The Agent Skills Directory

[COMMAND_EXECUTION]: The AGENTS.md file directs the agent to execute the compose-agentsmd command to ensure the local project rules remain synchronized with the latest standards.
[REMOTE_CODE_EXECUTION]: Instructions in the README.md and AGENTS.md recommend running the compose-agentsmd tool via npx or installing it globally, which results in the download and execution of packages from the public npm registry.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch its governing rules from an external GitHub repository hosted by the author at github.com/metyatech/agent-rules.
[PROMPT_INJECTION]: The skill uses a dynamic rule-loading mechanism where core instructions are retrieved from an external repository. This creates a surface for indirect prompt injection if the source repository or the update process were compromised. The mechanism includes a human-in-the-loop safeguard requiring approval before applying updates.

cli-design