The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the grok CLI with the --always-approve flag, which allows the agent to edit files and execute shell commands without user confirmation during each turn. It also utilizes shell commands like mkdir, git diff, and git clean for environment management.- [EXTERNAL_DOWNLOADS]: The skill provides instructions to run grok update, which downloads and installs updates from xAI's infrastructure, introducing a third-party software dependency.- [REMOTE_CODE_EXECUTION]: By sending task specifications to xAI's Grok model and allowing it to return and execute code locally via the CLI, the skill establishes a remote-controlled execution path.- [DATA_EXFILTRATION]: Local code context and task specifications are sent to xAI's servers. While the skill advises against sending secrets, the operation inherently involves transmitting project data externally.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection, where malicious instructions in a processed repository could influence the Grok agent's output.
Ingestion points: Project files within the repository specified in the cwd flag.
Boundary markers: Use of a structured Markdown task specification template to isolate instructions.
Capability inventory: File writing and shell execution capabilities granted to the grok CLI.
Sanitization: A mandatory review gate where the orchestrator agent must inspect diffs and run acceptance tests before committing changes.

grok-build