AgentBay:AgentBay Skill Suite

Obter chave de API

1. Acesse o console do AgentBay.

2. No painel de navegação à esquerda, escolha Service Management e clique em no botão de cópia da chave de API desejada.

   Nota

   Se não houver nenhuma chave de API disponível, clique em Create API Key, insira um nome e clique em OK para criar uma chave de API.

Defina variáveis de ambiente

Linux/macOS:

export AGENTBAY_API_KEY=your_api_key_here

Windows:

setx AGENTBAY_API_KEY your_api_key_here

from agentbay import AgentBay

agent_bay = AgentBay()

# Get the metadata and root path for all available skills (no sandbox startup required)
metadata = agent_bay.beta_skills.get_metadata()
print(f"Skills root path: {metadata.skills_root_path}")
print(f"Number of available skills: {len(metadata.skills)}")
for skill in metadata.skills:
    print(f"  - {skill.name}: {skill.description}")

# Optional: Filter by skill name
filtered = agent_bay.beta_skills.get_metadata(skill_names=["my-skill"])

from agentbay import AgentBay, CreateSessionParams

agent_bay = AgentBay()

# Create a session with skills enabled
# With load_skills=True, skills from the enterprise namespace are automatically mounted.
result = agent_bay.create(CreateSessionParams(
    load_skills=True               # ← Enable skill loading
))
session = result.session

# You can also load only specified skills
result = agent_bay.create(CreateSessionParams(
    load_skills=True,
    skill_names=["skill-1", "skill-2"]
))

# Skills are mounted as files in the sandbox and can be read through the file system
metadata = agent_bay.beta_skills.get_metadata()
root = metadata.skills_root_path  # e.g., "/home/wuying/skills"

# Read the skill's documentation file
skill_md = session.file_system.read_file(f"{root}/my-skill/SKILL.md")
print(f"Skill documentation:\n{skill_md.content}")

# List all skills loaded in the sandbox
ls_result = session.command.execute_command(f"ls {root}")
print(f"Loaded skills:\n{ls_result.output}")

# After the Agent decides to use a skill, run it using the sandbox command execution capability
# Skill files are already mounted in the session and can be read and executed directly

# Method 1: Directly execute the skill's script
result = session.command.execute_command(
    f"python3 {root}/my-skill/scripts/main.py"
)
print(f"Execution result: {result.output}")

# Method 2: Read the script content for flexible handling
script_content = session.file_system.read_file(f"{root}/my-skill/scripts/main.py")
print(f"Script content: {script_content.content}")

from agentbay import AgentBay, CreateSessionParams

# ========== 1. Initialize ==========
agent_bay = AgentBay()

# ========== 2. Query skill metadata (without creating a sandbox) ==========
metadata = agent_bay.beta_skills.get_metadata()
root = metadata.skills_root_path
print(f"Skills root path: {root}")
print(f"Number of available skills: {len(metadata.skills)}")

# ========== 3. Create a session with skills enabled ==========
result = agent_bay.create(CreateSessionParams(
    load_skills=True  # Automatically mount skills from the enterprise namespace
))
session = result.session

try:
    # ========== 4. Verify that skills are loaded ==========
    ls_result = session.command.execute_command(f"ls {root}")
    print(f"Loaded skills:\n{ls_result.output}")

    # ========== 5. Use an open-source skill ==========
    seo_md = session.file_system.read_file(f"{root}/seo-audit/SKILL.md")
    print(f"SEO skill description:\n{seo_md.content}")
    result = session.command.execute_command(f"python3 {root}/seo-audit/scripts/audit.py")
    print(f"SEO audit result: {result.output}")

    # ========== 6. Use an enterprise-created skill ==========
    result = session.command.execute_command(f"python3 {root}/crm-integration/scripts/sync_customer.py")
    print(f"CRM sync result: {result.output}")

    # ========== 7. Combine multiple skills ==========
    # Data collection → Data cleaning → Data analysis → Report generation
    for step_skill in ["data-collector", "data-cleaner", "data-analyzer", "report-generator"]:
        step_result = session.command.execute_command(f"python3 {root}/{step_skill}/scripts/main.py")
        print(f"{step_skill} execution result: {step_result.output}")

finally:
    session.delete()
    print("Session deleted.")

# Install the find-skills skill
npx skills add https://github.com/agentbay-ai/agentbay-skills -s find-skills

User: "I need a sandbox that can run Python code."
↓
Cursor Agent automatically calls find-skills to search
↓
Agent: "I found 'agentbay-aio-skills', which can run Python/JS/R/Java code in a secure sandbox. Do you want to install it?"
↓
User: "Install it."
↓
Agent automatically installs the skill
↓
User: "Help me run this code: print('Hello from sandbox!')"
↓
Agent uses the newly installed sandbox skill to execute the code
↓
Agent: "Execution result: Hello from sandbox!"