The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the terminal and process tools to interact with the grok CLI and system utilities.
Evidence: Examples include managing background processes, utilizing tmux for interactive sessions, and using git worktrees for isolation.
Evidence: Commands such as stat, ps, and lsof are used for monitoring session progress and detecting stalled runs, which is a legitimate use case for process management.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the delegation of tasks to an external AI agent (Grok), which creates an indirect prompt injection surface where the external agent's output could influence the primary agent's context.
Ingestion points: The skill accepts user-supplied task briefs that are passed as arguments to grok -p in SKILL.md.
Boundary markers: The documentation explicitly recommends using --permission-mode dontAsk and --deny rules for review-only runs to limit the capabilities of the external agent.
Capability inventory: The skill utilizes terminal and process tools which allow for file system writes, command execution, and network access (if not denied via grok flags).
Sanitization: The instructions strongly emphasize the need for independent verification of diffs, running tests, and performing git diff --check before committing any changes produced by the external agent.

grok-build-cli