The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill implements a 'fetch and execute' pattern by downloading a SKILL.md file from a user-provided GitHub repository and instructing the agent to 'follow the instructions' within it. This enables the execution of unvalidated logic inside the agent's current session.
[COMMAND_EXECUTION]: The skill executes shell commands using the GitHub CLI (gh api) and base64 to pull content from the internet into the agent's context.
[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by treating remote data as authoritative instructions.
Ingestion points: Data is ingested from external SKILL.md files hosted on GitHub via the gh api command.
Boundary markers: None. The agent is explicitly told to treat the fetched content as a valid skill and obey its instructions.
Capability inventory: The agent can run shell commands and interact with APIs, which loaded instructions could exploit.
Sanitization: No validation or filtering is performed on the content retrieved from GitHub before processing.

try