Interactive coding agent
Read, edit, run commands, search, and review diffs in a full-screen terminal interface.
The executable is `grok`; Grok Build is the product name, not a shell command.
Open official xAI documentation ↗TERMINAL AGENT / SYSTEM MAP
Grok Build system map
Understand the coding-agent surfaces you operate from the terminal, from a reviewed plan to parallel work and machine-readable automation.
CHANNEL08 GROK BUILD SURFACES
SOURCES CHECKED2026-07-26
ORIGINOFFICIAL DOCS / PRIMARY
Plan, review, approve
Explore first, comment on the proposed plan, then explicitly approve implementation.
Plan mode gates edit tools, not every possible shell write.
Open official xAI documentation ↗Permissions and sandbox
Combine approval rules with an OS-level filesystem and network boundary.
Always-approve is not a substitute for deny rules or sandbox isolation.
Open official xAI documentation ↗Repository instructions
Load AGENTS.md and scoped rule files so conventions travel with the codebase.
Use `grok inspect` to verify exactly which instructions were discovered.
Open official xAI documentation ↗Sessions and worktrees
Resume context, rewind turns, export transcripts, and isolate concurrent changes.
Worktrees persist until you remove them; inspect before garbage collection.
Open official xAI documentation ↗Subagents and workflows
Delegate focused work to child sessions or orchestrate verified multi-phase runs.
Workflow details are new and volatile; do not infer an undocumented shell subcommand.
Open official xAI documentation ↗Skills, plugins, hooks, and MCP
Package reusable guidance and connect tools, lifecycle scripts, and external systems.
Review third-party code and trust boundaries before enabling an extension.
Open official xAI documentation ↗Headless and ACP
Run one-shot prompts with structured output or embed Grok through ACP over stdio.
Automation needs explicit permissions, stable output parsing, and bounded turns.
Open official xAI documentation ↗