Nonblocking post connect capability broadcast

SKILL: Nonblocking Post-Connect Capability Broadcast

Domain: bridge-lifecycle Trigger: Apply when a transport has reached its ready or connected state, but optional capability discovery and hello metadata still need to be broadcast to the remote peer. Source Pattern: Distilled from reviewed bridge-lifecycle, capability-broadcast, and reconnect-flow implementations.

Core Method

Treat connection visibility and capability advertisement as two separate phases. In the state-change handler, synchronously commit the host's connected/session-active flags first so the UI and downstream effects can react immediately. Only after that state commit should you launch a detached async task that resolves optional capabilities, re-checks cancellation or liveness, snapshots any current state needed for the payload, and sends the post-connect broadcast. Keep the late path best-effort: if capability loading or broadcast assembly fails, log it locally and leave the connection marked ready.

Key Rules

• Commit the connected or session-active state before awaiting any optional capability load; readiness must not depend on slow skill or metadata discovery.
• Run the post-connect broadcast in fire-and-forget async work guarded by a local try/catch so failures do not bubble back into the connection transition.
• Re-check cancellation or handle liveness after the awaited capability work and before writing to the transport; the peer may already be gone.
• Read mutable payload state as late as practical inside the async task so the broadcast reflects the current permission mode, agents, or feature flags rather than a stale pre-await snapshot.
• Keep the broadcast path optional and feature-gated when the metadata is additive rather than required for connection correctness.

Example Application

When a mobile or web client connects to a background bridge, flip the app state to connected immediately so forwarding and UI badges start right away. Then, in a detached async block, load the bridge-safe skill list, collect the latest runtime metadata, and send one system/init message. If that late load fails, record the error, but do not bounce the session back out of connected.

Anti-Patterns (What NOT to do)

• Do not await capability discovery before setting the connected state; that turns optional metadata into user-visible connection latency.
• Do not clear or roll back the connected flag just because the late capability broadcast fails; the transport can still be healthy.
• Do not capture all payload state before the async work and reuse it blindly after awaits; you risk broadcasting stale session metadata.