The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions require the agent to use the grok CLI with the --always-approve flag for headless task delegation. This flag enables the autonomous execution of both filesystem edits and arbitrary shell commands by the external agent. This creates a significant capability gap where the delegated AI can perform unauthorized system actions before the user or the orchestrating agent reviews the resulting changes.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it instructs the delegated agent (Grok) to read and analyze content from the user's repository, which may contain adversarial instructions embedded in comments or documentation.
Ingestion points: Local repository files read by the grok CLI during the execution loop defined in SKILL.md.
Boundary markers: Present. The skill utilizes a specific Markdown task specification template to scope Grok's context and constraints, and requires user approval of the task spec before dispatch.
Capability inventory: Full filesystem modification rights and shell command execution on the host machine via the grok CLI as documented in references/cli.md.
Sanitization: Absent. There is no automated sanitization or filtering of the repository content before it is processed by the external model.

grok-build