Grok Build
Delegates coding tasks to xAI's grok CLI headlessly, reviews every diff before commit
Test report
- Verdict
- Works with setup
- Score
- Tested
- Jul 21, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · de64b8a
Cloned the repo and installed skills/grok-build into a throwaway HOME=$(mktemp -d); frontmatter parsed to name+description (489 chars), the one file the body references (references/cli.md) returned HTTP 200 raw, and a grep for curl|sh, base64, secret exfiltration, injection text and hardcoded /Users or /home paths found nothing (only match was the word "token" in prose about token cost). `which grok` returned "grok not found" and the CLI needs a paid grok.com OAuth login, so I could not execute the preflight, the dispatch, or the resume fix-up — the skill's entire execution path. I did measure the part I could: I built a real git repo (wordcount.py + a passing pytest), then wrote two dispatch artifacts (A-baseline-dispatch.md = the one-liner `grok -p "Add a --json flag..."` I'd send by default; B-skill-dispatch.md = the skill's full spec template with Context/Files/Task/Constraints/Acceptance criteria) and implemented each from its prompt text alone on separate branches. Concrete difference: the baseline branch renamed the dict keys to line_count/word_count/char_count and `python3 -m pytest -q` reported "1 failed ... AssertionError"; the skill branch kept lines/words/chars and pytest reported "1 passed", and `--json` printed exactly `{"lines": 2, "words": 3, "chars": 6}` matching the spec's acceptance line — the delta traces to two things only the template made room for (the "the test imports it directly" constraint and the exact expected-output strings). Caveat that keeps outputMeasured false: I substituted myself for Grok, so this measures spec quality, not Grok's real behavior, and none of cli.md's "verified against grok 0.2.93" claims (acceptEdits cancelling headlessly, sessionId JSON shape, --check ~48s) could be checked.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 5/5
- Output vs. baseline 4/10
- Docs & honesty 4/5
What Grok Build does
Turns the assistant into an orchestrator that writes self-contained task specs, dispatches them to xAI's Grok Build CLI in headless mode, then reads the resulting diff and runs acceptance commands before committing. Includes a spec template, the headless flag set (--prompt-file, --output-format json, --always-approve, --max-turns), a resume-based fix-up loop capped at two rounds, and a failure table. Triggers when the user says "use grok", "delegate to grok", "have grok implement", or "execute this plan with grok".
How to install Grok Build
git clone --depth 1 https://github.com/sanjay3290/ai-skills.git /tmp/grok-build-src
mkdir -p ~/.claude/skills
cp -R /tmp/grok-build-src/skills/grok-build ~/.claude/skills/grok-build
# REQUIRED EXTERNAL DEP: xAI's Grok Build CLI (`grok`) is NOT installed by the above.
# Install it per xAI's docs for your OS, then verify: grok --version
# Auth needs a grok.com subscription OAuth login: grok login
# Check auth with: grok models (skill's preflight STOPs if this errors)
# Works on macOS, Linux, Windows PowerShell. Repo also ships 23 other skills under skills/.
# Skill body is 135 lines + references/cli.md loaded on demand.
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Grok Build
-
/grok-buildDelegates coding tasks to xAI's grok CLI headlessly, reviews every diff before commit
It also activates on plain-language prompts like these:
-
Have grok implement this plan step by step -
Delegate this task spec to grok build -
Execute this markdown plan with grok
Frequently asked questions
- Is the Grok Build skill free?
- Yes. The skill itself is free from sanjay3290/ai-skills. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Grok Build work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 21, 2026. Verdict: Works with setup. Cloned the repo and installed skills/grok-build into a throwaway HOME=$(mktemp -d); frontmatter parsed to name+description (489 chars), the one file the body references (references/cli.md) returned HTTP 200 raw, and a grep for curl|sh, base64, secret exfiltration, injection text and hardcoded /Users or /home paths found nothing (only match was the word "token" in prose about token cost). `which grok` returned "grok not found" and the CLI needs a paid grok.com OAuth login, so I could not execute the preflight, the dispatch, or the resume fix-up — the skill's entire execution path. I did measure the part I could: I built a real git repo (wordcount.py + a passing pytest), then wrote two dispatch artifacts (A-baseline-dispatch.md = the one-liner `grok -p "Add a --json flag..."` I'd send by default; B-skill-dispatch.md = the skill's full spec template with Context/Files/Task/Constraints/Acceptance criteria) and implemented each from its prompt text alone on separate branches. Concrete difference: the baseline branch renamed the dict keys to line_count/word_count/char_count and `python3 -m pytest -q` reported "1 failed ... AssertionError"; the skill branch kept lines/words/chars and pytest reported "1 passed", and `--json` printed exactly `{"lines": 2, "words": 3, "chars": 6}` matching the spec's acceptance line — the delta traces to two things only the template made room for (the "the test imports it directly" constraint and the exact expected-output strings). Caveat that keeps outputMeasured false: I substituted myself for Grok, so this measures spec quality, not Grok's real behavior, and none of cli.md's "verified against grok 0.2.93" claims (acceptEdits cancelling headlessly, sessionId JSON shape, --check ~48s) could be checked.
- What is the Grok Build SkillProof Score?
- 7.2/10 — installs cleanly 5/5, triggers reliably 5/5, output vs. baseline 4/10, docs & honesty 4/5.
- How do I install Grok Build?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Grok Build with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.