From 45bb219331a2dc2be8a4a7c05bd41b95aaa924f0 Mon Sep 17 00:00:00 2001
From: Marcos Filipe <mfilipe@renderedtext.com>
Date: Wed, 6 May 2026 13:19:53 -0300
Subject: [PATCH] fix: fix connectivity issues

---
 internal/providers/semaphore/api.go     | 24 +++--------
 internal/providers/semaphore/backend.go | 54 +++++++++++++++++++------
 2 files changed, 47 insertions(+), 31 deletions(-)

diff --git a/internal/providers/semaphore/api.go b/internal/providers/semaphore/api.go
index 927f0f2..5c886fd 100644
--- a/internal/providers/semaphore/api.go
+++ b/internal/providers/semaphore/api.go
@@ -8,6 +8,7 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"os"
 	"time"
 
 	core "github.com/openclaw/crabbox/internal/cli"
@@ -43,25 +44,9 @@ func (c *apiClient) CreateJob(ctx context.Context, project, machine, osImage str
 		return "", fmt.Errorf("resolve project %q: %w", project, err)
 	}
 
-	durationSecs := int(idleTimeout.Seconds()) * 2 // max duration = 2x idle timeout
-	idleSecs := int(idleTimeout.Seconds())
+	durationSecs := int(idleTimeout.Seconds())
 
-	keepalive := fmt.Sprintf(
-		`echo crabbox-testbox-ready && touch /tmp/.testbox-activity && python3 -c "
-import os, time, sys
-max_duration = %d
-idle_timeout = %d
-start = time.time()
-f = '/tmp/.testbox-activity'
-while True:
-    if time.time() - start >= max_duration:
-        sys.exit(0)
-    try:
-        if time.time() - os.path.getmtime(f) >= idle_timeout:
-            sys.exit(0)
-    except: pass
-    time.sleep(5)
-"`, durationSecs, idleSecs)
+	keepalive := fmt.Sprintf("sudo mkdir -p /work/crabbox && sudo chown $(whoami) /work/crabbox && echo crabbox-testbox-ready && sleep %d", durationSecs)
 
 	body := map[string]any{
 		"apiVersion": "v1alpha",
@@ -79,6 +64,8 @@ while True:
 		},
 	}
 
+	fmt.Fprintf(os.Stderr, "DEBUG: CreateJob url=https://%s/api/v1alpha/jobs project_id=%s\n", c.host, projectID)
+
 	var result struct {
 		Metadata struct {
 			ID string `json:"id"`
@@ -261,6 +248,7 @@ func (c *apiClient) post(ctx context.Context, path string, payload any, target a
 	req.Header.Set("Authorization", "Token "+c.token)
 	req.Header.Set("Content-Type", "application/json")
 	req.Header.Set("User-Agent", "crabbox-semaphore-provider")
+	fmt.Fprintf(os.Stderr, "DEBUG POST: %s transport=%T content-length=%d\n", req.URL, c.http.Transport, req.ContentLength)
 
 	resp, err := c.http.Do(req)
 	if err != nil {
diff --git a/internal/providers/semaphore/backend.go b/internal/providers/semaphore/backend.go
index f4539ea..3a8fcb6 100644
--- a/internal/providers/semaphore/backend.go
+++ b/internal/providers/semaphore/backend.go
@@ -4,6 +4,8 @@ package semaphore
 import (
 	"context"
 	"fmt"
+	"os"
+	"path/filepath"
 
 	core "github.com/openclaw/crabbox/internal/cli"
 )
@@ -60,18 +62,24 @@ func (b *semaphoreBackend) Acquire(ctx context.Context, req core.AcquireRequest)
 		return core.LeaseTarget{}, err
 	}
 
-	// 3. Get SSH key
+	// 3. Get SSH key and write to file (crabbox expects a file path)
 	sshKey, err := b.client.GetSSHKey(ctx, jobID)
 	if err != nil {
 		return core.LeaseTarget{}, err
 	}
 
+	keyPath, err := storeSSHKey(leaseID, sshKey)
+	if err != nil {
+		return core.LeaseTarget{}, fmt.Errorf("store SSH key: %w", err)
+	}
+
 	target := core.SSHTarget{
-		User:     "semaphore",
-		Host:     ip,
-		Key:      sshKey,
-		Port:     fmt.Sprintf("%d", sshPort),
-		TargetOS: core.TargetLinux,
+		User:       "semaphore",
+		Host:       ip,
+		Key:        keyPath,
+		Port:       fmt.Sprintf("%d", sshPort),
+		TargetOS:   core.TargetLinux,
+		ReadyCheck: "true", // Semaphore job is ready once SSH is reachable
 	}
 
 	server := core.Server{
@@ -94,9 +102,14 @@ func (b *semaphoreBackend) Acquire(ctx context.Context, req core.AcquireRequest)
 	return core.LeaseTarget{Server: server, SSH: target, LeaseID: leaseID}, nil
 }
 
-// Resolve looks up an existing Semaphore job by ID.
+// Resolve looks up an existing Semaphore job by ID or slug.
 func (b *semaphoreBackend) Resolve(ctx context.Context, req core.ResolveRequest) (core.LeaseTarget, error) {
-	jobID := stripLeasePrefix(req.ID)
+	// Resolve slug → lease ID via claim file
+	id := req.ID
+	if claim, found, err := core.ResolveLeaseClaim(id); err == nil && found {
+		id = claim.LeaseID
+	}
+	jobID := stripLeasePrefix(id)
 
 	state, ip, sshPort, err := b.client.GetJobStatus(ctx, jobID)
 	if err != nil {
@@ -112,12 +125,18 @@ func (b *semaphoreBackend) Resolve(ctx context.Context, req core.ResolveRequest)
 	}
 
 	leaseID := "sem_" + jobID
+	keyPath, err := storeSSHKey(leaseID, sshKey)
+	if err != nil {
+		return core.LeaseTarget{}, fmt.Errorf("store SSH key: %w", err)
+	}
+
 	target := core.SSHTarget{
-		User:     "semaphore",
-		Host:     ip,
-		Key:      sshKey,
-		Port:     fmt.Sprintf("%d", sshPort),
-		TargetOS: core.TargetLinux,
+		User:       "semaphore",
+		Host:       ip,
+		Key:        keyPath,
+		Port:       fmt.Sprintf("%d", sshPort),
+		TargetOS:   core.TargetLinux,
+		ReadyCheck: "true",
 	}
 	server := core.Server{
 		CloudID:  jobID,
@@ -166,6 +185,15 @@ func (b *semaphoreBackend) Touch(ctx context.Context, req core.TouchRequest) (co
 	return req.Lease.Server, nil
 }
 
+func storeSSHKey(leaseID, keyContent string) (string, error) {
+	dir := os.TempDir()
+	path := filepath.Join(dir, ".crabbox-sem-"+leaseID+".key")
+	if err := os.WriteFile(path, []byte(keyContent), 0600); err != nil {
+		return "", err
+	}
+	return path, nil
+}
+
 func stripLeasePrefix(leaseID string) string {
 	if len(leaseID) > 4 && leaseID[:4] == "sem_" {
 		return leaseID[4:]