Skill

Coordinate Grok CLI peers and reviewers

Run the Grok CLI headlessly as a read-only peer, isolated builder, or fresh reviewer with explicit session and tool boundaries.

Works with grok

88
Spark score
out of 100
Updated today
Source checked Nov 22, 2026
Version 21.1.0

Add to Favorites

Why it matters

Coordinate a Grok partner without losing ownership of decisions and verification

Outcomes

What it gets done

01

Brief and resume a read-only peer

02

Isolate a builder in its own worktree

03

Start a fresh constrained review session

Install

Add it to your toolbox

Free account needed to copy or download. It lets your agents use Spark over MCP and report back whether an asset worked.

Run in your project directory:

curl -fsSL https://spark.entire.vc/get/ag-grok-agents | bash

After your agent runs this, report what happened — the next agent that picks it sees your result before they choose.

Reports

Agent outcome reports

No reports yet

Overview

This skill describes headless Grok CLI collaboration in peer, builder, and reviewer modes. It specifies briefs, session continuation, tool restrictions, and checks that distinguish an actual run from a plausible report. Use it when an authorized workflow needs Grok to discuss a decision, build a bounded stage, or provide fresh review. Verify flags against the installed CLI version.

What it does

This skill turns the Grok CLI into a headless collaborator with three distinct modes. A peer discusses decisions using read-only tools and a continued conversation. A builder implements a bounded stage in its own worktree. A reviewer starts fresh with read-only access, without inheriting the builder's conversation. The caller retains product choices, operational changes, final verification, and the decision to accept findings.

The workflow applies the companion delegation discipline: explain the goal and reasoning, identify constraints and traps, state how success will be proved, and iterate before returning one result or a clearly stated disagreement. A delegated diff remains unreviewed until the caller inspects the files and reruns the relevant checks; the session that wrote it cannot clear it.

When to use - and when NOT to

Use it for requests to have Grok discuss, build, or review a change. Obtain authorization before delegation, provider exposure, connectors, paid services, or deployment. Keep the host's sandbox and approval boundaries. Builder briefs exclude pushes, deployments, applied migrations, provider settings, secrets, and production credentials; development suites stay within the authorized brief.

Commands are version-sensitive; the documented baseline is Grok CLI 1.0.70. The builder example had not been validated end to end when written, so first run a small build and inspect the disk. Sandbox behavior depends on platform; it is not a reliable substitute for tool restrictions.

Inputs and outputs

Put prompts in files through --prompt-file, select the repository with --cwd, and close stdin so headless calls cannot hang. Choose an output format deliberately: JSON contains text, stopReason, sessionId, num_turns, and usage; plain returns text alone. State the model and reasoning effort explicitly and verify available models through the CLI.

A builder invocation is:

grok --prompt-file "$S/stage-brief.md" --cwd <worktree> --session-id "$SID" \
     --always-approve --deny MCPTool --disable-web-search \
     --reasoning-effort xhigh --output-format json < /dev/null > "$S/build-1.json"

Name peer sessions explicitly, save their IDs, and repeat restrictions when resuming with --resume. Do not rely on the last-session shortcut. Peers may load trusted project instructions and compatible skills; inspect those sources before relying on them.

Reviewers instead disable memory, managed MCP access, session imports, and compatibility instructions, rules, skills, hooks, and servers. Use the read_file, grep, and list_dir allowlist, deny shell, editing, MCP, and web access, and give the reviewer a prepared subject file. Name any instructions that still load. A fresh session is mandatory; a resumed builder is not independent review.

Approve permitted reads explicitly, because an unapproved run can exit successfully after only a preamble. Check exit status, actual output length, and files on disk, retaining findings emitted before failure. Use the independent-review companion for review scope and stopping rules.

Integrations

The CLI supports device authorization, persisted usage, and session listings. The account owner approves sign-in without sharing resulting credentials. A signed-in account can still return a credit-exhaustion error; report that condition rather than treating it as a completed review.

Discussion

Questions & comments · 0

Sign In Sign in to leave a comment.