A critical OpenClaw security policy gate that limits agent risk, isolates actions, verifies outcomes, and enables safe recovery.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install sandbox-blast-radius-control-engine
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install sandbox-blast-radius-control-engine using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Sandbox & Blast-Radius Control Engine X∞?
Sandbox & Blast-Radius Control Engine X∞ is a distributed runtime security layer for OpenClaw. It sits between an agent’s intent and tool execution, requiring every medium-, high-, and critical-risk action to pass through assess, authorize, isolate, execute, verify, and recover or rollback stages.
This Openclaw Skills security engine applies least privilege, risk classification, dry-run planning, change previews, backup verification, atomic operations, audit trails, observability, and explicit confirmation for destructive or production actions. It helps agents remain powerful while ensuring each action has a defined scope, authorization boundary, verification method, and recovery path.
Sandbox & Blast-Radius Control Engine X∞ Use Cases
- Safely deleting or modifying important files, especially in production or system directories.
- Running database updates, deletes, drops, truncations, or migrations with appropriate controls.
- Deploying configuration or application changes to production environments.
- Executing shell commands, installation scripts, builds, or package operations without blind privilege escalation.
- Testing changes in sandboxes, temporary directories, separate branches, staging systems, or test databases.
- Protecting API keys, credentials, secrets, and sensitive data from logs, traces, prompts, and external content.
- Defending against prompt injection and untrusted instructions from websites, repositories, email, plugins, documents, and uploaded files.
- Managing long-running or low-resource workflows on Termux and Android with bounded retries, timeouts, concurrency, and checkpoints.
- Handling incidents involving secret leaks, data loss, unauthorized access, corruption, or privilege escalation.
- Providing Openclaw Skills workflows with traceable success, failure, rollback, recovery, and evaluation outcomes.
How Sandbox & Blast-Radius Control Engine X∞ Works
- Detect the trigger: Activate as a pre-flight gate for medium, high, and critical actions, or when users mention sandboxing, risk, rollback, production, secrets, destructive commands, or verification.
- Collect context: Identify the environment, target, explicit authorization, reversibility, backup status, content origin, available resources, and potential symlink or path hazards.
- Classify risk: Assign LOW, MEDIUM, HIGH, CRITICAL, or UNKNOWN based on the action and its potential impact.
- Minimize scope: Select the smallest feasible blast radius, from local file or project scope through service, account, system, or global scope.
- Select the least-privileged tool: Match tool capability, permission, and risk; never escalate to sudo, root, or admin access merely because a command failed.
- Isolate the operation: Prefer a sandbox, temporary directory, separate workspace, Git branch, staging environment, or test database instead of production or the main branch.
- Plan and preview: Use a dry run or
--planoption when available; otherwise preview what will change, why, where, expected results, risks, and rollback steps. - Protect recoverability: Create and verify backups for high- and critical-risk actions, and create checkpoints for long-running tasks.
- Execute atomically: Prefer
TEMP → VERIFY → MOVE/REPLACEover deleting the old state before creating a replacement. - Verify actual state: Check the real target, resulting state, dependencies, smoke-test behavior, security exposure, side effects, and rollback availability rather than trusting an exit code.
- Trace and evaluate: Emit START, ACTION, ERROR, RETRY, SUCCESS, FAILURE, and ROLLBACK events with risk and blast-radius metadata, excluding secrets.
- Commit or recover: Use bounded retries, an alternative method, rollback, checkpoint restoration, safe stop, or explicit user intervention according to the recovery hierarchy.
- Report and exit: Stop only after verified success, verified restoration, user cancellation, or incident isolation and reporting.
Sandbox & Blast-Radius Control Engine X∞ Setup
Installation and activation
This skill is a runtime policy and guardrail layer, not a standalone command-line tool or executable package. The specification does not define a package manager command, environment variable set, or separate daemon. Install it through the standard OpenClaw skill-loading mechanism as a workspace-level security skill, using the supplied SKILL.md metadata and policy content.
Configuration checklist
- Keep the skill active and available as a pre-flight gate for MEDIUM, HIGH, and CRITICAL actions.
- Preserve the default risk level of
CRITICALfor the guardrail itself. - Configure or connect
agent-observability-trace-enginefor trace events. - Configure or connect
agent-evaluation-benchmark-enginefor success, failure, regression, recovery, and performance evaluation. - Ensure tools expose or support capability, permission, risk, dry-run, timeout, and rollback information where possible.
- Define workspace, staging, temporary, backup, checkpoint, and production boundaries before enabling destructive workflows.
- On Termux or Android, account for RAM, storage, process lifetime, background execution, battery, network, architecture, and permission limits.
Operational modes
- LIGHT: LOW-risk, local, read-only, reversible work with minimal trace volume.
- STANDARD: MEDIUM-risk changes with preview, standard controls, and dry-run support where available.
- STRICT: HIGH- and CRITICAL-risk work requiring tight scope, backup or checkpoint controls, verification, and confirmation where applicable.
No action should be considered complete until the actual state has been verified.
Sandbox & Blast-Radius Control Engine X∞ Data Schema & Taxonomy
Skill metadata
| Field | Value |
|---|---|
| Name | Sandbox & Blast-Radius Control Engine X∞ |
| Slug | sandbox-blast-radius-control-engine |
| Version | 1.1.0 |
| Type | Security / runtime policy |
| Risk level | Critical |
| License | MIT |
| Status | Active |
| Compliance | X∞ 21-node mandatory structure v1.0.0 |
| Principles | 15 |
| Owner | pmuhammadagus-byte |
Risk and scope taxonomy
- Risk: LOW, MEDIUM, HIGH, CRITICAL, UNKNOWN.
- Blast radius: LOCAL, PROJECT, SERVICE, ACCOUNT, SYSTEM, GLOBAL.
- Execution mode: LIGHT, STANDARD, STRICT.
- Action lifecycle: ASSESS, AUTHORIZE, ISOLATE, EXECUTE, VERIFY, RECOVER or ROLLBACK.
- Tool classes: read/file-fetch/web_fetch as LOW; write and build/install execution as MEDIUM; production configuration, database updates, and important deletes as HIGH; database drops, production deletes, secret access, and irreversible actions as CRITICAL.
Audit and checkpoint records
Important operations should record who, what, when, target, authorization, result, and rollback without storing secrets. Long-running work should checkpoint task, progress, completed, remaining, dependencies, and last-safe-state.
Observability event schema
Emit START, ACTION, ERROR, RETRY, SUCCESS, FAILURE, and ROLLBACK events. Each event should include risk_level and blast_radius; credentials and secrets must never be included in logs, memory, output, or traces.
Sandbox & Blast-Radius Control Engine X∞ Advanced Features
- 21-node X∞ compliance structure with a canonical 51-section security specification.
- Automatic pre-flight gating for medium-, high-, and critical-risk agent actions.
- Positive and negative trigger detection, including safe stops for policy-bypass and prompt-injection attempts.
- Adaptive LIGHT, STANDARD, and STRICT control modes based on risk and available resources.
- Least-privilege tool selection with capability, permission, and risk mapping.
- Blast-radius optimization across local, project, service, account, system, and global scopes.
- Dry-run,
--plan, manual preview, backup verification, checkpointing, and atomic replacement patterns. - Explicit confirmation requirements for production deletion, database destruction, irreversible changes, sensitive-data exposure, and account-level destructive actions.
- File, path, database, network, plugin, credential, skill, and agent-handoff boundaries.
- Bounded retry loops, timeouts, adaptive concurrency, alternative methods, safe-stop behavior, and user intervention.
- Incident response flow: STOP → ISOLATE → PRESERVE EVIDENCE → REPORT → RECOVER.
- Integration with observability tracing and agent evaluation contracts without leaking secrets.
- Termux and Android low-resource optimization that reduces trace volume, concurrency, retries, memory use, and background work.
- Semver-based policy versioning with changelog expectations and non-breaking guidance improvements in version 1.1.0.
- Self-audit prompts that continually assess authorization, safer alternatives, blast-radius reduction, rollback readiness, and verification strategy.
SKILL.md
Loading
Related Openclaw Skills
AI Content Disclosure Check – AI-HIVE
Turn AI-generated content disclosure requirements into an auditable review plan, labeling strategy, metadata ledger, and pre-publication checklist.
AI Unboxing Video Generator - AI-HIVE
Turn authorized product photos, packaging details, and launch goals into reviewable AI unboxing video plans, scripts, generation tasks, and validated deliverables.
AI Literacy Expert V8.2-AIPC
An AI literacy teaching and lesson-planning skill that transforms course materials into privacy-aware, interactive, and discipline-specific learning experiences.
AI Ecommerce Expert: Etsy Image and Video Content
An AI-powered IMIVA MCP workflow for creating conversion-focused Etsy product images, detail pages, videos, and marketing content.
Livestream Highlight Clips AI-HIVE
Turn authorized livestream replays into evidence-based highlight timestamps, short-video scripts, subtitles, covers, and publish-ready vertical clips.
Token Hub Alternative & AI-HIVE Migration Guide
A practical AI-HIVE migration and Token Hub alternative skill for auditing model compatibility, routing workloads, validating costs, and safely scaling image, video, and commerce generation.
Featured*
