The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from URLs and local files. Ingestion points occur during the source gathering step (匶材) as defined in SKILL.md and 教合案.md. Boundary markers are minimal, primarily relying on Markdown headers in the structured notes template. The skill has capabilities for file writing and web fetching but lacks explicit technical sanitization of the ingested data beyond the agent's natural language processing, making it susceptible to indirect prompt injection.
[DYNAMIC_EXECUTION]: The skill generates HTML files containing JavaScript for quizzes and interactive demonstrations. These files are executed in the user's browser, representing a script generation vector where the generated code could be influenced by malicious instructions or content within the source materials.
[DATA_EXPOSURE]: The skill instructs the agent to search for study materials in sensitive local project files such as MEMORY.md, CLAUDE.md, and AGENTS.md if no external URL is provided. This could lead to the unintended exposure of project configurations, agent history, or other sensitive metadata into the agent's context and subsequent study notes.
[COMMAND_EXECUTION]: The skill utilizes shell commands like mkdir -p and requires absolute file paths to manage local study directories and generate study artifacts.
[INDIRECT_PROMPT_INJECTION]: A potential cross-site scripting (XSS) vulnerability exists in the templates/核馆数换样.html file which uses innerHTML to render content from studied articles. If the source material contains malicious HTML tags or scripts that are not properly escaped by the agent during file generation, they could execute in the user's browser environment.
[EXTERNAL_DOWNLOADS]: The skill uses web fetch tools (like agent-reach) to download content from arbitrary external URLs provided by the user to serve as study material.

article-study