agents_and_robots/cmd/agentctl/main.go

// Command agentctl manages Matrix agents: list, start, stop, remove.
//
// Usage:
//
//	agentctl list                  # all agents with their status
//	agentctl start                 # start all enabled agents
//	agentctl start assistant-bot   # start a specific agent
//	agentctl stop                  # stop all running agents
//	agentctl stop assistant-bot    # stop a specific agent
//	agentctl remove assistant-bot  # disable agent (keeps data)
package main

import (
	"fmt"
	"os"
	"os/exec"
	"path/filepath"
	"strconv"
	"strings"
	"syscall"

	"github.com/spf13/cobra"

	"github.com/enmanuel/agents/internal/config"
)

const (
	runDir     = "run"      // PID + log files
	agentsGlob = "agents/*/config.yaml"
)

// ── entry point ───────────────────────────────────────────────────────────

func main() {
	var binPath string

	root := &cobra.Command{
		Use:   "agentctl",
		Short: "Manage Matrix agents",
		PersistentPreRunE: func(cmd *cobra.Command, args []string) error {
			return os.MkdirAll(runDir, 0o755)
		},
	}

	root.PersistentFlags().StringVar(&binPath, "bin", "",
		"Launcher binary path. Defaults to ./bin/launcher, falls back to 'go run ./cmd/launcher'")

	root.AddCommand(
		listCmd(),
		startCmd(&binPath),
		stopCmd(),
		removeCmd(),
	)

	if err := root.Execute(); err != nil {
		os.Exit(1)
	}
}

// ── list ──────────────────────────────────────────────────────────────────

func listCmd() *cobra.Command {
	return &cobra.Command{
		Use:     "list",
		Short:   "List all agents and their current status",
		Aliases: []string{"ls"},
		RunE: func(cmd *cobra.Command, args []string) error {
			agents, err := scanAgents()
			if err != nil {
				return err
			}
			if len(agents) == 0 {
				fmt.Println("No agents found under agents/*/config.yaml")
				return nil
			}

			fmt.Printf("%-20s  %-12s  %-8s  %s\n", "ID", "STATUS", "VERSION", "DESCRIPTION")
			fmt.Println(strings.Repeat("─", 72))
			for _, a := range agents {
				fmt.Printf("%-20s  %-12s  %-8s  %s\n",
					a.ID,
					statusLabel(a),
					a.Version,
					truncate(a.Desc, 36),
				)
			}
			return nil
		},
	}
}

// ── start ─────────────────────────────────────────────────────────────────

func startCmd(binPath *string) *cobra.Command {
	return &cobra.Command{
		Use:   "start [agent-id...]",
		Short: "Start one or all enabled agents",
		RunE: func(cmd *cobra.Command, args []string) error {
			agents, err := scanAgents()
			if err != nil {
				return err
			}

			targets := filterTargets(agents, args)
			if len(targets) == 0 {
				return fmt.Errorf("no matching agents found")
			}

			bin := resolvedBin(*binPath)
			started := 0

			for _, a := range targets {
				if !a.Enabled {
					fmt.Printf("skip  %-20s  (disabled in config)\n", a.ID)
					continue
				}
				if isRunning(a.ID) {
					fmt.Printf("skip  %-20s  (already running, PID %d)\n", a.ID, readPID(a.ID))
					continue
				}

				if err := startAgent(a, bin); err != nil {
					fmt.Fprintf(os.Stderr, "fail  %-20s  %v\n", a.ID, err)
					continue
				}

				fmt.Printf("start %-20s  PID %d  log → %s\n",
					a.ID, readPID(a.ID), logPath(a.ID))
				started++
			}

			if started == 0 {
				fmt.Println("Nothing started.")
			}
			return nil
		},
	}
}

// ── stop ──────────────────────────────────────────────────────────────────

func stopCmd() *cobra.Command {
	return &cobra.Command{
		Use:   "stop [agent-id...]",
		Short: "Stop one or all running agents",
		RunE: func(cmd *cobra.Command, args []string) error {
			agents, err := scanAgents()
			if err != nil {
				return err
			}

			targets := filterTargets(agents, args)
			if len(targets) == 0 {
				return fmt.Errorf("no matching agents found")
			}

			stopped := 0
			for _, a := range targets {
				pid := readPID(a.ID)
				if pid == 0 || !isRunning(a.ID) {
					fmt.Printf("skip  %-20s  (not running)\n", a.ID)
					continue
				}
				if err := syscall.Kill(pid, syscall.SIGTERM); err != nil {
					fmt.Fprintf(os.Stderr, "fail  %-20s  kill: %v\n", a.ID, err)
					continue
				}
				removePIDFile(a.ID)
				fmt.Printf("stop  %-20s  sent SIGTERM to PID %d\n", a.ID, pid)
				stopped++
			}

			if stopped == 0 {
				fmt.Println("Nothing stopped.")
			}
			return nil
		},
	}
}

// ── remove ────────────────────────────────────────────────────────────────

func removeCmd() *cobra.Command {
	return &cobra.Command{
		Use:   "remove <agent-id>",
		Short: "Disable an agent (sets enabled: false). Does not delete data.",
		Args:  cobra.ExactArgs(1),
		RunE: func(cmd *cobra.Command, args []string) error {
			id := args[0]

			agents, err := scanAgents()
			if err != nil {
				return err
			}

			var target *agentInfo
			for i := range agents {
				if agents[i].ID == id {
					target = &agents[i]
					break
				}
			}
			if target == nil {
				return fmt.Errorf("agent %q not found", id)
			}

			// Stop if running
			if isRunning(id) {
				pid := readPID(id)
				_ = syscall.Kill(pid, syscall.SIGTERM)
				removePIDFile(id)
				fmt.Printf("stop  %-20s  sent SIGTERM to PID %d\n", id, pid)
			}

			// Disable in config (preserves comments)
			if err := setEnabled(target.ConfigPath, false); err != nil {
				return fmt.Errorf("update config: %w", err)
			}

			fmt.Printf("ok    %-20s  marked as disabled in %s\n", id, target.ConfigPath)
			fmt.Printf("      Data preserved at agents/%s/data/\n", id)
			return nil
		},
	}
}

// ── agent scanning ────────────────────────────────────────────────────────

type agentInfo struct {
	ID         string
	Version    string
	Enabled    bool
	Desc       string
	ConfigPath string
}

func scanAgents() ([]agentInfo, error) {
	matches, err := filepath.Glob(agentsGlob)
	if err != nil {
		return nil, err
	}

	var agents []agentInfo
	for _, path := range matches {
		// Use LoadMeta so list works even when env vars aren't set.
		cfg, err := config.LoadMeta(path)
		if err != nil {
			fmt.Fprintf(os.Stderr, "warn  skipping %s: %v\n", path, err)
			continue
		}
		agents = append(agents, agentInfo{
			ID:         cfg.Agent.ID,
			Version:    cfg.Agent.Version,
			Enabled:    cfg.Agent.Enabled,
			Desc:       cfg.Agent.Description,
			ConfigPath: path,
		})
	}
	return agents, nil
}

func filterTargets(agents []agentInfo, ids []string) []agentInfo {
	if len(ids) == 0 {
		return agents // no filter → all
	}
	set := make(map[string]bool, len(ids))
	for _, id := range ids {
		set[id] = true
	}
	var out []agentInfo
	for _, a := range agents {
		if set[a.ID] {
			out = append(out, a)
		}
	}
	return out
}

// ── process management ────────────────────────────────────────────────────

func startAgent(a agentInfo, bin string) error {
	logFile, err := os.OpenFile(logPath(a.ID), os.O_CREATE|os.O_APPEND|os.O_WRONLY, 0o644)
	if err != nil {
		return fmt.Errorf("open log: %w", err)
	}

	var cmd *exec.Cmd
	if strings.HasPrefix(bin, "go run") {
		// dev mode: go run ./cmd/launcher -c <config>
		cmd = exec.Command("go", "run", "./cmd/launcher", "-c", a.ConfigPath)
	} else {
		cmd = exec.Command(bin, "-c", a.ConfigPath)
	}

	cmd.Stdout = logFile
	cmd.Stderr = logFile
	// Detach from the parent process group so it keeps running after agentctl exits
	cmd.SysProcAttr = &syscall.SysProcAttr{Setsid: true}

	if err := cmd.Start(); err != nil {
		logFile.Close()
		return fmt.Errorf("exec: %w", err)
	}

	// Write PID file — the subprocess owns its lifecycle now
	if err := os.WriteFile(pidPath(a.ID), []byte(strconv.Itoa(cmd.Process.Pid)), 0o644); err != nil {
		return fmt.Errorf("write PID: %w", err)
	}

	// Detach: don't wait for the process
	go func() { _ = cmd.Wait() }()

	return nil
}

func isRunning(id string) bool {
	pid := readPID(id)
	if pid == 0 {
		return false
	}
	err := syscall.Kill(pid, 0) // signal 0 checks existence without killing
	return err == nil
}

func readPID(id string) int {
	raw, err := os.ReadFile(pidPath(id))
	if err != nil {
		return 0
	}
	pid, _ := strconv.Atoi(strings.TrimSpace(string(raw)))
	return pid
}

func removePIDFile(id string) {
	_ = os.Remove(pidPath(id))
}

func pidPath(id string) string { return filepath.Join(runDir, id+".pid") }
func logPath(id string) string { return filepath.Join(runDir, id+".log") }

// ── config editing ────────────────────────────────────────────────────────

// setEnabled flips `enabled: true/false` in the agent section of the YAML.
// Uses text replacement to preserve all comments.
func setEnabled(configPath string, enabled bool) error {
	raw, err := os.ReadFile(configPath)
	if err != nil {
		return err
	}

	current := "enabled: true"
	replacement := "enabled: false"
	if enabled {
		current = "enabled: false"
		replacement = "enabled: true"
	}

	updated := strings.Replace(string(raw), current, replacement, 1)
	if updated == string(raw) {
		return nil // already in the desired state
	}

	return os.WriteFile(configPath, []byte(updated), 0o644)
}

// ── display helpers ───────────────────────────────────────────────────────

func statusLabel(a agentInfo) string {
	switch {
	case !a.Enabled:
		return "disabled"
	case isRunning(a.ID):
		return "● running"
	default:
		return "○ stopped"
	}
}

func truncate(s string, max int) string {
	if len(s) <= max {
		return s
	}
	return s[:max-1] + "…"
}

// resolvedBin returns the launcher binary path to use.
// Priority: --bin flag > ./bin/launcher (if exists) > go run fallback.
func resolvedBin(flagVal string) string {
	if flagVal != "" {
		return flagVal
	}
	if _, err := os.Stat("bin/launcher"); err == nil {
		return "bin/launcher"
	}
	return "go run ./cmd/launcher"
}