The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions by interpolating untrusted user input (such as product pages, descriptions, or goals) directly into a structured prompt template. This creates a surface for indirect prompt injection where malicious instructions embedded in the user-provided content could potentially influence the agent's behavior.
Ingestion points: User input is collected via placeholders in SKILL.md: 【粘贴产品页面、网页、方案、流程说明、数据看板或其他成品】 and 【填写你希望从中学会什么】.
Boundary markers: The skill does not use explicit delimiters or instructions to the agent to ignore embedded commands within the user-provided content.
Capability inventory: Based on the provided files, the skill does not possess dangerous capabilities such as file system writes, network operations, or subprocess execution. Its primary function is text generation.
Sanitization: There is no evidence of sanitization, escaping, or validation performed on the user input before it is processed by the agent.

reverse-teardown