The Agent Skills Directory

[COMMAND_EXECUTION]: The apm.yml manifest supports a scripts field that enables the execution of arbitrary shell commands. These can be triggered manually via apm run, providing a direct path for command execution.
[REMOTE_CODE_EXECUTION]: The tool includes a postinstall hook in the scripts section that automatically executes specified shell commands upon successful installation of a skill. This allows for immediate execution of potentially malicious code when installing skills from unverified repositories.
[EXTERNAL_DOWNLOADS]: The skill explains how to download and install packages from remote platforms like GitHub and GitLab. The support for unpinned version references (e.g., using branch names like #main) presents a supply chain risk where the remote content can be modified after an initial audit.
[REMOTE_CODE_EXECUTION]: The manifest allows for the configuration of MCP (Model Context Protocol) servers via remote URLs and environment variables, which could lead to interaction with malicious servers capable of executing unauthorized logic.

apm-usage