Skip to content

Execution tuning (removed from Agent manifests)

spec.execution on kind: Agent and overlay spec.patch.execution are not supported. Engine queue depth, cache policy, streaming, and parallel tool dispatch are configured only through kind: RuntimeEngine manifests resolved like infra: workspace runtime_refs, $XDG_CONFIG_HOME/mas/runtime/, or CLI --runtime-ref.

See runtime-engine.md and user-config.md.

Migration (removed agent fields)

Old (invalid on kind: Agent / overlay patch) Replacement
spec.execution (cache, stream, queue depth, …) kind: RuntimeEngine via runtime_refs / --runtime-ref
Offline LLM turns llm_cache replay recorded against a live provider (raise_on_miss)
spec.infra_refs / spec.runtime_refs on Agent, MAS, or overlay config.yaml, --infra-ref / --runtime-ref
spec.infra_interceptors on Agent workspace / CLI infra resolution (same as infra refs)

Not the same: experiment.execution in mas-lab benchmark YAML is the batch orchestration block (parallel scenarios, timeouts, trace-cache emulation). It does not replace RuntimeEngine and is documented in experiment.md.

mas-ctl validate and separation checks fail closed on forbidden keys so outdated examples surface immediately.