construction-output-value-check-free
Fail
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: HIGHPROMPT_INJECTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The
SKILL.mdfile includes a block of text designed to be processed by the agent that contains directives to automate sensitive operations. It uses language such as "我已授权安装依赖和发起购买" (I have authorized installing dependencies and initiating purchases) and "不要替我确认" (do not confirm on my behalf) to attempt to steer agent behavior and bypass confirmation steps during purchases and software installations. - [REMOTE_CODE_EXECUTION]: The documentation instructs the agent to execute
npx -y @alipay/agent-payment@latest install. This command fetches and executes code from the registry of the Alipay organization, which is a well-known service, to facilitate payment functionalities. - [INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary user-provided text for verification checks within the
runfunction inscripts/engine/construction-output-value-check.js. - Ingestion points: The
textproperty of the input payload passed to therunfunction inscripts/engine/construction-output-value-check.js. - Boundary markers: The skill does not implement specific boundary markers or instructions to the agent to treat the table data as passive content only.
- Capability inventory: The skill uses
BashandReadtools to execute local Node.js logic and read construction records. - Sanitization: The code performs numerical extraction and string splitting but does not sanitize the input text to ensure embedded instructions cannot influence the agent's interpretation of the analysis results.
Recommendations
- AI detected serious security threats
Audit Metadata