Security audit
Security checks for vulnerabilities and agentic risk
This animation skill needs review because crafted project metadata can overwrite image files outside the promised project folder.
Until destination filenames and final paths are validated, use trusted, self-generated project metadata and restrict the process's filesystem access. Imported asset indexes can otherwise overwrite unrelated images. Cloud voice synthesis sends dialogue to Microsoft; image generation sends prompts and reference images to Volcengine. Review and pin the external app before installation, since downloaded dependencies and the optional voice-cloning engine were not audited here.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
"""
kwargs.setdefault("capture_output", True)
kwargs.setdefault("text", True)
return subprocess.run(list(cmd), **kwargs)
def die(msg, code=1):
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def run(cmd):
"""跑外部命令的唯一入口：列表传参、不经过 shell。"""
p = subprocess.run(list(cmd), capture_output=True)
if p.returncode != 0:
sys.stderr.write(p.stderr.decode("utf-8", "ignore")[-3000:])
raise SystemExit(f"外部命令失败：{' '.join(cmd[:6])}…")
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def run(cmd):
r = subprocess.run(cmd, capture_output=True, text=True)
if r.returncode != 0:
sys.exit(f"命令失败：{' '.join(cmd)}\n{r.stderr}")
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def run(cmd):
r = subprocess.run(cmd, capture_output=True, text=True)
if r.returncode != 0:
sys.exit(f"命令失败：{' '.join(cmd)}\n{r.stderr}")
Without declared permissions the skill's intent is opaque and cannot be validated.
L229 explicitly requires the generated JSON's emotion values to use Chinese emotion names. Although this is a limited metadata field rather than the entire response, the skill provides neither a language opt-in nor an explicit justification for requiring Chinese rather than language-neutral identifiers.
L03 explicitly requires the script’s emotion column to use only the listed Chinese names. Although the document describes engine mappings, it does not explain why users must author this field in Chinese rather than have localized labels normalized internally.