Security audit
Security checks for vulnerabilities and agentic risk
This looks like a test-runner skill, but it can run any supplied command and gives unsafe sudo/admin retry advice, so it needs review before installation.
Install only if you are comfortable with the agent running exact commands you provide. Use simple project-local test commands, review the command before execution, do not pass untrusted text as the command, and do not rerun failures with sudo or administrator rights. Treat expected_behavior as a note only; the script checks exit code, not semantic behavior.
The skill enables shell command execution via a user-supplied test_command, but it does not declare any explicit tool scope or command restrictions. In an agent environment, this creates a permission ambiguity that can lead to overbroad execution of arbitrary commands under the guise of test verification.
The activation condition, 'When you need to verify functionality after code changes,' is broad enough to match many ordinary development requests. That increases the chance the skill is invoked in situations where executing shell commands was not explicitly intended, enabling unsafe or surprising command execution paths.
The skill advertises autonomous execution of commands based on user input, which delegates a sensitive decision—what code to run—to the skill without meaningful guardrails. In context, this is more dangerous because the skill's sole purpose is to launch subprocesses, making command execution the primary behavior rather than an incidental one.
## Usage
Provide test command, script will automatically execute and check results.
## Parameters
The usage text says the script will 'automatically execute' a provided test command, but it does not define safety checks, approval requirements, or command limitations. Because the parameter is free-form, this ambiguity can be exploited to run arbitrary shell commands rather than merely safe test invocations.
Natural-language strings throughout the docstrings, help text, and console output are exclusively in Chinese. For an all-purpose code verification utility, this imposes a language/locale constraint without user opt-in or a documented region-specific justification.
The function documents and prints expected_behavior but never actually validates program output or runtime behavior against it, so success is determined only by exit code. This can produce false assurances that a code change is correct even when the observed behavior contradicts expectations, which is especially risky in automated verification pipelines.
The script accepts an arbitrary --test-command string and executes it as a subprocess, which gives the skill a general command-execution capability rather than a narrowly scoped code-verification function. In an agent/skill context, this is dangerous because upstream prompts or untrusted inputs could cause execution of destructive commands, data exfiltration tools, package installers, or other system-impacting binaries under the user's privileges.
The presence of a direct suggestion to use sudo is security-relevant in this context because the script executes arbitrary commands and may prompt operators toward privileged reruns. If followed, the same command-execution pathway could operate with root/admin permissions, turning otherwise limited mistakes into full host compromise or destructive changes.
print(f" 命令: {test_command}")
print("\n建议:")
print(" 1. 检查文件/目录权限")
print(" 2. 尝试使用 sudo (Linux/macOS)")
print(" 3. 以管理员身份运行 (Windows)")
return False
except Exception as e:
When a permission error occurs, the tool recommends retrying with sudo or administrator privileges without warning about the security implications. In a skill that already executes user-supplied commands, encouraging privilege escalation increases the chance that unsafe or overly broad commands will be rerun with elevated rights, amplifying any accidental or malicious system impact.