controlclaw-firewall

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill requires sudo (root) access to configure system-wide network filtering via nftables, modify DNS settings, and manage system services. It includes explicit warnings to the user about the risks of running with root and instructs the agent to ask for permission before using sudo.
  • [COMMAND_EXECUTION]: The skill executes multiple shell commands to verify the environment (uname, node, docker) and perform the installation/connection tasks. It uses complex command patterns, such as file redirection combined with immediate deletion, to securely handle sensitive inputs.
  • [EXTERNAL_DOWNLOADS]: The skill uses npx to download and run the controlclaw package directly from the npm registry. The package name corresponds to the skill author's known repository.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data in the form of pairing codes and firewall addresses. It implements a robust evidence chain to mitigate risks:
  • Ingestion points: Pairing codes and connection addresses provided by the user (SKILL.md).
  • Boundary markers: The agent is instructed to ask for explicit user consent before performing installations or joining remote firewalls.
  • Capability inventory: The skill has the capability to execute commands with sudo, download external packages, and modify system network configurations.
  • Sanitization: To protect pairing codes, the skill writes them to a restricted file (umask 077, chmod 700) and uses shell redirection (<) to pass the data to the tool's stdin, preventing the code from appearing in public process lists (ps).
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 09:12 AM