The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the creation of commands that execute shell operations. It provides examples using git, npm, and npx, and documents the platform's dynamic context injection feature using the backtick syntax. While intended for local development workflows, this grants the agent the ability to execute code on the host system.
[PROMPT_INJECTION]: The "Deep Research" and "PR Summary" skill patterns documented in examples.md ingest untrusted external data (via WebFetch) or local repository content (via git diff). These patterns represent a surface for indirect prompt injection (Category 8), as the templates do not implement boundary markers or instructions to isolate the data from the agent's control logic, potentially allowing external content to influence agent actions.
[EXTERNAL_DOWNLOADS]: The "Deep Research" template utilizes tools like WebSearch and WebFetch to retrieve content from the internet. While these are standard agent capabilities, they involve ingesting data from arbitrary remote sources which are then processed by the LLM.

create-skill