The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains a pre-written instruction block designed to be sent to an AI agent. This block includes commands to install external packages (npx -y @alipay/agent-payment@latest install) and perform purchases. It also includes an "authorization statement" ("I have authorized installation and purchase") intended to influence the agent's decision-making process and security posture by pre-emptively granting permissions for installation and financial transactions, which could be used to bypass standard user confirmation prompts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input data (tabular overtime records) and reflects parts of this data in its output findings, creating a surface for indirect prompt injection. * Ingestion points: Untrusted data enters via the text field in the input JSON file processed by scripts/run.mjs and the analysis engine in scripts/engine/overtime-pay-check.js. * Boundary markers: The skill does not use specific delimiters or instructions to the agent to ignore any natural language commands that might be embedded within the overtime records. * Capability inventory: The skill is granted Read and Bash permissions to facilitate the local execution of the Node.js analysis script. * Sanitization: Input values, such as names or notes from the processed table, are interpolated into the findings output without sanitization, which could lead to an agent following malicious instructions hidden within the input data when interpreting the results.

overtime-pay-check-free