The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is designed to download and execute arbitrary scripts from user-provided remote repositories. The agent is explicitly instructed to run these scripts after downloading them, which is a high-risk pattern for executing unverified code from the internet.
[PROMPT_INJECTION]: The skill fetches remote Markdown files and instructs the agent to follow their instructions as if they were locally installed. This creates a surface for indirect prompt injection where a remote source could provide malicious instructions to override safety guidelines or exfiltrate context.
Ingestion points: Remote SKILL.md and referenced files are downloaded to a temporary directory for processing.
Boundary markers: Absent; the skill does not use strict delimiters for the fetched content, although it includes a 'Safety' section informing the agent that the content is untrusted.
Capability inventory: The agent is instructed to execute scripts and follow the instructions contained in the downloaded files.
Sanitization: No validation or sanitization of the remote content is performed before the agent reads and executes it.
[EXTERNAL_DOWNLOADS]: The skill fetches data and files from GitHub's API and raw content domains (api.github.com and raw.githubusercontent.com) to resolve skill references and download assets.
[COMMAND_EXECUTION]: The skill executes a local Node.js script (scripts/fetch-skill.mjs) which performs network operations and file system writes based on external input from the user.

run-skill