The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from external Git repositories based on a templateUrl provided by the user at runtime.
[COMMAND_EXECUTION]: The skill utilizes multiple shell commands to manage the project setup, including git init, git remote add, git fetch, git merge, and npm install.
[REMOTE_CODE_EXECUTION]: There is a potential path for remote code execution because the skill merges external code and then offers to run npm install. Although the skill explicitly requires user confirmation before running the installation, executing scripts from an untrusted template repository poses a risk.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the files it processes after a merge.
Ingestion points: The agent reads and performs search-and-replace operations on various files (e.g., package.json, angular.json, src/app/app.ts) sourced from the external template repository.
Boundary markers: No specific delimiters or instructions are provided to the agent to treat the content of the merged files as data rather than instructions.
Capability inventory: The skill possesses the ability to execute shell commands (git, npm) and perform broad filesystem modifications.
Sanitization: While the skill validates the projectId and projectName using regular expressions, it does not sanitize or verify the integrity of the content merged from the templateUrl before processing it.
[PERSISTENCE]: The instructions specifically target the modification of files located in .cursor/hooks/telemetry/scripts/. Because IDE hooks are designed to execute code automatically during development workflows, they can be utilized as a persistence mechanism if the template repository contains malicious script content.

project-create