Grok BotWorkPlanningAI score · 75/100AI-assisted edit · no obvious prompt-injection riskX

Grok Bot Human-in-the-Loop Approval Workflow

Dispatch coding tasks to agents via 1acp, with machine checks and human approval at key nodes to prevent vague-requirement waste

ExperiencedStepsWorkflow
Who
Tech leads or senior developers who already have AI agent infrastructure and need to control automated coding quality
You get
Understand and reference a human-in-the-loop approval architecture: Grok Bot → 1acp → coding agent → machine checks → human approval → release / reject
You need
Grok Bot

Constraints

Requires existing Grok Bot access and a Mac environment capable of running 1acp (the author's self-built MCP+CLI layer). Not a generic SaaS; cannot simply sign up and use.

What they did

ScottZeng93 explored automated coding from 2024–2026. On Programmer's Day 2025 he proposed SPEC programming; in H1 2026 he shipped the open-source product 1agents, aiming for 'one sentence in, fully automatic development out.' On June 25, 2026, the repo logged 44 PRs (43 merged same day), but later review revealed much output was half-finished or unusable—including the 'Agents roundtable brainstorm' feature that went from MVP to rewrite to patches to deprecation in five days. Root cause diagnosed as vague requirements letting agents fill gaps themselves, making late-stage acceptance and modification impossible.

How it went

The author evolved the 1agents experience into the 1acp architecture: Grok Bot dispatches → coding agent executes → machine checks at key nodes → human approval workbench. As of 2026-10-26, the ledger shows 82 jobs (45 completed, 26 failed, 7 canceled) from one week of actual use, not a benchmark. One README report went from acceptance to human approval in about 2 minutes. (author_report)

Worth knowing

1acp currently runs on one Mac for one owner; it is not a generic workflow engine and does not coordinate across users. 'One-sentence coding' remains viable for small projects (scripts, pages, one-off tools) but risks spinning out of control for large projects with fuzzy requirements. An unsolved problem the author raises: human reviewers may grow lazy when agents produce long, confident reports with green checks, turning approval into a rubber stamp.

Steps

  1. Grok Bot creates a task workflow

    Describe requirements in Grok Bot; the Bot generates a task workflow. The workflow dispatches tasks to an appropriate coding agent (Codex, Claude, or Grok Build, depending on the job) through 1acp, an MCP and CLI layer built on the Agent Client Protocol.

  2. Set acceptance standards at key nodes

    Set acceptance standards at key nodes in the workflow, not just at final output. Results that fail checks are automatically sent back to the agent for rework.

  3. Machine checks run automatically

    After the agent produces a result, machine checks run first. For example, one README report was accepted at 23:08:57 and passed machine checks about 36 seconds later.

  4. Human approval workbench decision

    Results that pass machine checks enter the approval workbench. The operator has two options: - Approve: releases the result; approval is bound to that version and its SHA-256 hash. If content changes, the approval no longer counts. - Request changes: sends it back to the author's agent session, which produces a new version that goes through checks again. Agents cannot approve on the operator's behalf.

Prompt you can copy

Reference prompt · Editorial suggestion。Confirm the destination, input variables and output format before use.

This block is the prompt. Ads are never placed here.

No prompt needed — just follow the steps.

Source

Why my Grok Bot waits for approval：人在环内审批工作流 · ScottZeng93 · X · 2026-10-26T11:20:43.000Z

Comments & feedback

Tried this method? Share results, problems or improvements. Comments appear after moderation. Please avoid sensitive information.

Loading…

    Related