The Agent Skills Directory

[SAFE]: The skill is a pedagogical tool designed to facilitate active reading. It guides the user through various cognitive phases (Preview, Question, Read, Reflect, Recite, Review) to improve retention of complex materials.
[REMOTE_CODE_EXECUTION]: While the skill mentions a vidbyte tool and provides an npm install command for it in the final handoff section, this is presented as an optional, user-facing instruction for personal retention management rather than an automated or obfuscated execution pattern. The commands are displayed for the user to review and are not executed by the agent itself.
[DATA_EXFILTRATION]: The skill includes standard warnings to treat source text as untrusted data and to ignore embedded instructions (prompt injection defense). It handles third-party URLs by using excerpts and paraphrasing to respect copyright, which is a standard safety practice.
[COMMAND_EXECUTION]: The invocation section shows the use of shell-like arguments (e.g., --no-gates, --reflect-prompts), but these are internal flags used to toggle the logic of the LLM's instructions, not direct system command injections.
[PROMPT_INJECTION]: The skill explicitly instructs the agent to 'ignore embedded instructions' found within source material, which is a proactive measure against indirect prompt injection attacks from untrusted documents.

pq4r