The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the qianwen and alipay-bot CLI tools for sensitive operations including account authentication, balance summaries, and recharge order creation. These operations are performed using structured argument arrays to prevent shell injection, and critical transaction steps require explicit user confirmation with a mandatory AI risk notice.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of official Node.js packages @qianwenai/qianwen-cli and @alipay/agent-payment via NPM. These resources originate from the skill author and a well-known payment service provider, and their provenance is documented for manual verification.
[DYNAMIC_EXECUTION]: The skill employs a custom resolver to locate executables on the system PATH. This utility includes defensive logic to exclude the current Git worktree, which prevents the agent from being tricked into running malicious local binaries. Additionally, the skill uses a restrictive regular expression (SAFE_ARG_RE) to validate arguments whenever Windows shell execution is required.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as recharge status reports and payment URLs. To prevent data poisoning, the skill implements strict schema validation and a hostname allowlist for payment links (e.g., cashier.alipay.com), ensuring the agent only interacts with verified endpoints.
Ingestion points: Recharge order creation responses and periodic status polling results processed in scripts/preflight.mjs and scripts/poll-recharge.mjs.
Boundary markers: The instructions include explicit rules against improvising payment links or using data from untrusted sources like user history or logs.
Capability inventory: Execution of CLI tools via spawn and spawnSync is utilized to interact with the underlying account services.
Sanitization: Amount normalization logic and JSON shape validators are used to ensure that all transaction data is correctly formatted and matches confirmed anchors.

qianwen-payment