feat(channels): add sidecar delivery behavior overrides

Closes #144.

- Retire user-facing Tool Status Messages and deterministic tool-status channel text.
- Add sidecar-generated Quick Acknowledgement and Intermediate Replies with provider/model/timeout/token/char caps.
- Resolve delivery behavior as Channel > Agent > Workspace with agent overrides in other_config.delivery_behavior.
- Preserve legacy block_reply defaults for Intermediate Replies.

cmd/gateway_consumer.go

@@ -15,6 +15,7 @@ import (
 	"github.com/nextlevelbuilder/goclaw/internal/bus"
 	"github.com/nextlevelbuilder/goclaw/internal/channels"
 	"github.com/nextlevelbuilder/goclaw/internal/config"
+	"github.com/nextlevelbuilder/goclaw/internal/providers"
 	"github.com/nextlevelbuilder/goclaw/internal/scheduler"
 	"github.com/nextlevelbuilder/goclaw/internal/store"
 	"github.com/nextlevelbuilder/goclaw/internal/tools"
@@ -26,7 +27,7 @@ import (
 // and routes them through the scheduler/agent loop, then publishes the response back.
 // Also handles subagent announcements: routes them through the parent agent's session
 // (matching TS subagent-announce.ts pattern) so the agent can reformulate for the user.
-func consumeInboundMessages(ctx context.Context, msgBus *bus.MessageBus, agents *agent.Router, cfg *config.Config, sched *scheduler.Scheduler, channelMgr *channels.Manager, teamStore store.TeamStore, quotaChecker *channels.QuotaChecker, sessStore store.SessionStore, agentStore store.AgentStore, contactCollector *store.ContactCollector, postTurn tools.PostTurnProcessor, subagentMgr *tools.SubagentManager, usageCapSvc *usagecaps.Service) {
+func consumeInboundMessages(ctx context.Context, msgBus *bus.MessageBus, agents *agent.Router, cfg *config.Config, sched *scheduler.Scheduler, channelMgr *channels.Manager, teamStore store.TeamStore, quotaChecker *channels.QuotaChecker, sessStore store.SessionStore, agentStore store.AgentStore, contactCollector *store.ContactCollector, postTurn tools.PostTurnProcessor, subagentMgr *tools.SubagentManager, usageCapSvc *usagecaps.Service, providerReg *providers.Registry) {
 	slog.Info("inbound message consumer started")
 
 	// Inbound message deduplication (matching TS src/infra/dedupe.ts + inbound-dedupe.ts).
@@ -59,6 +60,7 @@ func consumeInboundMessages(ctx context.Context, msgBus *bus.MessageBus, agents
 		ContactCollector: contactCollector,
 		SubagentMgr:      subagentMgr,
 		UsageCaps:        usageCapSvc,
+		ProviderReg:      providerReg,
 		GetAnnounceMu:    getAnnounceMu,
 	}
 

cmd/gateway_consumer_deps.go

@@ -7,6 +7,7 @@ import (
 	"github.com/nextlevelbuilder/goclaw/internal/bus"
 	"github.com/nextlevelbuilder/goclaw/internal/channels"
 	"github.com/nextlevelbuilder/goclaw/internal/config"
+	"github.com/nextlevelbuilder/goclaw/internal/providers"
 	"github.com/nextlevelbuilder/goclaw/internal/scheduler"
 	"github.com/nextlevelbuilder/goclaw/internal/store"
 	"github.com/nextlevelbuilder/goclaw/internal/tools"
@@ -30,6 +31,7 @@ type ConsumerDeps struct {
 	TaskRunSessions  sync.Map
 	SubagentMgr      *tools.SubagentManager
 	UsageCaps        *usagecaps.Service
+	ProviderReg      *providers.Registry
 	BgWg             sync.WaitGroup
 	GetAnnounceMu    func(string) *sync.Mutex
 }

cmd/gateway_consumer_normal.go

@@ -242,14 +242,17 @@ func processNormalMessage(
 	if lk := msg.Metadata["local_key"]; lk != "" {
 		chatIDForRun = lk
 	}
-	blockReply := deps.ChannelMgr != nil && deps.ChannelMgr.ResolveBlockReply(msg.Channel, deps.Cfg.Gateway.BlockReply)
 	chatBehavior := channels.ResolvedChatBehavior{}
 	if deps.ChannelMgr != nil {
-		chatBehavior = deps.ChannelMgr.ResolveChatBehavior(msg.Channel, deps.Cfg.Gateway.ChatBehavior)
+		workspaceBehavior := channels.ChatBehaviorConfigWithIntermediateDefault(deps.Cfg.Gateway.ChatBehavior, deps.Cfg.Gateway.BlockReply)
+		agentBehavior := channels.ParseAgentDeliveryBehaviorConfig(agentLoop.OtherConfig())
+		chatBehavior = deps.ChannelMgr.ResolveChatBehaviorWithAgent(msg.Channel, workspaceBehavior, agentBehavior)
 	}
+	blockReply := deps.ChannelMgr != nil && chatBehavior.IntermediateReplies.Enabled
+	deliveryRuntime := buildDeliveryRuntime(ctx, deps, agentLoop, chatBehavior, msg, peerKind, resolveChannelType(deps.ChannelMgr, msg.Channel), agentID)
 	toolStatus := deps.Cfg.Gateway.ToolStatus == nil || *deps.Cfg.Gateway.ToolStatus // default true
 	if deps.ChannelMgr != nil {
-		deps.ChannelMgr.RegisterRunWithBehavior(runID, msg.Channel, chatIDForRun, messageID, outMeta, msg.TenantID, channelStream, blockReply, toolStatus, chatBehavior, reasoningDelivery)
+		deps.ChannelMgr.RegisterRunWithDelivery(runID, msg.Channel, chatIDForRun, messageID, outMeta, msg.TenantID, channelStream, blockReply, toolStatus, chatBehavior, deliveryRuntime, reasoningDelivery)
 	}
 
 	// Group-aware system prompt: help the LLM adapt tone and behavior for group chats.
@@ -608,6 +611,60 @@ func processNormalMessage(
 	}(agentID, msg.Channel, msg.ChatID, sessionKey, runID, peerKind, msg.Content, outMeta, blockReply, chatBehavior, channelStream, ptd, msg.TenantID, agentLoop.UUID(), agentLoop.OtherConfig())
 }
 
+func buildDeliveryRuntime(ctx context.Context, deps *ConsumerDeps, agentLoop agent.Agent, behavior channels.ResolvedChatBehavior, msg bus.InboundMessage, peerKind, channelType, agentKey string) channels.DeliveryRuntime {
+	locale := msg.Metadata["locale"]
+	if locale == "" {
+		locale = "auto"
+	}
+	runtime := channels.DeliveryRuntime{
+		Locale:    locale,
+		Inbound:   msg.Content,
+		PeerKind:  peerKind,
+		Channel:   channelType,
+		AgentName: agentKey,
+	}
+	if behavior.Enabled && behavior.QuickAck.Enabled {
+		switch behavior.QuickAck.Mode {
+		case channels.QuickAckModeLLMGenerated, channels.QuickAckModeSidecar, "":
+			runtime.QuickAckGenerator = buildDeliveryGenerator(ctx, deps, agentLoop, msg.TenantID, behavior.QuickAck.Provider, behavior.QuickAck.Model)
+		}
+	}
+	if behavior.Enabled && behavior.IntermediateReplies.Enabled {
+		switch behavior.IntermediateReplies.Mode {
+		case channels.IntermediateModeSidecar, channels.QuickAckModeLLMGenerated, "":
+			runtime.ProgressGenerator = buildDeliveryGenerator(ctx, deps, agentLoop, msg.TenantID, behavior.IntermediateReplies.Provider, behavior.IntermediateReplies.Model)
+		}
+	}
+	return runtime
+}
+
+func buildDeliveryGenerator(ctx context.Context, deps *ConsumerDeps, agentLoop agent.Agent, tenantID uuid.UUID, providerName, model string) channels.DeliveryMessageGenerator {
+	provider := agentLoop.Provider()
+	resolvedProviderName := agentLoop.ProviderName()
+	if providerName != "" && deps.ProviderReg != nil {
+		if p, err := deps.ProviderReg.Get(ctx, providerName); err == nil {
+			provider = p
+			resolvedProviderName = providerName
+		} else {
+			slog.Warn("channel delivery: configured provider unavailable, falling back to agent provider", "provider", providerName, "error", err)
+		}
+	}
+	if provider == nil {
+		return nil
+	}
+	if model == "" {
+		model = agentLoop.Model()
+	}
+	return channels.ProviderDeliveryMessageGenerator{
+		Provider:     provider,
+		ProviderName: resolvedProviderName,
+		Model:        model,
+		UsageCaps:    deps.UsageCaps,
+		TenantID:     tenantID,
+		AgentID:      agentLoop.UUID(),
+	}
+}
+
 // isSafeBitrixEntityToken validates a webhook-sourced Bitrix entity token before
 // it is interpolated into the agent system prompt. Rejects empty, oversized, or
 // control-character payloads to prevent prompt-injection from a crafted portal

cmd/gateway_lifecycle.go

@@ -147,7 +147,7 @@ func (d *gatewayDeps) runLifecycle(
 		d.channelMgr.SetContactCollector(contactCollector)
 	}
 
-	go consumeInboundMessages(ctx, d.msgBus, d.agentRouter, d.cfg, deps.sched, d.channelMgr, deps.consumerTeamStore, deps.quotaChecker, d.pgStores.Sessions, d.pgStores.Agents, contactCollector, deps.postTurn, deps.subagentMgr, d.usageCapSvc)
+	go consumeInboundMessages(ctx, d.msgBus, d.agentRouter, d.cfg, deps.sched, d.channelMgr, deps.consumerTeamStore, deps.quotaChecker, d.pgStores.Sessions, d.pgStores.Agents, contactCollector, deps.postTurn, deps.subagentMgr, d.usageCapSvc, d.providerRegistry)
 
 	// Webhook callback worker — delivers async webhook_calls rows to receiver callback_url.
 	// Runs in both editions: Standard (PG, concurrency=4) and Lite (SQLite, concurrency=1).