The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to interact with the local grok CLI tool and git. While these commands are central to the skill's functionality, they involve executing logic based on interpolated prompts.
Evidence: Section 1 and 3 provide shell command templates for grok --version, grok --help, and grok -p using heredocs and subshells.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates a subagent workflow where the parent agent processes and reviews code changes generated by an external tool (Grok CLI), which could theoretically contain malicious instructions or insecure code patterns.
Ingestion points: Output, design decisions, and file edits produced by the grok CLI command (SKILL.md, Section 4).
Boundary markers: The skill uses explicit completion criteria, heredoc delimiters for prompts, and clear behavioral constraints (e.g., "Do not commit, tag, push").
Capability inventory: The parent agent is instructed to use shell execution for tool invocation and git commands for repository inspection.
Sanitization: The workflow mandates a central review phase (SKILL.md, Section 4) where the user/parent agent must inspect git diff, verify public interfaces, and run independent tests to validate the subagent's work.

grok-build-subagent