The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file includes a text block intended for the user to send to the AI assistant. This block contains the phrase '我已授权安装依赖和发起购买' (I have authorized the installation of dependencies and the purchase). This is a prompt injection technique designed to bypass the agent's standard confirmation protocols for sensitive operations like package installation and financial transactions.\n- [EXTERNAL_DOWNLOADS]: The documentation references the installation of the @alipay/agent-payment package via npx. This involves downloading and executing code from the official Alipay registry, which is a well-known and established service provider.\n- [REMOTE_CODE_EXECUTION]: The skill suggests the execution of remote code via the npx -y @alipay/agent-payment@latest install command. This command downloads and runs a package from a remote registry. As the package belongs to a well-known service provider, this is documented as a functional requirement.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data (cash flow tables) through the scripts/run.mjs entry point.\n
Ingestion points: Data is read from files or direct text input in scripts/run.mjs and processed by the run function in scripts/engine/cash-flow-variance-check.js.\n
Boundary markers: No explicit delimiters or instructions are used to separate user data from the agent's internal logic.\n
Capability inventory: The skill has the capability to read files from the local file system and execute shell commands via the Bash tool.\n
Sanitization: The processing engine employs strict numeric parsing (normNumber) to extract data, which significantly mitigates the risk of the agent interpreting the input as instructions.

cash-flow-variance-check-free