ClaudeSoftwareShipping codeAI score · 72/100AI-assisted edit · no obvious prompt-injection riskX

Orchestrate 40+ AI Agents with Claude Code Shortcuts for Code Changes

A non-coder routes risky code changes through /qnew → /qplan → /qcode → /qgit shortcuts, letting specialized agents collaborate in stages

Some experienceStepsWorkflowTip
Who
Technical leads or product managers who don't write code but need to manage complex code changes
You get
Encapsulate the full development workflow as Claude Code shortcuts, composed by risk level, enabling 40+ agents to split tasks and collaborate
You need
Claude Code

Constraints

Requires Claude Code with the Superpowers open-source plugin loaded for 14 helper skills. Each shortcut maps to a written playbook triggered by human keystrokes, not autonomously. Normal and high-risk changes use different command combos.

What they did

The author saved their entire development workflow as Claude Code shortcuts, each backed by a written playbook that tells the AI exactly what to do step by step. The AI cannot fire them on its own—the author presses every key. Which keys a change needs is decided together with a separate strategist Claude chat.

How it went

The author reports that a risky change with 10 small tasks can mobilize 30 build agents, 6 deep-review agents, 4 automatic checks (Claude, Greptile, security scan, /mika-lint), plus the lead agent, /qsec, ChatGPT (Codex), and the strategist—totaling 44 agents. Normal changes need far fewer; deep reviews run only when risk calls for it. (Source: author report)

Worth knowing

The author's four lessons: write the playbook once and reuse forever; pick names you can type half asleep; many small, fresh agents beat one big one; keep the trigger human. The /qstrategy experiment is already built but underperformed (the agent tended to slope toward approval rather than real assessment); they plan to retry with a better model in a couple of months. Playbook text, Superpowers plugin helper-skill details, and /mika-lint rules were not provided.

Steps

  1. Start session

    Type /qnew in Claude Code to load personal rules and 14 helper skills from the Superpowers plugin. One lead agent runs the session.

  2. Plan the change

    Type /qplan to have agents brainstorm 2–3 approaches with trade-offs; the human picks one. For hard features, run 3 planning agents in parallel and take the best plan. Then type /qrt to slice the plan into small tasks of 2–5 minutes each.

  3. Build task by task

    Type /qcode to execute tasks one by one. Each task spins up 3 fresh agents: one builds, one checks plan adherence, one checks code quality. Each agent sees only its own task so nothing carries over. If a real problem is found, a fourth agent fixes it. Type /qgoal for autopilot: it runs /qcode until all tests pass with a hard cap, then a judge agent verifies whether the issue is truly solved or just green on paper.

  4. Ship and review

    Type /qgit to prepare the change and pause for human approval. Type /mika-lint to check house rules. Then automatic checks run: Claude, Greptile, and a security scan review independently; ChatGPT (Codex) gives a second opinion. For risky changes, also type /qlayer3 to launch 6 specialist agents in parallel on the top model (hunt silent failures, check tests, data types, comments, code itself, simplification opportunities), and type /qsec for a security deep-dive in a fresh session.

  5. Clean up

    After merge, type /qcleanup to tidy up automatically. No agent needed—just a script.

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

不写代码的人把整套开发流程存成 Claude Code 快捷命令：/qnew、/qplan、/qcode（每任务 3 个新 Agent）到 /qgit · alexander_jais · X · 2026-10-25T15:35:08.000Z

Comments & feedback

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

Loading…

    Related