The Agent Skills Directory

[SAFE]: The skill serves as a set of instructional guidelines for the agent to follow when creating plans. It does not contain any malicious patterns or functional code.
[FRONTMATTER]: The allowed-tools field restricts the agent to Read, Glob, and Grep. This is a read-only set of tools that prevents the skill from performing any destructive actions on the filesystem or making network requests.
[COMMAND_EXECUTION]: While the skill mentions commands like npx create-next-app, npm run dev, and curl, these are clearly presented as illustrative examples for the agent to include in its planning documentation, not as commands to be executed automatically by the skill itself.
[INDIRECT_PROMPT_INJECTION]: The skill defines how to process user tasks into plans. While it lacks explicit boundary markers for user input, its limited toolset (read-only) prevents any significant security impact from potentially malicious task names.

plan-writing