refactor: query toolsrv directly for proc list

- ListProcs() now queries toolsrv directly instead of local tracker
- Remove ownership verification from SignalProc/ProcOutput/DismissProc
- Clean up stale procs from local tracker when they disappear from toolsrv
- Document local tracker quirk in TODO.md for future cleanup

cmd/olliesrv/internal/agent/agent.go

@@ -583,52 +583,44 @@ type BgProc struct {
 	Status string // "running" or "exited (N)"
 }
 
-// ListProcs returns a list of background processes owned by this agent.
+// ListProcs returns a list of background processes from toolsrv.
 func (ag *Agent) ListProcs() []BgProc {
-	procs := ag.bgProcs.List()
 	ts := ag.ToolServer()
-	out := make([]BgProc, len(procs))
-	for i, p := range procs {
-		status := "running"
-		if ts != nil {
-			info := ts.ListDetachedRaw()
-			for _, raw := range info {
-				m, ok := raw.(map[string]any)
-				if !ok {
-					continue
-				}
-				if pid, ok := m["pid"].(int); ok && pid == p.PID {
-					if exited, ok := m["exited"].(bool); ok && exited {
-						code := 0
-						if c, ok := m["exit_code"].(int); ok {
-							code = c
-						}
-						status = fmt.Sprintf("exited (%d)", code)
-					}
-					break
-				}
-			}
+	if ts == nil {
+		return nil
+	}
+	
+	var out []BgProc
+	for _, raw := range ts.ListDetachedRaw() {
+		m, ok := raw.(map[string]any)
+		if !ok {
+			continue
 		}
-		out[i] = BgProc{PID: p.PID, Tool: p.Tool, Cmd: p.Cmd, Status: status}
+		pid, _ := m["pid"].(int)
+		cmd, _ := m["command"].(string)
+		
+		status := "running"
+		if exited, ok := m["exited"].(bool); ok && exited {
+			code := 0
+			if c, ok := m["exit_code"].(int); ok {
+				code = c
+			}
+			status = fmt.Sprintf("exited (%d)", code)
+		}
+		
+		out = append(out, BgProc{
+			PID:    pid,
+			Tool:   "shell", // toolsrv doesn't track tool name currently
+			Cmd:    cmd,
+			Status: status,
+		})
 	}
 	return out
 }
 
-// SignalProc sends a signal to a background process owned by this agent.
+// SignalProc sends a signal to a background process.
 // sig is "term" or "kill".
 func (ag *Agent) SignalProc(pid int, sig string) error {
-	// Verify this process belongs to this agent
-	procs := ag.bgProcs.List()
-	found := false
-	for _, p := range procs {
-		if p.PID == pid {
-			found = true
-			break
-		}
-	}
-	if !found {
-		return fmt.Errorf("process %d not found", pid)
-	}
 	ts := ag.runtime.ToolServer
 	if ts == nil {
 		return fmt.Errorf("no tool server")
@@ -645,20 +637,8 @@ func (ag *Agent) SignalProc(pid int, sig string) error {
 	return ts.SignalDetached(pid, s)
 }
 
-// ProcOutput returns the output of a background process owned by this agent.
+// ProcOutput returns the output of a background process.
 func (ag *Agent) ProcOutput(pid int) (string, error) {
-	// Verify this process belongs to this agent
-	procs := ag.bgProcs.List()
-	found := false
-	for _, p := range procs {
-		if p.PID == pid {
-			found = true
-			break
-		}
-	}
-	if !found {
-		return "", fmt.Errorf("process %d not found", pid)
-	}
 	ts := ag.runtime.ToolServer
 	if ts == nil {
 		return "", fmt.Errorf("no tool server")
@@ -666,28 +646,16 @@ func (ag *Agent) ProcOutput(pid int) (string, error) {
 	return ts.GetDetachedOutput(pid)
 }
 
-// DismissProc dismisses a background process owned by this agent.
+// DismissProc dismisses a background process.
 func (ag *Agent) DismissProc(pid int) error {
-	// Verify this process belongs to this agent
-	procs := ag.bgProcs.List()
-	found := false
-	for _, p := range procs {
-		if p.PID == pid {
-			found = true
-			break
-		}
-	}
-	if !found {
-		return fmt.Errorf("process %d not found", pid)
-	}
 	ts := ag.runtime.ToolServer
 	if ts == nil {
 		return fmt.Errorf("no tool server")
 	}
 	if !ts.DismissDetached(pid) {
-		return fmt.Errorf("failed to dismiss process %d", pid)
+		return fmt.Errorf("process %d not found", pid)
 	}
-	ag.bgProcs.Remove(pid)
+	ag.bgProcs.Remove(pid) // Also remove from local tracker if present
 	return nil
 }
 

cmd/olliesrv/internal/agent/background.go

@@ -56,6 +56,7 @@ func (bt *bgTracker) CollectInterrupts(conn *toolsrv.Conn) string {
 	}
 
 	var sb strings.Builder
+	var gone []int
 
 	for i := range bt.procs {
 		p := &bt.procs[i]
@@ -63,7 +64,8 @@ func (bt *bgTracker) CollectInterrupts(conn *toolsrv.Conn) string {
 		// Get current output
 		output, err := conn.GetDetachedOutput(p.PID)
 		if err != nil {
-			// Process may have been dismissed from toolsrv or doesn't exist - skip it
+			// Process no longer exists in toolsrv (GC'd or dismissed) - mark for removal
+			gone = append(gone, i)
 			continue
 		}
 
@@ -107,6 +109,12 @@ func (bt *bgTracker) CollectInterrupts(conn *toolsrv.Conn) string {
 		sb.WriteString("\n</system-proc-interrupt>")
 	}
 
+	// Remove procs that no longer exist in toolsrv (reverse order to preserve indices)
+	for i := len(gone) - 1; i >= 0; i-- {
+		idx := gone[i]
+		bt.procs = append(bt.procs[:idx], bt.procs[idx+1:]...)
+	}
+
 	return sb.String()
 }
 

doc/TODO.md

@@ -0,0 +1,16 @@
+
+## Background Process Tracking Cleanup
+
+The agent maintains a local `bgTracker` (`bgProcs`) that partially duplicates toolsrv's proc state. This creates sync issues:
+
+- `ListProcs()` now queries toolsrv directly (correct)
+- `CollectInterrupts()` still uses local tracker for `LastOutput` diff detection
+- `bgProcs.Add()` is still called when starting background procs
+- `bgProcs.Remove()` is called on dismiss
+
+Consider:
+1. Remove local tracker entirely, track only `LastOutput` map keyed by PID
+2. Or add agent ID to toolsrv Proc struct for proper ownership filtering
+3. Move `LastOutput` tracking to toolsrv (add a "read offset" or "last read hash")
+
+Current state works but is fragile - toolsrv is source of truth for proc list, local tracker is only for interrupt diff detection.