fix: concurrency safety, regex caching, lint issues

- agent.go: guard reader goroutine msgCh send with sessionCtx.Done()
  to prevent permanent goroutine hang
- checks.go: cache regex.Compile (not MatchString) per HTTP check;
  cache x509.SystemCertPool via sync.Once across SSL checks
- lldp.go: remove dead walkErrors variable
- ssh_test.go: reset global host key store to fix TOFU port collision
  flakiness across tests
- .gitignore: add towerops-agent binary, .tool-versions

.gitignore

@@ -33,5 +33,8 @@ result
 # Local scripts
 monitor-deploy.sh
 
+# Tool versions (asdf, mise)
+.tool-versions
+
 # Claude
 CLAUDE.md

agent.go

@@ -196,7 +196,11 @@ func runSession(ctx context.Context, baseURL, token string) error {
 				sessionCancel()
 				return
 			}
-			msgCh <- data
+			select {
+			case msgCh <- data:
+			case <-sessionCtx.Done():
+				return
+			}
 		}
 	}()
 
@@ -319,7 +323,7 @@ func runSession(ctx context.Context, baseURL, token string) error {
 		case data := <-msgCh:
 			var msg channelMsg
 			if err := json.Unmarshal(data, &msg); err != nil {
-				slog.Warn("invalid message", "error", err)
+				slog.Debug("invalid message", "error", err)
 				continue
 			}
 			shouldEnd, endErr := handleMessage(sessionCtx, msg, pools, snmpResultCh, mikrotikResultCh, credTestResultCh, monitoringCheckCh, checkResultCh, lldpTopologyResultCh)

checks.go

@@ -11,6 +11,7 @@ import (
 	"regexp"
 	"strconv"
 	"strings"
+	"sync"
 	"time"
 
 	"codeberg.org/towerops-agent/towerops-agent/pb"
@@ -28,7 +29,18 @@ var (
 		MaxIdleConnsPerHost: 10,
 		IdleConnTimeout:     90 * time.Second,
 	}
-	sslRootCAs = x509.SystemCertPool
+	sslRootCAsOnce sync.Once
+	sslRootCAsPool *x509.CertPool
+	sslRootCAsErr  error
+
+	// sslRootCAs returns the system cert pool, cached after first load.
+	// Overridable for tests.
+	sslRootCAs = func() (*x509.CertPool, error) {
+		sslRootCAsOnce.Do(func() {
+			sslRootCAsPool, sslRootCAsErr = x509.SystemCertPool()
+		})
+		return sslRootCAsPool, sslRootCAsErr
+	}
 )
 
 // ExecuteCheck runs a service check and returns the result.
@@ -146,17 +158,16 @@ func executeHTTPCheck(ctx context.Context, config *pb.HttpCheckConfig, timeoutMs
 
 	// Check content regex if provided
 	if config.Regex != "" {
+		re, err := regexp.Compile(config.Regex)
+		if err != nil {
+			return 2, fmt.Sprintf("Invalid regex: %v", err), responseTime
+		}
 		body, err := io.ReadAll(io.LimitReader(resp.Body, 1<<20))
 		if err != nil {
 			return 2, fmt.Sprintf("Failed to read body: %v", err), responseTime
 		}
 
-		matched, err := regexp.MatchString(config.Regex, string(body))
+		if !re.Match(body) {
-		if err != nil {
-			return 2, fmt.Sprintf("Invalid regex: %v", err), responseTime
-		}
-
-		if !matched {
 			return 2, fmt.Sprintf("Content does not match pattern: %s", config.Regex), responseTime
 		}
 	}

lldp.go

@@ -106,7 +106,6 @@ func discoverLldpNeighbors(client *gosnmp.GoSNMP, deviceID, jobID string) (*pb.L
 
 	// Walk remote port descriptions
 	remotePorts := make(map[string]string)
-	var walkErrors []string
 	if err := client.Walk(oidRemPortDesc, func(pdu gosnmp.SnmpPDU) error {
 		key := parseRemoteKey(pdu.Name, oidRemPortDesc)
 		if key != "" {
@@ -114,7 +113,6 @@ func discoverLldpNeighbors(client *gosnmp.GoSNMP, deviceID, jobID string) (*pb.L
 		}
 		return nil
 	}); err != nil {
-		walkErrors = append(walkErrors, fmt.Sprintf("walk remote port descriptions: %v", err))
 		slog.Warn("failed to walk remote port descriptions", "error", err)
 	}
 
@@ -127,7 +125,6 @@ func discoverLldpNeighbors(client *gosnmp.GoSNMP, deviceID, jobID string) (*pb.L
 		}
 		return nil
 	}); err != nil {
-		walkErrors = append(walkErrors, fmt.Sprintf("walk remote port IDs: %v", err))
 		slog.Warn("failed to walk remote port IDs", "error", err)
 	}
 
@@ -140,7 +137,6 @@ func discoverLldpNeighbors(client *gosnmp.GoSNMP, deviceID, jobID string) (*pb.L
 		}
 		return nil
 	}); err != nil {
-		walkErrors = append(walkErrors, fmt.Sprintf("walk management addresses: %v", err))
 		slog.Warn("failed to walk management addresses", "error", err)
 	}
 
@@ -172,9 +168,6 @@ func discoverLldpNeighbors(client *gosnmp.GoSNMP, deviceID, jobID string) (*pb.L
 		result.Neighbors = append(result.Neighbors, neighbor)
 	}
 
-	// Walk errors are already logged; return result with whatever data was collected
-	_ = walkErrors
-
 	return result, nil
 }
 

ssh_test.go

@@ -7,7 +7,9 @@ import (
 	"crypto/rand"
 	"fmt"
 	"net"
+	"path/filepath"
 	"strings"
+	"sync"
 	"testing"
 	"time"
 
@@ -382,6 +384,15 @@ func TestExecuteMikrotikBackupWithOutput(t *testing.T) {
 }
 
 func TestExecuteMikrotikBackupSessionError(t *testing.T) {
+	// Reset global host key store to avoid TOFU collisions from other tests
+	origStore := globalHostKeys
+	defer func() {
+		hostKeysOnce = sync.Once{}
+		globalHostKeys = origStore
+	}()
+	hostKeysOnce = sync.Once{}
+	t.Setenv("TOWEROPS_HOST_KEYS_FILE", filepath.Join(t.TempDir(), "hosts.json"))
+
 	// SSH server that accepts connection but rejects all channel requests
 	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
 	if err != nil {