The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes advertising rebate tables provided as text, which involves ingesting untrusted data into the agent's context.
Ingestion points: The scripts/run.mjs wrapper loads data from files specified by the user or from built-in samples.
Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" markers for the processed text data.
Capability inventory: The skill utilizes the Bash and Read tools to execute a Node.js engine that parses and analyzes the input.
Sanitization: No specific sanitization or filtering is applied to the input text to prevent instructions embedded within the spreadsheet data from influencing the agent.
[PROMPT_INJECTION]: The SKILL.md file contains a block of text intended to be used by the agent to facilitate a purchase, which uses language patterns characteristic of injection.
Evidence: The text includes phrases such as "我已授权安装依赖和发起购买" (I have authorized installation of dependencies and initiated purchase) which attempts to direct the agent's execution flow and authorization logic.
[EXTERNAL_DOWNLOADS]: The documentation references an external package installation to enable premium features.
Evidence: npx -y @alipay/agent-payment@latest install is provided in the manual upgrade instructions.
Context: The package originates from the official @alipay scope on NPM, representing a well-known technology company.
[REMOTE_CODE_EXECUTION]: The skill instructs the agent/user to perform remote script execution during the upgrade process.
Evidence: The use of npx with the @alipay/agent-payment package fetches and executes code from the NPM registry.
[DYNAMIC_EXECUTION]: The wrapper script dynamically imports the processing logic at runtime.
Evidence: scripts/run.mjs uses createRequire and path.join to load the engine from scripts/engine/ad-agency-rebate-check.js.

ad-agency-rebate-check-free