Security audit
Security checks for vulnerabilities and agentic risk
The skill appears to be a local AgentScope test-harness generator, but it requires an OpenAI API key despite claiming offline, no-API testing.
Review this before installing because it asks for an OpenAI-compatible API key even though the included generator does not use one and the documentation presents the workflow as offline. Prefer installing only if the publisher removes or makes that credential optional, or if you are comfortable exposing the key in this skill's runtime environment.
The skill advertises executable capabilities that include reading files, writing files, and invoking shell commands, but it does not declare any explicit tool scope such as permissions or allowed-tools. That ambiguity weakens least-privilege enforcement and can allow the skill to operate with broader agent runtime powers than users or reviewers would expect.
The skill is described as an offline test harness that avoids external API calls, yet the manifest requires an OPENAI_API_KEY and explicitly centers that credential. This mismatch can mislead users into supplying sensitive credentials to a skill they believe is offline, and it increases the chance of unintended live model calls, data egress, or billing exposure during testing.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
main()
# main uses sys.argv internally; call via script instead
import subprocess
subprocess.run([sys.executable, str(Path(__file__).parent.parent / "scripts" / "generate_harness.py"), "--agent", str(agent), "--output", str(out)], check=True)
assert out.exists()
assert "FakeModel" in out.read_text(encoding="utf-8")
The documentation claims tests are offline by default and that no live API calls are made unless a real model is used, but the manifest still requires a real API key. This inconsistency reduces operator trust, encourages unnecessary credential exposure, and makes accidental online execution more likely in a context where users expect safe local-only testing.
No suspicious patterns detected.