Sniffer
Investigation and requirements analysis. Finds facts, traces issues, gathers evidence.
Open source CLI orchestrator
Turn Grok CLI into an autonomous engineering crew
Multi-agent orchestration for the xAI Grok CLI — real subagents, durable workflows, lifecycle hooks, memory, and worktrees that keep messy work isolated until it is verified.
$ npx grokgoblin@latest setup
$ goblin cruise "add a /health endpoint with tests"
[RALPH] iteration 07/50
> reflect → tests failing; endpoint not wired
> act → add route, handler, and test
> verify → npm test
✓ 24 passed, 0 failed
[COMPLETE] evidence captured → PR ready
512K
context window
10
specialist agents
5
workflow stages
∞
autonomous loops
The squad
Meet the crew
Specialist goblins with defined roles, tool boundaries, and routing discipline. Click any card to launch their sandbox output.
Schemer
Architecture and planning. Evaluates tradeoffs and writes implementation strategy.
Basher
Precise implementation inside approved scope. No wandering, no accidental expansions.
Squasher
Root-cause debugging. Minimizes surface area and fixes the actual failure path.
Warden
Security review for auth flaws, secrets, injection risks, and data exposure.
Prover
Runs tests, validates behavior, and confirms evidence before work is called done.
…plus tinker, nitpick, forager & grunt — 10 specialists in all.
Pipeline
Workflow stages
Autonomous modes that drive work from ambiguity to verified completion.
1
dig
Clarify scope, success criteria, non-goals, and constraints.
2
goblinplan
Turn scope into architecture decisions and an execution plan.
3
quest
Execute larger work as checkpointed goals with a ledger.
4
tdd
Move Red → Green → Refactor before polishing behavior.
5
code-review
Self-review for correctness, security, edge cases, and regressions.
$
goblin cruise "add a /health endpoint with tests"
Reference
Command map
Everything important at your fingertips, from quick questions to isolated worktrees.
Execution
goblin ask <question>
goblin explore <topic>
goblin forage <topic>
goblin exec <prompt>
goblin exec --check
Workflows
goblin cruise <goal>
goblin quest <goal>
goblin ralph <task>
goblin hunt <objective>
goblin swarm [N] <task>
Review & ship
goblin review
goblin review <PR#>
goblin ship <goal>
goblin worktree
goblin -w <name>
Memory
goblin memory status
goblin memory search "..."
goblin memory on|off
goblin memory edit
goblin memory clear
Worktrees
goblin worktree
goblin worktree new [name]
goblin worktree rm <name>
goblin worktree clean
goblin -w feature-x
Config & tools
goblin config get <key>
goblin config set <key> <val>
goblin skills list
goblin hooks list
goblin doctor --verbose
Comparison
Grok CLI vs. GrokGoblin
What makes GrokGoblin a complete autonomous workflow layer compared to standard CLI wrappers.
| Capabilities | Standard Grok CLI | GrokGoblin Layer |
|---|---|---|
| Agent Collaboration | ❌ Single linear chat agent | ✓ 10 collaborating specialists |
| Workspace Safety | ❌ Modifies files directly on main | ✓ Sandbox worktree isolation |
| Durable Pipelines | ❌ Forgetful chat states | ✓ Multi-stage checkpoint ledger |
| Behavior Loops | ❌ Manual compile/test run loops | ✓ Self-healing test execution |
| Persistent Memory | ❌ Session-only token caching | ✓ Cross-session project memory |
| Custom Automation | ❌ Default prompts only | ✓ Extensible skills & lifecycle hooks |
Built-in
Memory, hooks, worktrees
Built-in primitives for durable, reliable autonomous work.
Persistent memory
Grok's native cross-session memory, enabled at setup, plus a local project file injected through the launch overlay.
.grokgoblin/memory/project.md
Lifecycle hooks
Tool and session events log state transitions, notifications, and runtime evidence.
~/.grok/hooks/hooks.json
Isolated worktrees
Risky work lands on gg/ branches in sibling worktrees before integration.
../repo.gg-worktrees/name
FAQ
Frequently Asked Questions
Get answers to common structural, keying, and execution questions.
A leader goblin (the orchestrator) decomposes your goal into tasks and delegates each to the right specialist subagent — Sniffer for investigation, Schemer for planning, Warden for security review — running independent ones in parallel. Each role has strict tool boundaries and a defined input/output, which reduces context drift and prevents conflicting file writes. Read-only roles are capability-locked so they physically cannot edit files.
No. By default, any execution task initialized via goblin worktree or goblin cruise clones your current active state into a sibling git worktree directory. Coding changes and testing compile loops are run within this isolated space, checking changes in only after test coverage validations succeed.
GrokGoblin routes all actions directly through your authenticated local grok CLI shell. If your terminal is logged in (using grok login), GrokGoblin wraps its commands automatically, utilizing your active account context without requiring additional monthly token fees.
Skills are markdown instruction folders that become / slash commands inside a Grok session — GrokGoblin installs its own (/dig, /goblinplan, /swarm, /hunt, /review, /ship, /forage, and more) to drive each workflow. Hooks are scripts grok runs before/after tool and session events — GrokGoblin uses them to track state transitions and keep the orchestration overlay loaded.
Get started
Install GrokGoblin
One command, then launch the enhanced Grok CLI.
$ npx grokgoblin@latest setup
$ grok login
$ goblin
✓ GrokGoblin active — 10 agents ready
Setup installs
- GrokGoblin skills —
/dig,/goblinplan,/quest,/ralph,/cruise,/tdd,/code-review,/swarm,/hunt,/review,/ship,/forage - Native Grok subagents for the specialist crew
- Hooks and runtime overlay generation
- Config, memory, and orchestration state
Supports Node 20+, macOS, Linux, and standard Grok CLI sessions.