The Agent Skills Directory

[PERSISTENCE]: The documentation in references/onboarding.md provides instructions for users to persist their API keys by appending export commands to shell configuration files such as ~/.zshrc and ~/.bashrc. While this is a standard configuration practice for development tools, it constitutes persistent storage of sensitive credentials in plain text.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external 1688 API endpoints and possesses the capability to perform significant write operations (order creation, cancellation, etc.), creating an attack surface for indirect prompt injection if external fields contain malicious instructions.
Ingestion points: Data is entering the context via scripts/sku.py, scripts/order_status.py, scripts/logistics.py, and scripts/invoice_amount.py which fetch product and order details from 1688.
Boundary markers: The instructions in SKILL.md and references/workflow.md include warnings for the agent to require separate confirmation for high-risk actions, but the skill does not use specific delimiters or "ignore instructions" warnings for the interpolated API data itself.
Capability inventory: The skill can execute high-risk operations via create_order.py, payment_url.py, confirm_receive.py, cancel_order.py, and invoice_apply.py.
Sanitization: The _alibaba1688_common.py script implements a redact function that masks sensitive keys (e.g., tokens, secrets) in responses before they are printed to stdout or saved to the workspace.

linkfox-1688-procurement