strix-pentest
Audited by Socket on Oct 27, 2026
5 alerts found:
Securityx3Anomalyx2SUSPICIOUS: the skill’s behavior matches its stated pen-testing purpose, but that purpose is itself a high-risk offensive capability for an AI agent. The main concerns are autonomous scanning/exploitation, explicit permission-bypass guidance, and installing executable tooling from a personal GitHub repo outside an official package channel. This is not confirmed malware, but it is a high-risk security skill that should not be treated as benign developer guidance.
No clear evidence of stealthy malware (no obfuscation, no exfiltration to hardcoded domains, no persistence), but the module is a high-privilege remote execution tool server: it can run arbitrary bash commands in a PTY, execute arbitrary Python code, edit/write files under /workspace, perform arbitrary outbound HTTP requests, and evaluate arbitrary browser JavaScript. The security model relies entirely on the bearer token; if misconfigured or compromised, this becomes extremely dangerous.
No direct evidence of stealthy malware (no obfuscation, no local persistence, no explicit exfiltration) is present in the provided fragment. However, the module is architected as a high-privilege proxy that forwards largely unvalidated, caller-controlled tool execution requests (including shell/Python/JS execution and arbitrary HTTP/filesystem operations) to an environment-configured remote endpoint using a bearer token. If the remote tool server is malicious/compromised or if an attacker can reach the MCP tool interface or influence tool parameters/configuration, this component can enable serious compromise. The snippet is also incomplete, so filesystem safety and authorization controls cannot be confirmed.
No definitive evidence of classic malware (e.g., backdoors, data-stealing routines, persistence) is visible in the provided fragment. However, the module is a high-risk orchestration component: it launches an external LLM CLI with explicit permission-bypass flags and forwards all process environment variables to that subprocess, while embedding an exploit-focused instruction prompt. Even if intended for sanctioned pentesting, this design increases the likelihood of unsafe actions, secret exposure, and supply-chain/control-plane abuse. Full confidence requires reviewing the rest of the module(s) that construct mcp_config_path, initial_prompt, sandbox enforcement, and how tool permissions/MCP access are actually restricted.
No explicit malicious payload (no exfiltration/backdoor/cryptomining/reverse shell) is evident in this fragment. However, there are meaningful supply-chain and blast-radius risks: (1) installs trivy via `curl ... | sh` inside the container (supply-chain execution risk), and (2) optionally mounts the host Docker socket and sets it to 666, giving the container near-unrestricted Docker control (major privilege escalation risk if the sandbox can be influenced). Overall, treat as a high-risk sandbox/orchestration component requiring strong trust boundaries and supply-chain controls (pinning/verifying installers, restricting image/local_sources).