Analyze codebases to generate optimal Claude Code Sandbox configurations. Use this skill when users need to set up sandbox security settings for their projects. This skill should be triggered when users ask about sandbox configuration, security settings, or when setting up Claude Code for a new project. It analyzes the codebase stack (Node.js, Python, Rust, Go, PHP, etc.), detects dependencies, and generates appropriate sandbox and permission settings through an interactive Q&A process.
npx claudepluginhub ohare93/claude-setup --plugin jmo-development-toolsThis skill uses the workspace's default tool permissions.
This skill analyzes a codebase to understand its technology stack, architecture, and dependencies, then generates tailored Claude Code Sandbox configurations. The skill uses an interactive question-and-answer approach to achieve 95% confidence before generating the final configuration.
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
This skill analyzes a codebase to understand its technology stack, architecture, and dependencies, then generates tailored Claude Code Sandbox configurations. The skill uses an interactive question-and-answer approach to achieve 95% confidence before generating the final configuration.
Before asking questions, analyze the codebase automatically:
Run the analysis script to detect the stack:
python3 scripts/analyze_codebase.py
If the script is unavailable, manually check for:
package.json → Node.js/JavaScriptrequirements.txt, pyproject.toml, setup.py → PythonCargo.toml → Rustgo.mod → Gocomposer.json → PHPGemfile → Rubypom.xml, build.gradle → Java*.csproj → .NETDockerfile, docker-compose.yml → Container-basedIdentify key patterns:
Ask clarifying questions until 95% confident. Number each question and provide labeled options.
**[N]/ [Question Title]**
[Brief context about why this matters]
Options:
- **a)** [Option description]
- **b)** [Option description]
- **c)** [Custom input option if applicable]
**Recommendation**: [Your recommended choice] - [Reason why]
1/ Security Posture
Different projects require different security levels based on their context.
Options:
Recommendation: b) Standard Development - Provides good security while maintaining productivity. Switch to c) if handling credentials or untrusted code.
2/ Development Server Usage
Dev servers need to bind to localhost ports (especially on macOS).
Options:
Recommendation: Based on detected framework. If web framework detected, recommend a).
3/ Container/Docker Usage
Docker commands are incompatible with sandbox and need special handling.
Options:
Recommendation: If Dockerfile detected, recommend a) and add to excludedCommands.
4/ Git Operations Preference
Git push/commit can be sensitive operations.
Options:
Recommendation: b) - Allows routine git status/diff/log but prompts before changing history.
5/ SSH Agent Access (Linux/macOS)
Required for SSH-based git authentication.
Options:
Recommendation: If .git/config shows SSH remote URLs, recommend a).
6/ Sensitive Files to Protect
Beyond the defaults (.env, secrets), identify project-specific sensitive files.
Options:
If b) or c) selected, ask: "Please list the file patterns to protect (e.g., config/production.yml, *.key)"
Recommendation: a) unless detected files like credentials.py, *.pem, *.key.
7/ Package Registry Access
Network access needed for installing dependencies.
Options:
Recommendation: Based on detected package manager and any .npmrc/.pypirc files.
8/ Escape Hatch Behavior
When a command fails in sandbox, should Claude retry outside sandbox?
Options:
Recommendation: a) for new projects (learning phase), b) for established configs.
9/ Auto-Allow Sandboxed Commands
Should sandboxed bash commands run without prompting?
Options:
Recommendation: a) for productivity, b) when learning Claude Code or auditing behavior.
10/ Additional Tools/Commands (if applicable)
Based on codebase analysis, ask about specific tools detected:
After collecting answers, generate the configuration by:
references/stack-templates.md for the base template// .claude/settings.json
// Generated by Sandbox Architect for [detected stack]
{
"sandbox": {
// [Comment explaining each setting]
},
"permissions": {
// [Comment explaining each rule]
}
}
After generating the config:
Every question must include:
For complex scenarios, reference references/sandbox-guide.md which contains:
analyze_codebase.py - Automatic stack and dependency detectionsandbox-guide.md - Complete Claude Code Sandbox documentationstack-templates.md - Pre-built configurations for common stacksReach 95% confidence before generating configuration. Factors that increase confidence:
If confidence is below 95%, continue asking clarifying questions about uncertain areas.