Show HN：DevBox – 終結 AI 代理指令疲勞的執行合約

Navigation Menu

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

To see all available qualifiers, see our documentation.

DevBox is a minimal, language-agnostic contract that standardizes how local systems are started, validated, observed, and safely operated by humans and AI agents.

License

Uh oh!

There was an error while loading. Please reload this page.

danieljhkim/DevBox

Folders and files

Latest commit

History

Repository files navigation

DevBox

While working as a data engineer, I repeatedly ran into the same bottleneck: the gap between local development and validation.

A small code change often meant spinning up a cloud cluster just to test a single component. This slowed iteration, introduced unnecessary friction, and—most importantly—created a hard disconnect between my local environment and the system actually running the code.

That disconnect became even more apparent when working with AI tools. Agents couldn’t:

Validation lived “somewhere else,” and progress suffered because of it.

So I moved the cluster onto my local machine. Once everything ran locally, something clicked.

AI agents could execute commands, read logs, detect failures, and retry—without human glue code in between. Development stopped being a linear, line-by-line activity and became something else entirely.

We weren’t just coding anymore. We were orchestrating.

To make that orchestration explicit and safe, ChatGPT and I designed DevBox: a development box for your agent and your codebase.

DevBox is a lightweight, language‑agnostic development execution contract for humans and AI agents.

It defines:

DevBox is not a framework, container, or runtime.
It is a thin control layer around your existing project.

Quick Start (Template)

See QUICK_START.md for the full workflow.

Why DevBox Exists

Modern development environments are no longer operated only by humans.

AI agents can now:

But most repositories expose implicit, undocumented, and unsafe execution surfaces:

To enforce tighter rules, we are forced to maintain fragmented, tool-specific rules (.cursorrules, .claudecode, etc.) for every new agentic workflow. And with every new agentic IDE or CLI tools we adopt, more of these fragmented instructions clutter our codebase and our mind.

DevBox aims to solve this by providing a universal source of truth for all agentic workflows.

At the least, I hope it can provide some inspirations.

What DevBox Is (and Is Not)

✅ Is

❌ Is Not

DevBox wraps what you already have — it does not replace it.

Core Concepts

1. Commands

All project actions are normalized into named commands.

Examples:

Each command maps to a single, deterministic implementation.

2. Policies

Policies define what agents are allowed to do.

They specify:

This prevents accidental or malicious actions while enabling autonomy.

Policies are managed via named profiles (for example: readonly, safe-write, admin)
stored under .box/policies/. One profile is active at a time.

You can inspect or switch the active policy using:

3. Signals

Signals are machine‑readable outputs produced by the system:

Agents consume signals, not human intuition.

4. MCP Integration (Optional)

DevBox can be exposed to AI agents via MCP:

MCP is optional and explicitly opt-in.
DevBox does not depend on MCP; MCP adapters depend on DevBox.

To enable MCP wiring for VS Code:

This creates .vscode/mcp.json pointing to the DevBox MCP server under .box/. And installs and builds the MCP server.

DevBox itself is editor‑agnostic; MCP is an optional adapter where supported.

Directory Layout

Example Flow

Human or agent interaction:

The same flow works for:

DevBox CLI

When devbox is on your PATH, all core commands are available from anywhere inside the repository:

The CLI discovers the repo root automatically by locating .box/.

Design Principles

If it is not in DevBox, it is not supported.

Status

DevBox is an emerging pattern, not a formal standard.
It represents a convergence of:

Expect evolution — not churn.

Philosophy

Humans design and orchestrate.
Systems execute deterministically.
Agents iterate within guardrails.

DevBox defines the boundary between intent and execution.

DevBox Conformance: Core
(Extended and Agent-Ready supported via configuration)

License

MIT

About

DevBox is a minimal, language-agnostic contract that standardizes how local systems are started, validated, observed, and safely operated by humans and AI agents.

Topics

Resources

License

Security policy

Uh oh!

There was an error while loading. Please reload this page.

Stars

Watchers

Forks

Releases

  2

Packages

  0

Languages

Footer

Footer navigation