The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to run various shell commands to perform code counting and to install dependencies if they are missing. This includes tool-specific commands like cloc and git, as well as system-level package manager commands.
[PRIVILEGE_ESCALATION]: In the 'Bootstrap When Missing' section, the instructions explicitly guide the agent to use sudo with apt, yum, and dnf to install software, which involves acquiring root-level permissions on the host system.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the cloc utility from several external sources, such as NPM, Homebrew, Chocolatey, and Docker Hub (aldanial/cloc). It also suggests downloading the latest upstream release directly if necessary.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes files within a repository, creating a vulnerability surface where malicious content in those files could attempt to influence the agent's behavior.
[INDIRECT_PROMPT_INJECTION] (Ingestion points): The skill uses cloc to read and analyze the content of files across the target repository.
[INDIRECT_PROMPT_INJECTION] (Boundary markers): There are no instructions to use delimiters or ignore potential commands embedded in the source code being analyzed.
[INDIRECT_PROMPT_INJECTION] (Capability inventory): The skill can execute shell commands, install system packages, and is documented as having the ability to edit files and run build or validation scripts.
[INDIRECT_PROMPT_INJECTION] (Sanitization): The skill lacks explicit sanitization or filtering of the file content before it is processed.

cloc