Security audit
Security checks for vulnerabilities and agentic risk
The skill coherently prepares 1688 product drafts, with scoped uploads and user-controlled final publication.
Use an authorized 1688 store and provide only materials you may share with its services. Review generated prices, inventory, specifications and claims before publishing. Local source copies and task records remain until you remove them.
Without declared permissions the skill's intent is opaque and cannot be validated.
These steps actively instruct URL conversion and uploading complete Base64-encoded image files, rather than merely listing available tools. Although the upload mechanics are described, the file does not warn that image contents or source URLs will be disclosed to an external service, which matters for private or confidential images.
The submission parameters include product image URLs, user descriptions, source materials, and SKU pricing and inventory; step 2 at L64 instructs the agent to submit the request. The file does not warn users that these potentially confidential materials are sent to an external service, so this is an operational instruction rather than merely a tool listing.
The run helper executes the supplied command via subprocess.run, but provides no confirmation, user-facing execution log, or explanatory docstring or comment. The module-level description mentions local-file helpers, which does not disclose external command execution; captured output is returned rather than visibly logged.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def run(argv, timeout=60):
result = subprocess.run(argv, capture_output=True, text=True, timeout=timeout)
if result.returncode:
raise ValueError((result.stderr or result.stdout or "command failed")[-1500:])
return result.stdout