feat: read agent_id from ctx (SDK now plumbs it)

Plexum-sdk-go now propagates the caller agent id via
`_meta.agent_id` on tools/call. AgentIDFromCtx prefers
plugin.AgentIDFromContext(ctx); falls back to the
single-active-calendar-slot heuristic for host-driven dispatch
paths (channel manager, CLI plugin-call) that lack ctx.

Drops bestEffortAgentID — the inline closure does the same thing
without the dead-Slot-iterate noise.

cmd/plexum-harborforge-plugin/main.go

@@ -131,15 +131,16 @@ func (p *harborForgePlugin) Init(ctx context.Context, host sdkplugin.HostAPI) er
 		Scheduler: p.sched,
 		Host:      host,
 		AgentIDFromCtx: func(ctx context.Context) string {
-			// Plexum stashes the calling agent id on the host-side
-			// context (via WithAgent) before dispatching tool calls.
-			// We can't directly import internal/agentloop from a
-			// plugin, so we rely on PLEXUM_TOOL_AGENT_ID env-style
-			// (set per-call by host when we add that wiring) or fall
-			// back to the only-active-agent heuristic. v1: prefer the
-			// only-active wake-target (deterministic in single-agent
-			// HF deployments).
-			return p.bestEffortAgentID()
+			// Host attaches the caller agent id via tools/call
+			// `_meta.agent_id`; SDK unpacks it into ctx.
+			if id := sdkplugin.AgentIDFromContext(ctx); id != "" {
+				return id
+			}
+			// Fallback for host paths that don't carry an agent
+			// (channel-driven, CLI plugin-call). When a single
+			// calendar slot is active we can deterministically
+			// attribute the call to that slot's owner.
+			return p.sched.SingleActiveAgentID()
 		},
 	}
 
@@ -209,28 +210,6 @@ func mapStateToCalendar(s string) calendar.AgentStatusValue {
 	return calendar.AgentStatusOffline
 }
 
-// bestEffortAgentID is a v1 stop-gap for tools that need the calling
-// agent's id but don't have it on the ctx (Plexum SDK doesn't yet
-// expose this — TODO upstream). v1: if exactly one agent has an
-// active calendar slot we return it; otherwise empty. The calendar
-// tools (the only ones that need agent context) usually fire when
-// exactly one slot is active.
-func (p *harborForgePlugin) bestEffortAgentID() string {
-	sch := p.sched.Status()
-	if len(sch.Active) != 1 {
-		return ""
-	}
-	// We don't track AgentID on Slot directly — the scheduler keeps
-	// activeByAgentID. Iterate to find the one.
-	for _, a := range sch.Active {
-		// Slot is shared between agents only via the scheduler's maps;
-		// here we have just the Slot struct without owner.
-		_ = a
-	}
-	// Fallback to scheduler's helper:
-	return p.sched.SingleActiveAgentID()
-}
-
 func manifestFromDisk() sdkplugin.Manifest {
 	// Bundled manifest.json is the authoritative shape; the binary
 	// version reads it next to itself to avoid hand-syncing two