOT) and how to keep stage boundaries inspectable.
Quick Start
Load this runtime charter alongside your target harness skill, then run the resulting harness so the runtime interpreter launches the correct child agent stages and produces the contract deliverable.
