AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻译准备中
尚未检测到更新。GitHub 发布版本后会显示在这里。
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权,自行管理介绍,无需管理员批准。
登录并生成验证文件,在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件,即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录,将前往登录页面。登录后请回到此处,再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify,即可验证所有权并自行管理介绍。
一分钟完成验证Ready-to-use AI agent that reviews pull requests against your team's rules and answers questions about your code. Built on Hubzoid.
翻译准备中
取自仓库 README
正在加载完整 README。
查看完整 README ↗The Engineering Desk is a ready-to-use Hubzoid hub for engineering teams. It reviews pull requests against your team's rules, answers questions about your code in chat, watches dependencies for security problems and writes a weekly digest. It runs on your machine or your server, with the Claude Code or Codex you already sign in to.
Each repository you list gets three switches, all off until you turn them on. Chat needs no switch.
reviews). Every two minutes, or within seconds through a webhook, the desk reviews each new pull request and push against your team's rules, the repository's own rules and a profile of the repository. A second pass tries to refute each finding, so what reaches you has survived a challenge. It starts in preview, showing the comment it would post and writing nothing to GitHub, and it tracks which findings get fixed.security). Checks dependencies daily against known advisories and audits the code weekly. It alerts only when something new and serious appears.reports). A weekly digest and monthly code and delivery reports, as pages in the web app and, if you add a list, email.Reviews, findings and schedules live in .hubzoid/. A restart carries on where it stopped, and each comment is posted once. A review is usually one model call plus a verification call, and daily limits are in workflows/settings.yaml. Setup guide
You need Python 3.11 or 3.12, Git, and Claude Code or Codex signed in on this machine. No API key. The quick start uses the fictional sample, so you need no GitHub token.
Paste this into Claude Code or Codex:
Clone https://github.com/hubzoid/engineering-desk-hub into a folder called
engineering-desk, cd in, and follow SETUP.md to run the fictional sample.
Keep publishing to GitHub off until I say so.
Or by hand:
git clone https://github.com/hubzoid/engineering-desk-hub engineering-desk
cd engineering-desk
python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
cp .env.example .env
.env uses Claude Code (MODEL=claude-local). For Codex, set MODEL=codex-local.
In workflows/settings.yaml, change source: github to source: sample. Then give yourself, the local owner, the chat tools and the delivery report, and start:
hubzoid grant admin@localhost engineering_desk .
hubzoid grant admin@localhost wf_delivery_report .
hubzoid run .
Open the address hubzoid run prints. Within two minutes the desk reviews the five sample pull requests: an endless retry, an unprotected admin export, unsafe SQL, a vulnerable package and one clean change. Nothing is posted anywhere. Ask in the web chat:
What is waiting or stuck right now?
Then ask "Show the findings on storefront PR 102 with fix prompts." You see each finding with a prompt you can hand to your own coding assistant to fix it. Full setup
| Works with | What the desk does |
|---|---|
| GitHub | Reviews pull requests and pushes in the repositories you list. Publishes a comment, inline findings and a status once you turn publishing on. Comment /hubzoid review, pause, resume, dismiss, defer or help on a pull request to act on it. |
| GitHub's MCP server, read only | Gives chat issues, pull requests, Actions runs and logs, releases and branches, for any repository the token can read. It never writes. |
| The web chat, and Claude Code over MCP | Answers questions about your code. Its own tools answer signed-in people only. |
| Sends reports and the weekly digest to a list, if you add one. |
Ready for your own repositories? Add a fine-grained, read-only GitHub token to .env yourself, never in chat (SETUP.md, step 2, keeps it off your screen and out of your shell history). List your repositories in workflows/settings.yaml with source: github, then run hubzoid hub doctor . and hubzoid run .. Or ask your coding assistant to follow SETUP.md.
The files you will edit, and what each one changes:
engineering-desk/
├── .env your model and your GitHub token. Never committed
├── workflows/settings.yaml your repos, what runs on each, limits, preview or publish
├── knowledge/
│ ├── review-rules/ your team's rules, one Markdown file each
│ │ └── examples/ rules to copy and adapt
│ └── *.md conventions you want chat and reviews to know
└── connectors/.mcp.json GitHub's read-only MCP server for chat
Choose what runs on each repository in the engineering_desk: block of workflows/settings.yaml:
engineering_desk:
repos:
- name: your-org/api
reviews: true
security: true
- name: your-org/web
reviews: true
reports: true
- name: your-org/docs # chat and a profile only
Reviews start in preview. When you like them, give the token write access and set review.post_to_github: true, and the desk comments on pull requests. A GitHub App gives it its own name. SETUP.md, "Publish reviews"
A team rule has a title, the paths it covers, a severity and an optional pattern. Copy one of the examples to start. In a repository, a .engineering-desk/review.md file sets what to skip, focus on or ignore there, and the desk also reads the repository's CLAUDE.md, AGENTS.md and CONTRIBUTING.md.
To save a rule from chat, tell the desk "Remember: retries must stop after three attempts." It drafts the rule, and an admin approves it before it applies. Your .env, .hubzoid/, workflows/settings.yaml and your rules are yours: updates never change them.
/hubzoid commands count only from people with write access to the repository, and only outside quotes and code blocks.SETUP.md, "Troubleshooting and costs", covers limits and what a review costs.
The desk runs on a server the same way any Hubzoid hub does: one durable folder, sign-in on, and one public address. Three things are specific to the desk:
HUBZOID_SCHEDULES=1, so sweeps, profiles, security checks and digests run on their own. It is set in .env.example.HUBZOID_WORKFLOW_USER, the account scheduled work runs as. Keep the folder, .hubzoid/ included, on persistent storage, and back up with hubzoid backup before upgrades.trigger: webhook and WEBHOOK_SECRET_GITHUB, and point GitHub at https://<your hub>/webhooks/engineering-desk/github. Polling every two minutes needs no public address.Grant each person what they need in Settings, Agents, Engineering Desk, Access: chat and tools, their own reports, approving team rules, or activity counts for digest groups. Counts describe activity. They never score people. SETUP.md, "Bring your team", has the permission table. Hubzoid deployment lists the hosting options and which are verified.
git pull
python -m pip install -r requirements.txt
hubzoid hub doctor .
hubzoid restart engineering-desk
If pip upgraded Hubzoid itself, stop hubzoid run and start it again. UPGRADING.md has the details, the move from the old GitZoid names, and the move from the GitZoid bundled with Hubzoid 1.3.
| Next step | Guide |
|---|---|
| Install, add a token and choose what runs | Setup |
| Publish reviews and use the merge gate | Publish reviews |
| Give your team access | Bring your team |
| Reviews within seconds | Instant reviews |
| Fix findings with your own coding agent | Fix findings |
| Every option | Options |
| Update, and move from GitZoid | Upgrading |
| What changed | Changes |
Questions go to GitHub Discussions. Bugs in the desk go to this repository's issues.
The Engineering Desk is Apache-2.0. Built on Hubzoid. Need it running for your team? Build with us.