The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads skill packages from GitHub. By default, it uses the official openai/skills repository, which is a trusted source. It allows user-specified repositories to support its core installation functionality.
[COMMAND_EXECUTION]: The helper scripts use the git command-line tool via subprocess.run to perform sparse checkouts of repositories. The implementation passes arguments as a list and disables shell execution, which prevents common command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the GitHub API and repository files.
Ingestion points: Fetches skill lists, metadata, and file contents from GitHub repositories during the listing and installation process.
Boundary markers: The skill instructions define specific output formats for listing skills, although they do not use strict data/instruction delimiters.
Capability inventory: The skill possesses capabilities for file system writes (installing skill files), network access (via GitHub API), and command execution (git CLI).
Sanitization: The skill includes validation logic to prevent directory traversal attacks (Zip Slip protection) and restricts network operations to the github.com domain.
[SAFE]: The skill follows security best practices for its intended use case, including the use of temporary directories for downloads, relative path validation, and verification of the SKILL.md file before proceeding with installation.

skill-installer