Daemon heartbeat becomes recovery-focused (gt-vdprb.4)

Change daemon from wake-focused to recovery-focused:

Before: Daemon pokes agents every 5-60min as primary wake
After: Daemon only checks for edge cases that feed-wake cannot handle

Recovery checks:
- Dead sessions that need restart (ensureDeaconRunning, ensureWitnessesRunning)
- Stale agents that crashed without updating state (checkStaleAgents)
- GUPP violations: agents with work-on-hook not progressing (checkGUPPViolations)
- Orphaned work: work assigned to dead agents (checkOrphanedWork)

Removed:
- pokeDeacon() - no longer sending HEARTBEAT messages
- pokeWitness()/pokeWitnesses() - no longer sending HEARTBEAT messages
- MOTD message arrays - only used by poke functions

Normal agent wake is now handled by feed subscription (bd activity --follow).
The daemon is the safety net for edge cases, not the primary propulsion.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

internal/daemon/daemon.go

@@ -20,15 +20,15 @@ import (
 )
 
 // Daemon is the town-level background service.
-// It ensures patrol agents (Deacon, Witnesses) are running and sends periodic heartbeats.
-// All health checking, nudging, and decision-making belongs in the patrol molecules.
+// It ensures patrol agents (Deacon, Witnesses) are running and detects failures.
+// This is recovery-focused: normal wake is handled by feed subscription (bd activity --follow).
+// The daemon is the safety net for dead sessions, GUPP violations, and orphaned work.
 type Daemon struct {
-	config        *Config
-	tmux          *tmux.Tmux
-	logger        *log.Logger
-	ctx           context.Context
-	cancel        context.CancelFunc
-	lastMOTDIndex int // tracks last MOTD to avoid consecutive repeats
+	config *Config
+	tmux   *tmux.Tmux
+	logger *log.Logger
+	ctx    context.Context
+	cancel context.CancelFunc
 }
 
 // New creates a new daemon instance.
@@ -169,35 +169,39 @@ func (d *Daemon) calculateHeartbeatInterval() time.Duration {
 }
 
 // heartbeat performs one heartbeat cycle.
-// The daemon ensures patrol agents are running and sends heartbeats.
-// All health checking and decision-making belongs in the patrol molecules.
+// The daemon is recovery-focused: it ensures agents are running and detects failures.
+// Normal wake is handled by feed subscription (bd activity --follow).
+// The daemon is the safety net for edge cases:
+// - Dead sessions that need restart
+// - Agents with work-on-hook not progressing (GUPP violation)
+// - Orphaned work (assigned to dead agents)
 func (d *Daemon) heartbeat(state *State) {
-	d.logger.Println("Heartbeat starting")
+	d.logger.Println("Heartbeat starting (recovery-focused)")
 
-	// 1. Ensure Deacon is running (process management)
+	// 1. Ensure Deacon is running (restart if dead)
 	d.ensureDeaconRunning()
 
-	// 2. Send heartbeat to Deacon (simple notification, no decision-making)
-	d.pokeDeacon()
-
-	// 3. Ensure Witnesses are running for all rigs
+	// 2. Ensure Witnesses are running for all rigs (restart if dead)
 	d.ensureWitnessesRunning()
 
-	// 4. Send heartbeats to Witnesses
-	d.pokeWitnesses()
-
-	// 5. Trigger pending polecat spawns (bootstrap mode - ZFC violation acceptable)
+	// 3. Trigger pending polecat spawns (bootstrap mode - ZFC violation acceptable)
 	// This ensures polecats get nudged even when Deacon isn't in a patrol cycle.
 	// Uses regex-based WaitForClaudeReady, which is acceptable for daemon bootstrap.
 	d.triggerPendingSpawns()
 
-	// 6. Process lifecycle requests
+	// 4. Process lifecycle requests
 	d.processLifecycleRequests()
 
-	// 7. Check for stale agents (timeout fallback)
+	// 5. Check for stale agents (timeout fallback)
 	// Agents that report "running" but haven't updated in too long are marked dead
 	d.checkStaleAgents()
 
+	// 6. Check for GUPP violations (agents with work-on-hook not progressing)
+	d.checkGUPPViolations()
+
+	// 7. Check for orphaned work (assigned to dead agents)
+	d.checkOrphanedWork()
+
 	// Update state
 	state.LastHeartbeat = time.Now()
 	state.HeartbeatCount++
@@ -214,38 +218,6 @@ const DeaconSessionName = "gt-deacon"
 // DeaconRole is the role name for the Deacon's handoff bead.
 const DeaconRole = "deacon"
 
-// deaconMOTDMessages contains rotating motivational and educational tips
-// for the Deacon heartbeat. These make the thankless patrol role more fun.
-var deaconMOTDMessages = []string{
-	"Thanks for keeping the town running!",
-	"You are Gas Town's most critical role.",
-	"You are the heart of Gas Town! Be watchful!",
-	"Tip: Polecats are transient - spawn freely, kill liberally.",
-	"Tip: Witnesses monitor polecats; you monitor witnesses.",
-	"Tip: Wisps are transient molecules for patrol cycles.",
-	"The town sleeps soundly because you never do.",
-	"Tip: Mayor handles cross-rig coordination; you handle health.",
-	"Your vigilance keeps the agents honest.",
-	"Tip: Use 'gt deacon heartbeat' to signal you're alive.",
-	"Every heartbeat you check keeps Gas Town beating.",
-	"Tip: Stale agents need nudging; very stale ones need restarting.",
-}
-
-// nextMOTD returns the next MOTD message, rotating through the list
-// and avoiding consecutive repeats.
-func (d *Daemon) nextMOTD() string {
-	if len(deaconMOTDMessages) == 0 {
-		return "HEARTBEAT: run your rounds"
-	}
-
-	// Pick a random index that's different from the last one
-	nextIdx := d.lastMOTDIndex
-	for nextIdx == d.lastMOTDIndex && len(deaconMOTDMessages) > 1 {
-		nextIdx = int(time.Now().UnixNano() % int64(len(deaconMOTDMessages)))
-	}
-	d.lastMOTDIndex = nextIdx
-	return deaconMOTDMessages[nextIdx]
-}
 
 // ensureDeaconRunning ensures the Deacon is running.
 // ZFC-compliant: trusts agent bead state, no tmux inference.
@@ -285,37 +257,6 @@ func (d *Daemon) ensureDeaconRunning() {
 	d.logger.Println("Deacon session started successfully")
 }
 
-// pokeDeacon sends a heartbeat message to the Deacon session.
-// ZFC-compliant: trusts agent bead state, no tmux inference.
-// The Deacon molecule decides what to do with heartbeats.
-func (d *Daemon) pokeDeacon() {
-	// Check agent bead state (ZFC: trust what agent reports)
-	beadState, beadErr := d.getAgentBeadState("gt-deacon")
-	if beadErr != nil || (beadState != "running" && beadState != "working") {
-		// Agent not running per bead - don't poke (ensureDeaconRunning should start it)
-		return
-	}
-
-	// Agent reports running - send heartbeat
-	motd := d.nextMOTD()
-	msg := fmt.Sprintf("HEARTBEAT: %s", motd)
-	if err := d.tmux.SendKeysReplace(DeaconSessionName, msg, 50); err != nil {
-		d.logger.Printf("Error poking Deacon: %v", err)
-		return
-	}
-
-	d.logger.Println("Poked Deacon")
-}
-
-// witnessMOTDMessages contains rotating tips for witness heartbeats.
-var witnessMOTDMessages = []string{
-	"Time to patrol! Check your polecats.",
-	"Tip: Survey polecat health via agent beads.",
-	"Tip: Verify git state before killing polecats.",
-	"Your vigilance keeps polecats honest.",
-	"Tip: Escalate stuck workers to Mayor.",
-	"Tip: Send MERGE_READY when work is done.",
-}
 
 // ensureWitnessesRunning ensures witnesses are running for all rigs.
 // Called on each heartbeat to maintain witness patrol loops.
@@ -365,38 +306,6 @@ func (d *Daemon) ensureWitnessRunning(rigName string) {
 	d.logger.Printf("Witness session for %s started successfully", rigName)
 }
 
-// pokeWitnesses sends heartbeat messages to all witnesses.
-func (d *Daemon) pokeWitnesses() {
-	rigs := d.getKnownRigs()
-	for _, rigName := range rigs {
-		d.pokeWitness(rigName)
-	}
-}
-
-// pokeWitness sends a heartbeat to a specific rig's witness.
-func (d *Daemon) pokeWitness(rigName string) {
-	agentID := beads.WitnessBeadID(rigName)
-	sessionName := "gt-" + rigName + "-witness"
-
-	// Check agent bead state (ZFC: trust what agent reports)
-	beadState, beadErr := d.getAgentBeadState(agentID)
-	if beadErr != nil || (beadState != "running" && beadState != "working") {
-		// Agent not running per bead - don't poke
-		return
-	}
-
-	// Agent reports running - send heartbeat
-	idx := int(time.Now().UnixNano() % int64(len(witnessMOTDMessages)))
-	motd := witnessMOTDMessages[idx]
-	msg := fmt.Sprintf("HEARTBEAT: %s", motd)
-
-	if err := d.tmux.SendKeysReplace(sessionName, msg, 50); err != nil {
-		d.logger.Printf("Error poking witness for %s: %v", rigName, err)
-		return
-	}
-
-	d.logger.Printf("Poked witness for %s", rigName)
-}
 
 // getKnownRigs returns list of registered rig names.
 func (d *Daemon) getKnownRigs() []string {
@@ -406,17 +315,14 @@ func (d *Daemon) getKnownRigs() []string {
 		return nil
 	}
 
-	// Simple extraction - look for rig names in the JSON
-	// Full parsing would require importing config package
-	var rigs []string
-	// Parse just enough to get rig names
-	type rigsJSON struct {
+	var parsed struct {
 		Rigs map[string]interface{} `json:"rigs"`
 	}
-	var parsed rigsJSON
 	if err := json.Unmarshal(data, &parsed); err != nil {
 		return nil
 	}
+
+	var rigs []string
 	for name := range parsed.Rigs {
 		rigs = append(rigs, name)
 	}

internal/daemon/lifecycle.go

@@ -699,3 +699,210 @@ func identityToBDActor(identity string) string {
 		return identity
 	}
 }
+
+// GUPPViolationTimeout is how long an agent can have work on hook without
+// progressing before it's considered a GUPP (Gas Town Universal Propulsion
+// Principle) violation. GUPP states: if you have work on your hook, you run it.
+const GUPPViolationTimeout = 30 * time.Minute
+
+// checkGUPPViolations looks for agents that have work-on-hook but aren't
+// progressing. This is a GUPP violation: agents with hooked work must execute.
+// The daemon detects these and notifies the relevant Witness for remediation.
+func (d *Daemon) checkGUPPViolations() {
+	// Check polecat agents - they're the ones with work-on-hook
+	rigs := d.getKnownRigs()
+	for _, rigName := range rigs {
+		d.checkRigGUPPViolations(rigName)
+	}
+}
+
+// checkRigGUPPViolations checks polecats in a specific rig for GUPP violations.
+func (d *Daemon) checkRigGUPPViolations(rigName string) {
+	// List polecat agent beads for this rig
+	// Pattern: gt-polecat-<rig>-<name>
+	cmd := exec.Command("bd", "list", "--type=agent", "--json")
+	cmd.Dir = d.config.TownRoot
+
+	output, err := cmd.Output()
+	if err != nil {
+		return // Silently fail - bd might not be available
+	}
+
+	var agents []struct {
+		ID          string `json:"id"`
+		Type        string `json:"issue_type"`
+		Description string `json:"description"`
+		UpdatedAt   string `json:"updated_at"`
+	}
+
+	if err := json.Unmarshal(output, &agents); err != nil {
+		return
+	}
+
+	prefix := "gt-polecat-" + rigName + "-"
+	for _, agent := range agents {
+		// Only check polecats for this rig
+		if !strings.HasPrefix(agent.ID, prefix) {
+			continue
+		}
+
+		// Parse agent fields
+		fields := beads.ParseAgentFieldsFromDescription(agent.Description)
+		if fields == nil {
+			continue
+		}
+
+		// Check if agent has work on hook
+		if fields.HookBead == "" {
+			continue // No hooked work - no GUPP violation possible
+		}
+
+		// Check if agent is actively working
+		if fields.AgentState == "working" || fields.AgentState == "running" {
+			// Check when the agent bead was last updated
+			updatedAt, err := time.Parse(time.RFC3339, agent.UpdatedAt)
+			if err != nil {
+				continue
+			}
+
+			age := time.Since(updatedAt)
+			if age > GUPPViolationTimeout {
+				d.logger.Printf("GUPP violation: agent %s has hook_bead=%s but hasn't updated in %v (timeout: %v)",
+					agent.ID, fields.HookBead, age.Round(time.Minute), GUPPViolationTimeout)
+
+				// Notify the witness for this rig
+				d.notifyWitnessOfGUPP(rigName, agent.ID, fields.HookBead, age)
+			}
+		}
+	}
+}
+
+// notifyWitnessOfGUPP sends a mail to the rig's witness about a GUPP violation.
+func (d *Daemon) notifyWitnessOfGUPP(rigName, agentID, hookBead string, stuckDuration time.Duration) {
+	witnessAddr := rigName + "/witness"
+	subject := fmt.Sprintf("GUPP_VIOLATION: %s stuck for %v", agentID, stuckDuration.Round(time.Minute))
+	body := fmt.Sprintf(`Agent %s has work on hook but isn't progressing.
+
+hook_bead: %s
+stuck_duration: %v
+
+Action needed: Check if agent is alive and responsive. Consider restarting if stuck.`,
+		agentID, hookBead, stuckDuration.Round(time.Minute))
+
+	cmd := exec.Command("gt", "mail", "send", witnessAddr, "-s", subject, "-m", body)
+	cmd.Dir = d.config.TownRoot
+
+	if err := cmd.Run(); err != nil {
+		d.logger.Printf("Warning: failed to notify witness of GUPP violation: %v", err)
+	} else {
+		d.logger.Printf("Notified %s of GUPP violation for %s", witnessAddr, agentID)
+	}
+}
+
+// checkOrphanedWork looks for work assigned to dead agents.
+// Orphaned work needs to be reassigned or the agent needs to be restarted.
+func (d *Daemon) checkOrphanedWork() {
+	// Get list of dead agents
+	deadAgents := d.getDeadAgents()
+	if len(deadAgents) == 0 {
+		return
+	}
+
+	// For each dead agent, check if they have hooked work
+	for _, agent := range deadAgents {
+		fields := beads.ParseAgentFieldsFromDescription(agent.Description)
+		if fields == nil || fields.HookBead == "" {
+			continue // No hooked work to orphan
+		}
+
+		d.logger.Printf("Orphaned work detected: agent %s is dead but has hook_bead=%s",
+			agent.ID, fields.HookBead)
+
+		// Determine the rig from the agent ID (gt-polecat-<rig>-<name>)
+		rigName := d.extractRigFromAgentID(agent.ID)
+		if rigName != "" {
+			d.notifyWitnessOfOrphanedWork(rigName, agent.ID, fields.HookBead)
+		}
+	}
+}
+
+// getDeadAgents returns all agent beads with state=dead.
+func (d *Daemon) getDeadAgents() []struct {
+	ID          string
+	Description string
+} {
+	cmd := exec.Command("bd", "list", "--type=agent", "--json")
+	cmd.Dir = d.config.TownRoot
+
+	output, err := cmd.Output()
+	if err != nil {
+		return nil
+	}
+
+	var agents []struct {
+		ID          string `json:"id"`
+		Type        string `json:"issue_type"`
+		Description string `json:"description"`
+	}
+
+	if err := json.Unmarshal(output, &agents); err != nil {
+		return nil
+	}
+
+	var dead []struct {
+		ID          string
+		Description string
+	}
+
+	for _, agent := range agents {
+		fields := beads.ParseAgentFieldsFromDescription(agent.Description)
+		if fields != nil && fields.AgentState == "dead" {
+			dead = append(dead, struct {
+				ID          string
+				Description string
+			}{agent.ID, agent.Description})
+		}
+	}
+
+	return dead
+}
+
+// extractRigFromAgentID extracts the rig name from a polecat agent ID.
+// Example: gt-polecat-gastown-max → gastown
+func (d *Daemon) extractRigFromAgentID(agentID string) string {
+	// Pattern: gt-polecat-<rig>-<name>
+	if !strings.HasPrefix(agentID, "gt-polecat-") {
+		return ""
+	}
+
+	rest := strings.TrimPrefix(agentID, "gt-polecat-")
+	// Find the rig name (everything before the last dash)
+	lastDash := strings.LastIndex(rest, "-")
+	if lastDash == -1 {
+		return ""
+	}
+
+	return rest[:lastDash]
+}
+
+// notifyWitnessOfOrphanedWork sends a mail to the rig's witness about orphaned work.
+func (d *Daemon) notifyWitnessOfOrphanedWork(rigName, agentID, hookBead string) {
+	witnessAddr := rigName + "/witness"
+	subject := fmt.Sprintf("ORPHANED_WORK: %s has hooked work but is dead", agentID)
+	body := fmt.Sprintf(`Agent %s is dead but has work on its hook.
+
+hook_bead: %s
+
+Action needed: Either restart the agent or reassign the work.`,
+		agentID, hookBead)
+
+	cmd := exec.Command("gt", "mail", "send", witnessAddr, "-s", subject, "-m", body)
+	cmd.Dir = d.config.TownRoot
+
+	if err := cmd.Run(); err != nil {
+		d.logger.Printf("Warning: failed to notify witness of orphaned work: %v", err)
+	} else {
+		d.logger.Printf("Notified %s of orphaned work for %s", witnessAddr, agentID)
+	}
+}
+