The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains a dedicated section for a "Buyout Edition" that provides a specific block of text for the user to send to their AI assistant. This block uses authoritative phrasing such as "I have authorized the installation of dependencies and initiation of purchases" (我已授权安装依赖和发起购买). This technique is designed to influence the agent's internal state to believe that user consent has already been granted for potentially sensitive actions like installing external software.
[REMOTE_CODE_EXECUTION]: The upsell instructions directed at the AI agent include the command npx -y @alipay/agent-payment@latest install. This command downloads and executes code from the npm registry. While targeting a well-known payment provider (Alipay), the inclusion of automated installation commands inside prompt-based instructions represents a significant security risk if the agent executes them without additional verification.
[METADATA_POISONING]: The skill uses its description and SKILL.md documentation to host instructions that are not merely informative but are designed to be executed by the agent to perform multi-step installations and purchases of other skills and services.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it analyzes untrusted user-provided text for advertising law compliance.
Ingestion points: User input is ingested via the text or texts fields in scripts/run.mjs and processed by scripts/engine/ad-compliance.js.
Boundary markers: The skill does not implement specific boundary markers or "ignore" instructions when processing external text.
Capability inventory: The skill is configured with access to Bash and Read tools.
Sanitization: The input is analyzed using static regular expressions and string matching within a local Node.js environment. No evidence of direct command injection or dynamic execution of the analyzed text was found in the core engine.

adcheckup-content-compliance-free