Security audit
Security checks for vulnerabilities and agentic risk
This skill performs a proportionate local check of business-license details against bundled filing rules, with optional OCR dependencies that require separate installation.
If you need the OCR fallback, install reviewed dependency versions in an isolated environment; the supplied requirements are not a lockfile. OCR returns the license's recognized text to your agent. Confirm extracted fields and treat results as preliminary screening rather than filing approval.
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
ordered_categories = dict(
sorted(categories.items(), key=lambda pair: (pair[1].order, pair[0]))
)
return RuleBook(
schema_version=schema_version,
version=version,
checked_at=checked_at,
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
)
rules_path = Path(args.rules) if args.rules is not None else references / "huawei-rules.json"
args.rules = str(rules_path)
return rules_path
def main(argv: Optional[Sequence[str]] = None) -> int:
Without declared permissions the skill's intent is opaque and cannot be validated.
The open-ended requirement permits future, unreviewed onnxruntime releases, creating an installation-time supply-chain and reproducibility weakness rather than demonstrating a currently exploitable package vulnerability. The comment is misleading: a minimum version neither pins a patched release nor prevents the separate onnx package from appearing transitively; separate, optional installation limits exposure during normal skill execution.
# Pin onnxruntime to latest patched version (CVE-2026-14647 affects
# the "onnx" package, NOT "onnxruntime"; pinning here prevents
# accidental inclusion of the "onnx" package via transitive deps).
onnxruntime>=1.19.0
Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.
}
terms = []
for item in result.matches:
terms.extend(getattr(item, category_attribute))
for item in result.province_rules:
rule_terms = getattr(item, rule_attribute)
if national_category_ids.intersection(item.rule.category_ids):
Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.
for item in result.matches:
terms.extend(getattr(item, category_attribute))
for item in result.province_rules:
rule_terms = getattr(item, rule_attribute)
if national_category_ids.intersection(item.rule.category_ids):
rule_terms = tuple(
term for term in rule_terms if term not in BROAD_PROVINCE_TERMS
Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.
)
terms.extend(rule_terms)
for item in result.province_policies:
terms.extend(getattr(item, rule_attribute))
return tuple(dict.fromkeys(terms))
Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.
dict.fromkeys(
term
for match in result.province_rules + result.province_policies
for term in getattr(match, attribute)
)
)
Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.
multiple_paths = len(row.paths) > 1
rendered = []
for path in row.paths:
values = getattr(path, attribute)
if isinstance(values, str):
values = (values,)
for value in values:
Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.
raise RuleError("{}过大".format(label))
flags = os.O_RDONLY
for name in ("O_CLOEXEC", "O_NOFOLLOW", "O_NONBLOCK"):
flags |= getattr(os, name, 0)
descriptor = os.open(str(path), flags)
opened = os.fstat(descriptor)
if not stat.S_ISREG(opened.st_mode):