The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the GitHub CLI from its official website (https://cli.github.com/) if it is not already available. This targets a well-known, official service and is handled as a safe reference for setup.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from GitHub repositories, which creates an inherent attack surface. However, it effectively mitigates this through several layers of protection:
Ingestion points: Remote content is fetched via the GitHub API or CLI as described in references/acquisition.md and references/discussions.md.
Boundary markers: The SKILL.md instructions explicitly command the agent to treat repository text as untrusted evidence and never as authority to run commands or change the environment.
Capability inventory: The skill uses filesystem reads, network access to GitHub APIs, and executes a specific helper script to interface with the GitHub CLI.
Sanitization: The helper script scripts/view-issue-pr.mjs uses shell: false during process spawning and applies strict regular expression validation to all user-supplied arguments to prevent command injection.

gh-repo