Your coding agent has the code. Does it know where you left off?

Itay Zrihan — Tue, 06 Oct 2026 08:35:44 +0000

You open a project, start a fresh agent session, and hit the familiar question: what were we actually doing here?

The code is in Git. The plan might be in a chat. A check passed at some point, but you've changed three files since then. Before you can get moving, you're piecing the story back together.

I'm Itay from VXU Labs. We just released DIP, short for Development Intelligence Platform, to give that project context somewhere to live alongside the code.

Keep the handoff with the project

DIP stores plans, unfinished tasks, checkpoints and verification evidence in a .dip/ directory. Its event files can travel with normal Git commits.

With supported hooks enabled and trusted, integrations for Codex and Claude Code capture development activity and bring compact context into a new session. The agent still needs to explain meaningful intent. A stream of file changes cannot tell the whole story of why you made a decision.

There is also a local dashboard for projects, tasks, activity and checkpoints, plus coordination across Git worktrees on one machine.

Done and checked are different things

Here's the part I'd really like developers to poke at.

You configure a named check, DIP runs it, and the result is tied to the code snapshot that was tested. If relevant code changes afterwards, that evidence becomes stale.

That keeps three questions separate: what state is the task in, what has been verified, and what is present in the selected Git history? A green check is useful when you know what it actually checked.

Give it a small test drive

The README covers installation. You'll need Node.js 24 or newer.

For a first look, use the documented single-project path in a disposable Git repository. Create a task, save a checkpoint explaining what remains, close the session, and come back to it. Is there enough context to continue without digging through the old conversation?

The full installer changes machine integrations, including agent settings and Git hooks, so read those setup notes before running it. Automatic capture depends on hooks being loaded and trusted. Live ownership coordination is local to one machine; Git alone does not provide distributed locking.

DIP is an early release under Apache-2.0. It makes no model or API calls of its own. Your coding agent still uses whatever model and account you already configured.

Want to help make the next first run better?

We've opened three concrete places to start:

A single-project walkthrough, scoped as a documentation contribution.
First-run compatibility reports for one OS, shell and agent combination.
A keyboard accessibility audit of a single dashboard flow.

Pick a small piece and comment on the issue with your intended scope. The contributor guide covers setup, isolated testing and commit sign-off.

Source and setup on GitHub | Product overview

If you give it a spin, tell us what the next session still needs to know. A clear report of missing context is a useful contribution too.

DEV Community: Itay Zrihan

Your coding agent has the code. Does it know where you left off?

Keep the handoff with the project

Done and checked are different things

Give it a small test drive

Want to help make the next first run better?