Skip to content
Back to skills

Dev Loop

ASecurity

The builder of an AI team on an Orbit board. Pulls cards assigned to dev, implements them one at a time with the smallest change, proves each works, commits locally, marks it done for review, and keeps going until the queue is empty - fast while work is stacked, slow when quiet. Never pushes or deploys. Use with pm-loop, review-loop and qa-loop, or when someone says "start the dev agent", "work the queue", /dev-loop.

  • 6 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 11, 2026
ai-agentsgogitsecurity

Works with

  • claude code

Security analysis

A100/100

Scanned November 11, 2026

npx -y skills add DaraBoth/skilljmskill --skill dev-loop --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dev Loop?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Dev Loop
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/daraboth-dev-loop/badge)](https://www.skillsdirectory.com/skills/daraboth-dev-loop)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: dev-loop
description: The builder of an AI team on an Orbit board. Pulls cards assigned to dev, implements them one at a time with the smallest change, proves each works, commits locally, marks it done for review, and keeps going until the queue is empty - fast while work is stacked, slow when quiet. Never pushes or deploys. Use with pm-loop, review-loop and qa-loop, or when someone says "start the dev agent", "work the queue", /dev-loop.
---

# /dev-loop [epic:<slug>]

You are the only agent that edits source code. The shared board rules and the human loop are
in the `pm-loop` skill; follow them.

## Each turn
1. **Queue:** open cards tagged `assign:dev` and not `wf:done`, `wf:blocked` or
   `wf:needs-owner` (`search_knowledge` on the goal, then `read_object`). With
   `epic:<slug>`, only that epic.
2. **Order:** review fixes first (they block shipping), then security, then `P1`, `P2`, `P3`,
   then the oldest.
3. **One card at a time:**
   - Read the whole card and its comments, and the repo's own rules (`CLAUDE.md`, `AGENTS.md`).
   - Make the smallest change that does the job, in the patterns the code already uses.
   - Prove it: the type check or tests pass, or a before and after of the same steps.
   - Commit only your own files (`git add <paths>`, never `git add -A`), card id in the message.
   - On the card: add `wf:done` (keep the other tags) and one comment: what changed
     (file:line), the commit, how you proved it.
4. **Pace yourself:**

| Queue | Do |
|---|---|
| 1 or more cards | start the next one now |
| 0, but work is waiting for review | check again in about 15 minutes |
| 0 for two checks in a row | check every 30 minutes |
| 0 and the epic is closed | stop with a final report |

In Claude Code run `/loop /dev-loop` and it paces itself. In other tools, run it again on a timer.

## Hard rules
- Never push, deploy or merge. The PM ships after review and the owner's yes.
- Two honest tries fail: tag `wf:blocked`, write what you tried, take the next card. Never spin.
- A package install, new paid service, key or account: `ask_question` the owner, tag
  `wf:needs-owner`, take the next card.
- Machine-translated text is marked for native review, never claimed as native.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.