Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
arc-skill-sandbox 1.1.0 - Updated scripts/sandbox.py (details not included). - No user-facing documentation changes.
What This Skill Does
Runs untrusted skills in a monitored environment before installation. Tracks filesystem access, environment variable reads, network connections, and subprocess calls during execution. Produces a JSON report with a safety verdict (SAFE / SUSPICIOUS / DANGEROUS).
Static analysis misses runtime behavior, so executing the skill in a monitored environment reveals what it actually does with your data and credentials.
When to use it
- Testing a ClawHub skill before installing it on your real agent
- Checking if a downloaded script reads API keys or tokens
- Catching outbound network calls made by an unknown skill
- Running a skill with fake credentials to detect exfiltration attempts
- Generating a safety report before a team-wide skill rollout
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: sandbox this skill at ~/.openclaw/skills/some-skill/ and tell me if it's safe to install
- 1Runs sandbox.py against the skill directory with --monitor-network and --fake-env flags
- 2Captures all filesystem operations, environment variable accesses, and outbound network requests
- 3Injects fake API keys and tokens to observe whether the skill attempts to exfiltrate them
- 4Generates a JSON report listing all observed activity and flags suspicious patterns
- 5Returns the safety verdict with a summary of findings
JSON safety report covering filesystem, network, subprocess, and env var activity with a final SAFE / SUSPICIOUS / DANGEROUS verdict