Skills in Sandboxes

Coding agents discover skill files by scanning specific directories at startup. By placing TensorLake skill files in the right paths inside a sandbox image, any agent running in the sandbox will automatically pick them up without manual installation.

How Agents Discover Skills

Each coding agent scans a different directory for skill files:

Agent	Skill File	Discovery Path
Claude Code	`SKILL.md`	`~/.claude/skills/<name>/SKILL.md`
OpenAI Codex	`AGENTS.md`	`~/.agents/skills/<name>/SKILL.md` or `AGENTS.md` in working dir
Google ADK	`SKILL.md`	Loaded explicitly via `load_skill_from_dir()`
Cursor	`.mdc`	`.cursor/rules/*.mdc`
Cline	`.md`	`.clinerules/`
Windsurf	`.md`	`.windsurf/rules/*.md`
GitHub Copilot	`.md`	`.github/copilot-instructions.md`

To make skills work inside a sandbox, bake the skill files into the image at the paths the agent expects.

Create a Skills Image

Any Agent

The simplest way to install skills for any agent is with the skills CLI. It places skill files in the correct discovery paths for Claude Code, Codex, Cursor, Windsurf, and other supported agents.

Python
TypeScript
Dockerfile

from tensorlake import Image

image = (
    Image(name="with-skills", base_image="tensorlake/ubuntu-systemd")
    .run("apt-get update && apt-get install -y nodejs npm python3 python3-pip")
    .run("npm install -g skills")
    .run("skills add tensorlakeai/tensorlake-skills --all -y --copy")
    .run("python3 -m pip install --break-system-packages tensorlake")
)

import { Image } from "tensorlake";

const image = new Image({
  name: "with-skills",
  baseImage: "tensorlake/ubuntu-systemd",
})
  .run("apt-get update && apt-get install -y nodejs npm python3 python3-pip")
  .run("npm install -g skills")
  .run("skills add tensorlakeai/tensorlake-skills --all -y --copy")
  .run("python3 -m pip install --break-system-packages tensorlake");

Dockerfile

FROM tensorlake/ubuntu-systemd

RUN apt-get update && apt-get install -y nodejs npm python3 python3-pip
RUN npm install -g skills
RUN skills add tensorlakeai/tensorlake-skills --all -y --copy
RUN python3 -m pip install --break-system-packages tensorlake

--all installs skills to all detected agents
-y skips confirmation prompts for non-interactive use
--copy copies files instead of symlinking, which is more reliable inside containers

Claude Code Only

If you only need Claude Code support, copy the skill into ~/.claude/skills/ inside the image:

Python
TypeScript
Dockerfile

from tensorlake import Image

image = (
    Image(name="claude-code-skills", base_image="tensorlake/ubuntu-systemd")
    .run("apt-get update && apt-get install -y git python3 python3-pip")
    .run("git clone https://github.com/tensorlakeai/tensorlake-skills /tmp/tensorlake-skills")
    .run(
        "mkdir -p /root/.claude/skills/tensorlake && "
        "cp -r /tmp/tensorlake-skills/SKILL.md /tmp/tensorlake-skills/references "
        "/root/.claude/skills/tensorlake/"
    )
    .run("rm -rf /tmp/tensorlake-skills")
    .run("python3 -m pip install --break-system-packages tensorlake")
)

import { Image } from "tensorlake";

const image = new Image({
  name: "claude-code-skills",
  baseImage: "tensorlake/ubuntu-systemd",
})
  .run("apt-get update && apt-get install -y git python3 python3-pip")
  .run("git clone https://github.com/tensorlakeai/tensorlake-skills /tmp/tensorlake-skills")
  .run(
    "mkdir -p /root/.claude/skills/tensorlake && " +
      "cp -r /tmp/tensorlake-skills/SKILL.md /tmp/tensorlake-skills/references " +
      "/root/.claude/skills/tensorlake/",
  )
  .run("rm -rf /tmp/tensorlake-skills")
  .run("python3 -m pip install --break-system-packages tensorlake");

Dockerfile

FROM tensorlake/ubuntu-systemd

RUN apt-get update && apt-get install -y git python3 python3-pip
RUN git clone https://github.com/tensorlakeai/tensorlake-skills /tmp/tensorlake-skills
RUN mkdir -p /root/.claude/skills/tensorlake \
 && cp -r /tmp/tensorlake-skills/SKILL.md /tmp/tensorlake-skills/references /root/.claude/skills/tensorlake/
RUN rm -rf /tmp/tensorlake-skills
RUN python3 -m pip install --break-system-packages tensorlake

Claude Code scans ~/.claude/skills/ at startup. The SKILL.md file and references/ directory at /root/.claude/skills/tensorlake/ are auto-discovered.

Create a Reusable Sandbox Image

CLI
TypeScript

tl sbx image create ./Dockerfile --registered-name claude-code-skills

npx tl sbx image create ./Dockerfile --registered-name claude-code-skills

import { createSandboxImage, Image } from "tensorlake";

const image = new Image({
  name: "claude-code-skills",
  baseImage: "tensorlake/ubuntu-systemd",
})
  .run("apt-get update && apt-get install -y nodejs npm python3 python3-pip")
  .run("npm install -g skills")
  .run("skills add tensorlakeai/tensorlake-skills --all -y --copy")
  .run("python3 -m pip install --break-system-packages tensorlake");

await createSandboxImage(image, {
  contextDir: ".",
});

Then launch sandboxes from that image:

tl sbx new --image claude-code-skills

Use with the SDK

You can also install skills programmatically each time you create a sandbox:

Python
TypeScript

from tensorlake.sandbox import SandboxClient

client = SandboxClient()

with client.create_and_connect() as sandbox:
    sandbox.run("bash", ["-c", "apt-get update && apt-get install -y nodejs npm"])
    sandbox.run("bash", ["-c", "npm install -g skills"])
    sandbox.run("bash", ["-c", "skills add tensorlakeai/tensorlake-skills --all -y --copy"])

    result = sandbox.run(
        "find",
        ["/", "-name", "SKILL.md", "-type", "f", "-not", "-path", "*/node_modules/*"],
    )
    print(result.stdout)

import { SandboxClient } from "tensorlake";

const client = SandboxClient.forCloud({
  apiKey: process.env.TENSORLAKE_API_KEY,
});

const sandbox = await client.createAndConnect();

try {
  await sandbox.run("bash", {
    args: ["-lc", "apt-get update && apt-get install -y nodejs npm"],
  });
  await sandbox.run("bash", {
    args: ["-lc", "npm install -g skills"],
  });
  await sandbox.run("bash", {
    args: [
      "-lc",
      "skills add tensorlakeai/tensorlake-skills --all -y --copy",
    ],
  });

  const result = await sandbox.run("find", {
    args: [
      "/",
      "-name",
      "SKILL.md",
      "-type",
      "f",
      "-not",
      "-path",
      "*/node_modules/*",
    ],
  });
  console.log(result.stdout);
} finally {
  await sandbox.terminate();
  client.close();
}

For sandboxes you create frequently, use the sandbox image approach to avoid reinstalling skills on every launch.

What Gets Included

The skill repo contains SDK references that the agent uses as context:

tensorlake-skills/
├── AGENTS.md                       # Skill definition (OpenAI Codex)
├── SKILL.md                        # Skill definition (Claude Code, Google ADK)
└── references/
    ├── applications_sdk.md         # Orchestrate API reference
    ├── sandbox_sdk.md              # Sandbox API reference
    ├── documentai_sdk.md           # DocumentAI API reference
    └── integrations.md             # Integration patterns

Agent Skills

Learn about TensorLake skills and how to install them for your coding agent.

Sandbox Images

Create reusable sandbox images from Dockerfiles or the TensorLake image DSLs.

Overview

Guides

Use Cases

How Agents Discover Skills

Create a Skills Image

Any Agent

Claude Code Only

Create a Reusable Sandbox Image

Use with the SDK

What Gets Included

See Also

Agent Skills

Sandbox Images

Overview

Guides

Use Cases

​How Agents Discover Skills

​Create a Skills Image

​Any Agent

​Claude Code Only

​Create a Reusable Sandbox Image

​Use with the SDK

​What Gets Included

​See Also

Agent Skills

Sandbox Images

How Agents Discover Skills

Create a Skills Image

Any Agent

Claude Code Only

Create a Reusable Sandbox Image

Use with the SDK

What Gets Included

See Also