sandbox:sandbox-config-management

sandbox-config-management

Read, write, and manage Sandbox.toml configuration files that define sandbox setup and preferences.

Purpose

This skill provides knowledge for working with Sandbox.toml files - the configuration format used to define and share sandbox setups. Handle reading existing configurations, creating new ones, cloning and customizing configurations, and validating compatibility.

Sandbox.toml Structure

Complete Example

[sandbox]
name = "myproject"
location = "/Users/aaronbassett/Sandboxes/aaronbassett/myproject"
created = "2026-01-24T01:30:00Z"
base_image = "ubuntu:24.04"

[source]
type = "github"  # or "local" or "new"
repository = "aaronbassett/myproject"
branch = "main"  # optional

[languages]
rust = "1.93.0"  # or "stable", "nightly"
python = "3.14.2"  # or "current"
nodejs = "current"  # or "lts", "nightly", "18.20.0"

[languages.tools]
rust = ["clippy", "rustfmt", "cargo-dist", "cargo-deny", "cargo-release", "cocogitto"]
python = ["uv", "ruff", "mypy", "black", "pytest"]
nodejs = ["pnpm", "typescript", "ts-node"]

[claude]
marketplaces = [
    "anthropics/claude-plugins-official",
    "aaronbassett/agent-foundry"
]

[claude.plugins]
plugins = [
    "rust-analyzer-lsp@claude-plugins-official",
    "pyright-lsp@claude-plugins-official",
    "typescript-lsp@claude-plugins-official",
    "devs@agent-foundry",
    "git-lovely@agent-foundry",
    "settings-presets@agent-foundry"
]

[network]
ports = [3000, 3001, 3002]  # specific ports
# or
port_range = "3000-3999"  # range

[environment]
# Non-secret environment variables
NODE_ENV = "development"
RUST_BACKTRACE = "1"

[shell]
aliases = true  # use standard eza/git aliases
starship_theme = "red_container"
oh_my_zsh_plugins = ["git", "docker", "rust", "python", "node"]

[tools]
# Additional CLI tools beyond defaults
extra = ["bat", "delta", "hyperfine"]

Field Descriptions

[sandbox] section:

• name: Project name (required)
• location: Absolute path to sandbox directory (required)
• created: ISO 8601 timestamp of creation
• base_image: Docker base image (default: "ubuntu:24.04")

[source] section:

• type: "github", "local", or "new" (required)
• repository: GitHub repo (owner/name) if type="github"
• branch: Git branch (default: "main")

[languages] section:

• rust, python, nodejs: Version strings or keywords
• Keywords: "stable"/"current", "lts", "nightly"
• Specific versions: "1.93.0", "3.14.2", "18.20.0"

[languages.tools] section:

• Arrays of tool names to install for each language
• Empty array = no additional tools

[claude] section:

• marketplaces: List of marketplace repos
• plugins: List of plugins in "name@marketplace" format

[network] section:

• ports: Array of specific ports to forward
• port_range: String range like "3000-3999"
• Use one or the other, not both

[environment] section:

• Key-value pairs for environment variables
• Do not include secrets (use .env file instead)

[shell] section:

• aliases: Boolean to include standard aliases
• starship_theme: Theme identifier
• oh_my_zsh_plugins: List of plugin names

[tools] section:

• extra: Additional tools beyond standard set

Reading Configurations

Parse Sandbox.toml

Use scripts/parse_config.py:

python3 scripts/parse_config.py /path/to/sandbox/Sandbox.toml

Output: JSON representation of configuration

# In Python code
import tomli

with open("Sandbox.toml", "rb") as f:
    config = tomli.load(f)

# Access fields
project_name = config["sandbox"]["name"]
languages = config.get("languages", {})
rust_version = languages.get("rust")

Validate Configuration

Check for required fields and valid values:

def validate_config(config: dict) -> List[str]:
    """
    Validate Sandbox.toml structure.

    Returns list of validation errors (empty if valid).
    """
    errors = []

    # Required sections
    if "sandbox" not in config:
        errors.append("Missing [sandbox] section")
    else:
        sandbox = config["sandbox"]
        if "name" not in sandbox:
            errors.append("Missing sandbox.name")
        if "location" not in sandbox:
            errors.append("Missing sandbox.location")

    if "source" not in config:
        errors.append("Missing [source] section")
    else:
        source = config["source"]
        if "type" not in source:
            errors.append("Missing source.type")
        elif source["type"] not in ["github", "local", "new"]:
            errors.append(f"Invalid source.type: {source['type']}")

        if source.get("type") == "github" and "repository" not in source:
            errors.append("GitHub source requires repository field")

    # Warn about unknown fields (lenient)
    # Just log warnings, don't error

    return errors

Use scripts/validate_config.py for validation.

Read from Existing Sandbox

When user references an existing sandbox:

User: "Base this on ~/Sandboxes/aaronbassett/project1"

Process:

1. Check if directory exists
2. Look for Sandbox.toml in directory root
3. Parse configuration
4. Present configuration to user for customization

existing_config_path = Path("~/Sandboxes/aaronbassett/project1/Sandbox.toml").expanduser()

if not existing_config_path.exists():
    print("No Sandbox.toml found in that directory")
    # Offer to create from scratch
else:
    with open(existing_config_path, "rb") as f:
        existing_config = tomli.load(f)

    # Show user what's in existing config
    print(f"Found sandbox config for: {existing_config['sandbox']['name']}")
    print(f"Languages: {', '.join(existing_config.get('languages', {}).keys())}")
    # etc.

Writing Configurations

Create New Sandbox.toml

Build configuration from user answers and detected settings:

from datetime import datetime, timezone

config = {
    "sandbox": {
        "name": project_name,
        "location": str(sandbox_location),
        "created": datetime.now(timezone.utc).isoformat(),
        "base_image": base_image or "ubuntu:24.04",
    },
    "source": {
        "type": source_type,  # "github", "local", or "new"
    },
}

# Add repository if GitHub source
if source_type == "github":
    config["source"]["repository"] = repo_name
    config["source"]["branch"] = branch or "main"

# Add detected/specified languages
if languages:
    config["languages"] = languages
    config["languages"]["tools"] = tools_by_language

# Add Claude configuration
if marketplaces or plugins:
    config["claude"] = {}
    if marketplaces:
        config["claude"]["marketplaces"] = marketplaces
    if plugins:
        config["claude"]["plugins"] = plugins

# Add network config
if ports:
    config["network"] = {"ports": ports}
elif port_range:
    config["network"] = {"port_range": port_range}

# Add environment variables (non-secrets only)
if env_vars:
    config["environment"] = env_vars

# Shell configuration
config["shell"] = {
    "aliases": True,
    "starship_theme": "red_container",
    "oh_my_zsh_plugins": ["git", "docker", "rust", "python", "node"],
}

Write to File

import tomli_w  # or toml for writing

output_path = Path(sandbox_location) / "Sandbox.toml"

with open(output_path, "wb") as f:
    tomli_w.dump(config, f)

print(f"Configuration saved to {output_path}")

Use scripts/write_config.py helper script.

Cloning Configurations

Clone and Customize Workflow

When user wants to clone an existing sandbox for a new project:

User: "Create a duplicate of ~/Sandboxes/aaronbassett/project1 for myproject2"

Workflow:

1. Read existing config
2. Detect new project languages (if repository provided)
3. Compare and identify differences
4. Ask user about discrepancies
5. Merge preferences
6. Write new config

Example:

# Read existing
existing = read_config("~/Sandboxes/aaronbassett/project1/Sandbox.toml")

# Detect new project (if applicable)
if new_project_repo:
    detected_languages = detect_languages(new_project_repo)

# Compare
existing_langs = set(existing.get("languages", {}).keys())
detected_langs = set(detected_languages.keys())

if existing_langs != detected_langs:
    # Ask user
    print(f"Original sandbox configured for: {', '.join(existing_langs)}")
    print(f"New project uses: {', '.join(detected_langs)}")
    print("Should I configure for new project languages, or keep original?")

    # Wait for user response
    # Then merge accordingly

# Clone base config
new_config = existing.copy()

# Update with new values
new_config["sandbox"]["name"] = "myproject2"
new_config["sandbox"]["location"] = new_location
new_config["sandbox"]["created"] = datetime.now(timezone.utc).isoformat()

# Update languages if changed
if user_wants_new_languages:
    new_config["languages"] = detected_languages_with_versions

# Keep other settings (tools, plugins, etc.)
# unless user specifies changes

Partial Cloning

User may want only specific parts:

User: "Use the same Claude plugins as project1, but different languages"

new_config = {
    "sandbox": {...},  # New sandbox details
    "source": {...},   # New source
    "languages": {...},  # New languages
    "claude": existing_config["claude"],  # Cloned from existing
    "network": existing_config.get("network", default_network),  # Cloned
    "shell": existing_config.get("shell", default_shell),  # Cloned
}

User Preference Merging

Override Priority

When conflicts occur, resolution order:

1. Explicit user request (highest priority)
2. Detected from project files
3. Cloned from existing sandbox
4. User's default preferences (from .claude/sandbox.local.md)
5. Plugin defaults (lowest priority)

Example:

Detected: Python 3.12 (from pyproject.toml)
User says: "use current stable" (3.14.2)

Resolution: Use 3.14.2 (user request overrides detection)
Action: Warn user about mismatch, suggest updating pyproject.toml

Merge Algorithm

def merge_preferences(
    detected: dict,
    user_requested: dict,
    cloned: dict,
    user_defaults: dict,
    plugin_defaults: dict,
) -> dict:
    """Merge preferences with priority."""
    result = plugin_defaults.copy()

    # Layer in each level (lowest to highest priority)
    for source in [user_defaults, cloned, detected, user_requested]:
        for key, value in source.items():
            if value is not None:  # Only override if value provided
                result[key] = value

    return result

Validation

Compatibility Checks

Validate configuration before using:

Language version compatibility:

def check_language_compatibility(config: dict) -> List[str]:
    warnings = []

    # Check if versions are too old
    rust_version = config.get("languages", {}).get("rust")
    if rust_version and rust_version < "1.70":
        warnings.append(f"Rust {rust_version} is quite old, consider updating")

    # Check for known incompatibilities
    # e.g., certain tool combinations

    return warnings

Port conflicts:

def check_port_availability(ports: List[int]) -> List[int]:
    """Return list of ports that are already in use."""
    import socket

    in_use = []
    for port in ports:
        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
            if s.connect_ex(('localhost', port)) == 0:
                in_use.append(port)
    return in_use

Lenient Validation

Be lenient with unknown fields (warn, don't error):

KNOWN_SECTIONS = {"sandbox", "source", "languages", "claude", "network", "environment", "shell", "tools"}

def check_unknown_fields(config: dict):
    unknown = set(config.keys()) - KNOWN_SECTIONS
    if unknown:
        print(f"Warning: Unknown sections in config: {', '.join(unknown)}")
        print("These will be ignored but won't cause errors")

SANDBOX.md Generation

Create Documentation File

Alongside Sandbox.toml, create SANDBOX.md describing the setup:

# Sandbox: {project_name}

Created: {created_timestamp}
Location: {sandbox_location}

## Configuration

**Languages:**
- Rust: {rust_version}
- Python: {python_version}
- Node.js: {nodejs_version}

**Source:**
- Type: {source_type}
- Repository: {repository} (if applicable)

**Claude Code:**
- Marketplaces: {marketplaces}
- Plugins: {plugins}

## Setup

This sandbox was automatically configured based on:
{what influenced the config - detection, cloning, user preferences}

{Any version mismatches or warnings}

## Usage

### Starting the Sandbox

```bash
./sandbox/up.sh

The sandbox will build and start. First run may take 10-15 minutes.

Interactive Shell

./sandbox/shell.sh

Running Commands

./sandbox/run.sh <command>

# Examples:
./sandbox/run.sh cargo test
./sandbox/run.sh npm run dev
./sandbox/run.sh python -m pytest

Stopping

./sandbox/stop.sh

First Run

On first run in the container:

1. Authenticate Claude Code:

   ./sandbox/shell.sh
   cc auth
   # Follow authentication flow
   

2. Verify plugins loaded:

   cc plugin list
   

3. Exit shell:

   exit
   

Authentication persists across container restarts.

Network

Ports forwarded: {ports or port_range}

Access applications:

• http://localhost:{port}

Notes

{Any special notes, warnings, or instructions}

Generate this file automatically when creating sandbox.

## Integration with Other Skills

**With docker-sandbox-setup:**
- This skill defines WHAT to configure
- docker-sandbox-setup implements HOW (Dockerfile, scripts)

**With language-environment-config:**
- This skill stores language preferences
- language-environment-config implements installation

## Scripts

### Available Utilities

- **`scripts/parse_config.py`** - Parse Sandbox.toml to JSON
- **`scripts/validate_config.py`** - Validate configuration
- **`scripts/write_config.py`** - Create Sandbox.toml from dict
- **`scripts/merge_configs.py`** - Merge multiple configs with priority

### Example Usage

```bash
# Parse existing config
python3 scripts/parse_config.py ~/sandbox/Sandbox.toml

# Validate
python3 scripts/validate_config.py ~/sandbox/Sandbox.toml

# Create new config
python3 scripts/write_config.py \
  --name myproject \
  --location ~/Sandboxes/myproject \
  --rust stable \
  --python current \
  --output ~/Sandboxes/myproject/Sandbox.toml

Examples

See examples/ directory for:

• Sandbox.toml.rust - Rust-only project
• Sandbox.toml.python - Python project
• Sandbox.toml.fullstack - Multi-language monorepo

Quick Reference

Read config:

import tomli
with open("Sandbox.toml", "rb") as f:
    config = tomli.load(f)

Write config:

import tomli_w
with open("Sandbox.toml", "wb") as f:
    tomli_w.dump(config, f)

Priority order (highest first):

1. User explicit request
2. Project detection
3. Cloned config
4. User defaults
5. Plugin defaults

Validation: Lenient - warn on unknown fields, error only on critical issues