# Show HN: Devloop, a local code/review loop for Codex and Claude Code

> Source: <https://devloop.sh>
> Published: 2026-06-16 02:22:30+00:00



```
░█▀▄░█▀▀░█░█░█░░░█▀█░█▀█░█▀█
░█░█░█▀▀░▀▄▀░█░░░█░█░█░█░█▀▀
░▀▀░░▀▀▀░░▀░░▀▀▀░▀▀▀░▀▀▀░▀░░
```

Spec-driven code/review loop that runs until all acceptance criteria are met and bugs fixed by agents.

``` bash
$ curl -fsSL https://devloop.sh/install | bash
```

Requires Codex and Claude Code. Runs on macOS.

## ## WHY

Developers now spend most of their time in code
review, assisted by coding agents. But the same
model family can't both code and review. It
[favors](https://arxiv.org/abs/2404.13076) its own outputs. Devloop makes Codex implement and
Claude Code review adversarially (or vice versa),
until all criteria are met and reviews addressed, so
humans only come in where judgment pays off: at the
spec and PR sign-off stages.

## ## HOW IT WORKS

-
Spec:
`/devloop-spec`

skill generates a short markdown spec with a clear problem, outcome, acceptance criteria, and what's in scope as much as what's out. - Code: Codex implements the spec in isolated git worktrees.
-
Review: Claude
Code runs
`/devloop-review`

against the spec to ensure all acceptance gates along with engineering gates (security, maintainability, correctness, simplicity, etc.) are met. - Loop: Iterates on reject, stops on accept or max passes, and generates a PR ready for human review.

## ## SKILLS

Devloop ships two skills, installed for both agents:

-
`/devloop-spec`

: turns a rough idea, notes, a URL, or an interview into one concrete, devloop-ready spec. -
`/devloop-review`

: judges each pass against the spec and engineering quality gates, returning ACCEPT, REJECT, or UNCLEAR with fix instructions.

## ## WHEN (NOT) TO USE

### When to use

- The work can be pinned down in a clear spec through an interview with the coding agent.
- The change is large or multi-step enough that adversarial review actually catches something.
- You want the agents to iterate unattended and only pull you in for the spec and PR sign-off.

### When not to use

- The task is one-shotable and the scope is tiny: just ask the agent directly.
- The work is exploratory or open-ended, where acceptance criteria can't be written up front.
- You want hands-on control of every edit rather than a loop running on its own.

## ## FAQ

## Will it touch my working tree?

No. It runs in isolated sibling worktrees and
never pushes or opens a PR unless you pass
`--create-pr`

. Pass
`--in-place`

to work in your
current checkout instead.

## Which agent codes, which reviews?

Codex codes and Claude reviews by default.
Swap either with `--coder`

/
`--reviewer`

.

## Can it run my tests?

Yes. Add a `.devloop/verify`

hook
with your build, lint, or test command. It
gates acceptance and blocks the loop when it
fails.

## When does the loop stop?

On ACCEPT, on a stall, or when it runs out of passes (up to 5 by default, clamped 1-10). A 30-minute timeout caps each run.

## Any telemetry?

None. One auditable Bash script; tracks and
reports land under `.devloop/`

.