/dev for Claude Code

Claude Code as a Tech Lead with parallel Worker Agents

52 followers

Claude Code as a Tech Lead with parallel Worker Agents

52 followers

Visit website

AI Coding Agents

•

AI Code Editors

Most AI coding tools give you autocomplete. /dev gives you an engineering process. A 6-phase SOP turning Claude Code into a Tech Lead with parallel Worker Agents in isolated worktrees: PRD alignment, architecture, coding, QA, code review, and iteration. Includes conflict detection, counterexample checks, mandatory security audits, and DB migration guards. Open source, MIT.

Free

Launch tags:Developer Tools•Artificial Intelligence•GitHub

Launch Team / Built With

getviktor.com — An AI coworker that actually does the work

An AI coworker that actually does the work

Promoted

/dev for Claude Code

Maker

📌

Hey Product Hunt! I built /dev because I was frustrated watching Claude Code write great code but with zero engineering discipline. No PRD, no architecture phase, no code review. /dev makes Claude act as a real Tech Lead: request classification, PRD alignment, architecture and task decomposition, parallel Workers in isolated worktrees, QA verification, and structured code review with veto conditions. Workers run 6-category counterexample checks. Bandit and npm audit run before every merge. Open source, MIT licensed. Feedback welcome!

Report

11d ago

Raycast

@yuanheng_ma Claude Code already suggests splitting and assigning tasks I give it to subagents, natively. How is this different?

Report

8d ago

/dev for Claude Code

Maker

@chrismessina Great question! You're right that Claude Code has native subagent support — /dev builds on top of that, not replaces it.

The key difference is engineering process discipline. Native subagents are like giving junior devs tasks with no guardrails. /dev adds what a real Tech Lead would enforce:

Before coding — PRD alignment + architecture review, so agents aren't building the wrong thing.

During coding — each Worker runs in an isolated git worktree, does a 6-category counterexample self-check (null, boundary, concurrency, malicious input…), and checks for file-overlap conflicts with other open Issues.

Before merging — mandatory security audits (bandit/npm audit), structured code review with veto conditions (e.g., any direct DDL = auto-reject), and post-merge PR coordination.

Think of it as: Claude Code gives you the agents, /dev gives you the SOP that keeps them from shipping broken code. It's the difference between "write code in parallel" and "run a disciplined engineering team."

Report

8d ago

Raycast

@yuanheng_ma cool, thanks — that's helpful!

What level of complexity is best suited for this? I feel like the tasks I give Claude Code aren't that sophisticated...

Report

8d ago

Raycast

@yuanheng_ma Heh, I asked Claude Code to assess whether this is a good fit for me:

I'd recommend against installing this skill. Here's why:
What it does: It's a /dev slash command that turns Claude into a "Tech Lead" orchestrating multiple subagent workers via git worktrees. It enforces a rigid 6-phase workflow:
requirements → architecture → parallel development → QA → code review → merge.
Why it's not a good fit for you:
Massive overhead for your project. Porthole is a SwiftUI app where you're iterating on camera views and UI polish. This skill is designed for large team-style workflows with GitHub Issues, PR-based code review, and multi-agent parallel development. It would add a lot of ceremony to tasks that are currently quick and direct.
Conflicts with your existing superpowers skills. You already have superpowers:brainstorming, superpowers:executing-plans, superpowers:dispatching-parallel-agents, superpowers:using-git-worktrees, and superpowers:requesting-code-review — which cover the same ground but in a modular, pick-what-you-need way. This skill would be a monolithic replacement that's less flexible.
Rigid "never write code in main conversation" rule. The skill's core constraint is that all code must go through worker agents and PRs. That's overkill when you want me to just tweak a SwiftUI view or fix a layout issue.
Low community signal. 10 stars, 0 forks, 7 commits, single contributor. The skill files themselves are well-structured and I didn't find anything malicious — the install script just copies markdown files — but it's very early-stage.
Bottom line: Your current skill setup already gives you subagent orchestration, worktree isolation, and code review when you want them, without forcing every change through a heavyweight process. This skill would slow you down more than it helps.

Report

8d ago

Reviews

Most Informative

@yuanheng_ma Heh, I asked Claude Code to assess whether this is a good fit for me:

I'd recommend against installing this skill. Here's why:
What it does: It's a /dev slash command that turns Claude into a "Tech Lead" orchestrating multiple subagent workers via git worktrees. It enforces a rigid 6-phase workflow:
requirements → architecture → parallel development → QA → code review → merge.
Why it's not a good fit for you:
Massive overhead for your project. Porthole is a SwiftUI app where you're iterating on camera views and UI polish. This skill is designed for large team-style workflows with GitHub Issues, PR-based code review, and multi-agent parallel development. It would add a lot of ceremony to tasks that are currently quick and direct.
Conflicts with your existing superpowers skills. You already have superpowers:brainstorming, superpowers:executing-plans, superpowers:dispatching-parallel-agents, superpowers:using-git-worktrees, and superpowers:requesting-code-review — which cover the same ground but in a modular, pick-what-you-need way. This skill would be a monolithic replacement that's less flexible.
Rigid "never write code in main conversation" rule. The skill's core constraint is that all code must go through worker agents and PRs. That's overkill when you want me to just tweak a SwiftUI view or fix a layout issue.
Low community signal. 10 stars, 0 forks, 7 commits, single contributor. The skill files themselves are well-structured and I didn't find anything malicious — the install script just copies markdown files — but it's very early-stage.
Bottom line: Your current skill setup already gives you subagent orchestration, worktree isolation, and code review when you want them, without forcing every change through a heavyweight process. This skill would slow you down more than it helps.