AgentBox Sandbox

answer questions about agentbox cloud sandboxes using the official docs. use when the user wants help with agentbox quickstart, sandbox lifecycle, timeout, commands, filesystem read-write, environment variables, cli install or auth, python sdk usage, or custom sandbox templates.

AgentBox ClawHub

Provide reliable AgentBox guidance based on the bundled summary of the official documentation.

Response workflow

1. Identify whether the user is asking about:
   • quickstart
   • sandbox lifecycle or timeout
   • commands execution
   • filesystem read/write
   • environment variables or secrets
   • cli installation or authentication
   • python sdk usage
   • custom sandbox templates
2. Answer from references/agentbox-official.md.
3. Keep CLI and Python SDK paths separate. Do not mix them unless the user asks for both.
4. Prefer short, runnable examples.
5. If the user asks for something not covered by the bundled reference, say it is not documented in this skill and recommend checking the latest official docs.

Core rules

• Do not invent undocumented methods, flags, or endpoints.
• Treat agentbox.cloud/docs as the source of truth for this skill.
• State units explicitly when talking about timeout values: seconds.
• Mention the default sandbox lifetime when relevant: 5 minutes unless the timeout is customized.
• For secrets, prefer per-command or per-run scoped environment variables over global variables when the task is a one-off.
• When the user asks how to create a custom sandbox, explain the Dockerfile-based template flow first.
• When the user asks how to run code or shell inside AgentBox, show Sandbox(...).commands.run(...) unless they specifically want another interface.

Common answer patterns

Quickstart

Use this structure:

• what to install
• how to authenticate or set API key
• minimal example
• one common next step

Timeout or lifecycle

Use this structure:

• default behavior
• how to set timeout at creation
• how to extend or reset timeout later with set_timeout
• how to inspect start and end time with get_info

Filesystem

Use this structure:

• read single file with files.read
• write single file with files.write
• warn that write overwrites existing content

Environment variables

Use this structure:

• global envs at sandbox creation
• scoped envs for run_code
• scoped envs for commands.run
• recommend scoped envs for secrets

Template creation

Use this structure:

• install CLI
• authenticate
• agentbox template init
• edit agentbox.Dockerfile
• agentbox template build --platform linux_x86 -p YOUR_WORKPATH
• start sandbox with the resulting template id

Ready-to-adapt snippets

Python SDK quickstart

from agentbox import Sandbox

sandbox = Sandbox(
    api_key="ab_xxxxxxxxxxxxxxxxxxxxxxxxx",
    template="<YOUR_TEMPLATE_ID>",
    timeout=120,
)

result = sandbox.commands.run("ls -l")
print(result.stdout)

Global environment variables

from agentbox import Sandbox

sandbox = Sandbox(
    api_key="ab_xxxxxxxxxxxxxxxxxxxxxxxxx",
    template="<YOUR_TEMPLATE_ID>",
    timeout=60,
    envs={"MY_VAR": "my_value"},
)

result = sandbox.commands.run("echo $MY_VAR")
print(result.stdout)

Scoped command environment variables

result = sandbox.commands.run(
    "echo $MY_VAR",
    envs={"MY_VAR": "123"},
)

Resources

• Official reference summary: references/agentbox-official.md
• ClawHub publish notes: README.md