Add a judge subagent and two review passes
A solo developer used Claude Code to build an app with fresh sessions, a read-only judge, and two review layers for tighter quality control.
- What they did
- They split the work into short child sessions for each issue so the main thread stayed focused. For decisions that were still open, they sent the question to a separate judge subagent that could only inspect the work and had to mark each claim as confirmed, guessed, or unknown instead of filling gaps from guesswork. After implementation, they ran a general code review and then a second review based on coding-guideline skills, leaving any unresolved items for a person to handle in pull requests.
- How it went
- That workflow was used on a Windows app and reportedly helped land a Microsoft Store release. It also produced 142 merged PRs in 17 days. The tradeoff was extra overhead from the fresh sessions, the judge, and the two review stages, so it is not a light setup.
- Tips
- Use this only if you need strong checks and can afford the coordination cost. It is better suited to an advanced workflow than to a beginner trying to move quickly.
Try it yourself with Claude Code
Work in [repo path] on [app or feature]. For each task, implement the change in a fresh child session, then have a separate read-only judge subagent inspect the result against the requirements and tests. The judge should label every uncertain statement as confirmed, guessed, or unknown, and must not turn guesses into decisions. After that, run two review passes: first a general code review, then a second review against [coding guidelines or standards]. If anything is still unclear or would require sending, deleting, submitting, or making another irreversible change, stop and ask me for approval before proceeding. Keep a brief decision log so later sessions can pick up the context, and finish by summarizing the final state and any deferred items.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Build a project workflow after a guided interview
They used Claude Code to turn a repo interview into a custom workflow system, while the agent wrote, tested, reviewed, and shipped the code.
Move WordPress posts into MDX files
Pauline used Claude Code to pull a WordPress archive into MDX and sort it into year folders.
Run Microsoft Office on Linux with Wine
Use Wine and compatibility shims to get Microsoft 365 installing and opening on Linux without a virtual machine.
Keep many coding agents moving with local context cache
A developer ran several Claude Code sessions at once on local RTX 3090 hardware and avoided reloading long contexts after idle periods.
Hand web browsing tasks from Claude Code to Comet
A developer wired Claude Code to Perplexity Comet so it could browse, log in, and research while coding continued separately.
Use Claude Code from a phone with a remote VM
They accessed Claude Code on an iPhone by pairing a Tailscale-only Vultr VM with Termius, mosh, and tmux.