Context
Derived from the 2026-04-25 and 2026-04-28 Claude Code usage review.
Claude Code agent drivers had to manage inFlight guards, self-dispatch event handling, and shadow/driver coordination around dispatch / subscribe / on surfaces.
Current assessment
SPEC status: not currently provided as core runtime semantics.
The current runtime event model exposes causally complete events with correlation data such as intentId. It does not automatically suppress self-dispatch events, and it should not hide causality at the runtime layer.
Desired outcome
Design an agent-oriented helper layer above SDK dispatch events, potentially as an ecosystem package such as @manifesto-ai/agent.
This helper may provide source tags, ignore lists, driver utilities, in-flight handling, and safer patterns for AI-agent loops while preserving the underlying runtime telemetry.
Acceptance criteria
- Helper design is layered above SDK/Core semantics and does not change runtime causality.
- Self-dispatch suppression, if provided, is opt-in helper behavior rather than a default runtime behavior.
- Driver examples use event payloads and fresh runtime snapshots instead of React-mediated stale state.
- Non-goal is explicit: do not merge state events and telemetry events for agent convenience.
Context
Derived from the 2026-04-25 and 2026-04-28 Claude Code usage review.
Claude Code agent drivers had to manage
inFlightguards, self-dispatch event handling, and shadow/driver coordination arounddispatch/subscribe/onsurfaces.Current assessment
SPEC status: not currently provided as core runtime semantics.
The current runtime event model exposes causally complete events with correlation data such as
intentId. It does not automatically suppress self-dispatch events, and it should not hide causality at the runtime layer.Desired outcome
Design an agent-oriented helper layer above SDK dispatch events, potentially as an ecosystem package such as
@manifesto-ai/agent.This helper may provide source tags, ignore lists, driver utilities, in-flight handling, and safer patterns for AI-agent loops while preserving the underlying runtime telemetry.
Acceptance criteria