How to use Claude to run a reviewed coding pipeline
Codev combines git-tracked specifications, enforced phase transitions, isolated builder worktrees, and Claude, Gemini, and Codex reviews. Its author reported 20 bugs caught before shipping.
news.ycombinator.com
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Use [Codev project] to implement [feature]. Keep specifications and plans in git, read arch.md, and follow Spec → Plan → Implement → Review → PR. Use isolated builder worktrees, require passing tests, and obtain Claude, Gemini, and Codex reviews for each phase.
Reach people exploring AI agents
See placements →What happened
The author reported 20 bugs caught before shipping, including a severe security issue, while acknowledging longer execution and higher token use.
Reported by waleedk on Hacker NewsHow the agent did it
- Store specifications and plans in git alongside the source.
- Have agents read arch.md and their specific specifications.
- Launch builder agents in isolated git worktrees through an architect.
Where a person stepped in
The human directs the architect and writes specifications and review annotations; three models review each phase.
A person approved each step before the agent continued.
The author described a development process built around specifications, enforced phases, and multiple model reviewers. Specifications and plans live in git alongside source code. A new agent reads arch.md for the architectural picture and then reads its specific specification, keeping the rationale available beyond chat history.
A state machine enforces the sequence from specification to plan, implementation, review, and pull request. Tests must pass before work advances. An architect agent launches builder agents in isolated git worktrees, and the human directs the architect through specifications and review annotations. The agents communicate asynchronously. The author presents this as managing planning, review, integration, deployment scripts, and staging versus production alongside coding.
Each phase receives reviews from Claude, Gemini, and Codex. The author reported 20 bugs caught before shipping: Claude Code found five, while Gemini and Codex found another 15, including a severe security issue Claude missed. They also reported that no single model found more than 55 percent of issues. The stated tradeoffs were longer execution and greater token usage. The process was open-sourced as Codev.
Who it's for: Engineers who want an enforced development pipeline with explicit specifications and independent model reviews.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · waleedk
Frequently asked questions
What prevents phases from being skipped?
The author says a state machine enforces Spec → Plan → Implement → Review → PR, and tests must pass before advancing.
Why use three model reviewers?
The author reported different bug findings: Claude Code found five bugs, and Gemini plus Codex found another 15, including a severe security issue Claude missed.
More details
Remaining steps
- Enforce phase transitions with passing tests before advancement.
- Review each phase with Claude, Gemini, and Codex before a pull request.
About this account
An architect coordinates builder agents and lifecycle work within a state machine that enforces phases and test gates.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47208471. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: waleedk on Hacker News
Source published 1 Mar 2026
Source checked 14 Oct 2026
Tools and topics
Codev · Claude Code · Gemini · Codex · Git worktrees
Software development
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.