How sandbox-profile.md Maps Tools Across Different Runtime Environments in reverse-skill

The sandbox-profile.md file acts as a canonical lookup table that bridges the capabilities defined in bootstrap-manifest.json with the specific runtime environments—Android devices, Linux hosts, Windows systems, and Docker containers—where those reverse engineering tools execute.

The sandbox-profile.md in the zhaoxuya520/reverse-skill repository serves as a runtime-centric façade over the generic capability list. It translates high-level tool requirements into concrete environment specifications, ensuring that each capability lands in the correct execution context whether you are analyzing Android binaries, web applications, or native code.

Understanding the Runtime Mapping Architecture

The mapping system operates through a bidirectional relationship between two critical files. The skills/scripts/bootstrap-manifest.json defines the technical metadata—name, bootstrapKind, canAutoInstall, and verifyCommand—while skills/ops/sandbox-profile.md organizes these entries by operational context.

The Bootstrap Manifest Connection

In skills/scripts/bootstrap-manifest.json, each tool entry contains the fields necessary for automated installation. For example, frida is defined as a pip-package with canAutoInstall: true and a verifyCommand of frida. The sandbox profile does not duplicate this metadata; instead, it references these capabilities through the "能力 | 典型场景" (Capability | Typical Scenario) table, grouping tools by their target runtime.

Canonical Lookup Table Structure

The primary mapping occurs in the table spanning lines 9-19 of sandbox-profile.md. This section pairs each auto-installable tool with its target environment:

• Android Runtime: jadx, apktool, adb, frida, frida-ps
• Binary CLI Runtime: r2 (radare2), binary analysis utilities
• IDA MCP Runtime: IDA Pro-related tooling
• Web/JS Runtime: Browser-based and JavaScript analysis tools
• Ghidra Runtime: Native Ghidra script execution environment
• Pentest Runtime: Network security assessment tools

Runtime Environment Categories

The sandbox profile organizes tools into distinct runtime silos based on their execution requirements and target platforms.

Android Runtime Environment

Tools listed under the Android row require either a physical device connection via ADB or an emulator instance. The profile places frida alongside jadx and apktool, signaling that these tools operate within the Android runtime context. When bootstrap-reverse.ps1 processes the manifest, it consults this mapping to determine whether to verify Android SDK availability before attempting installation.

Binary CLI and IDA MCP Environments

For native binary analysis, the profile distinguishes between standalone command-line tools and IDA Pro plugin environments. The Binary CLI category encompasses r2 and similar utilities that run directly on the host operating system, while the IDA MCP category identifies tools requiring IDA Pro's specific plugin architecture and runtime hooks.

Web/JS and Ghidra Runtimes

The profile dedicates separate rows to browser-based analysis tools (Web/JS) and Ghidra script execution environments. This separation ensures that JavaScript-heavy analysis workflows do not conflict with Java-based Ghidra operations, each requiring distinct runtime dependencies and JVM configurations.

Handling Non-Auto-Installed Tools

Between lines 25-36, the "Z3r0 sandbox tools not auto-installed" section documents utilities that the upstream Z3r0 image provides but reverse-skill does not provision automatically. Each entry includes a "reverse-skill 策略" (reverse-skill strategy) column describing the provisioning method:

• Manual install: Tools like subfinder that require explicit user intervention
• Kali script: Utilities available through Kali Linux package repositories
• External MCP: Tools requiring Managed Code Provider integration outside the bootstrap process

This distinction tells a sandbox whether the runtime already contains the binary or if additional provisioning steps are required before execution.

Docker Profile Mappings

The sandbox profile provides three lightweight Docker configurations between lines 37-45, translating high-level use cases into specific container requirements.

Minimal, Mobile, and Reverse Engineering Profiles

The table defines three concise profiles:

1. 最小 (Minimal): Core utilities for basic operations
2. 移动 (Mobile): Android-specific tooling including jadx, apktool, and frida
3. 逆向 (Reverse Engineering): Comprehensive binary analysis suites

These profiles imply the underlying Docker image must include the relevant runtime libraries. For example, selecting the 移动 profile requires a container with Android SDK binaries and USB debugging capabilities.

Network Policy Integration

Between lines 49-55, the "network_profile 联动" (network_profile linkage) section enforces that container-based scanning respects the case scope's network_profile. This ties the runtime's network access level—offline versus authorized_target_only—to the allowed toolset.

Certain tools listed in the capability table trigger network restrictions automatically. The profile maps sandbox-profile modes to network policies, ensuring that nmap or nuclei operations execute only within authorized network boundaries.

Practical Implementation Examples

Querying the mapping system and translating profiles into actionable commands requires interaction with both the manifest and the profile documentation.

Querying Tool Runtime Assignments

To determine which runtime environment a specific tool requires, extract the mapping directly from the manifest and cross-reference the profile:

# PowerShell: Extract runtime category for frida from the bootstrap manifest

$manifest = Get-Content "$env:SKILL_ROOT\skills\scripts/bootstrap-manifest.json" | ConvertFrom-Json
$tool = $manifest.capabilities | Where-Object { $_.name -eq 'frida' }
Write-Host "Tool: $($tool.name) – Runtime: Android (implied by sandbox-profile table)"

Building Docker Commands from Profiles

The lightweight profiles translate directly into container orchestration commands:

# Bash: Assemble a Docker run command for the "移动" (Mobile) profile

TOOLS="jadx apktool frida"
docker run -it --rm \
  -v "$HOME/Tools:/opt/tools" \
  ghcr.io/zhaoxuya520/reverse-skill:mobile \
  /bin/bash -c "install-tools $TOOLS && start-reverse-session"

This command references the 移动 profile, automatically including the Android runtime dependencies specified in skills/ops/sandbox-profile.md.

Summary

• sandbox-profile.md serves as the authoritative mapping between bootstrap-manifest.json capabilities and their target runtime environments.
• The "能力 | 典型场景" table (lines 9-19) categorizes tools into Android, Binary CLI, IDA MCP, Web/JS, Ghidra, and Pentest runtimes.
• Z3r0 sandbox tools (lines 25-36) document manually-provisioned utilities with specific installation strategies.
• Docker profiles (lines 37-45) provide minimal image specifications for 最小, 移动, and 逆向 use cases.
• Network profile linkage (lines 49-55) enforces case-scope network restrictions on tool execution.
• The bootstrap-reverse.ps1 and refresh-tool-index.ps1 scripts consume these mappings to create reproducible environments.

Frequently Asked Questions

What is the relationship between bootstrap-manifest.json and sandbox-profile.md?

The bootstrap-manifest.json file provides the technical specification for each tool—including installation methods and verification commands—while sandbox-profile.md organizes these capabilities by runtime context. The manifest answers "how do we install this tool," while the profile answers "where does this tool run."

How does the sandbox profile determine which runtime to use for a tool?

The profile relies on the "能力 | 典型场景" table that pairs each capability with its typical operational scenario. For example, frida appears alongside apktool and jadx under the Android scenario, indicating that the Android runtime—whether a device or emulator—is required for its operation.

Can I use sandbox-profile.md without the auto-bootstrap system?

Yes. The profile functions as documentation for manual environment configuration. The Z3r0 sandbox tools section (lines 25-36) specifically addresses tools that are not auto-installed, providing manual installation strategies for Kali Linux, external MCPs, or direct package management.

How do network profiles restrict tool usage in sandbox environments?

The network_profile 联动 section (lines 49-55) defines which sandbox-profile modes are compatible with offline versus authorized_target_only network scopes. When a container initializes, the system checks this mapping to ensure that network-scanning tools like nmap or nuclei execute only within permitted network boundaries, preventing accidental scanning of unauthorized targets.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}