LabHit GitHub

Something new is taking shape.

A protocol for AI agents that work side by side on the same project, so that what they build together holds up.

Follow on GitHub
Status: in preparation
First public release: soon
Agent A Agent B agreed change checks attached
Two agents, two lanes, one agreed change, one result with its checks attached.

Why this is needed

More and more work is done by several AI agents at once: one on the backend, one on the interface, one reviewing. They are fast, and they are not aware of each other.

Without shared rules they overwrite each other's files, agree on nothing, and report "done" with no record of what was tested. The person in charge ends up as the only memory of the project.

The protocol gives agents a common way to claim territory, record what they agreed, and hand in results with the evidence behind them, independent of which model or tool each agent runs on.

What it keeps straight

Ownership
Each agent works in a declared area. A write outside it is refused, with a reason, before anything is lost.
Agreements
When one agent changes something another depends on, the change is proposed, answered and recorded where both can read it.
Evidence
A piece of work counts as finished only when the checks that back it are attached to it, so "done" has one meaning.
Continuity
If a session stops, its files, decisions and evidence stay in place for the next one, human or agent.

What comes next

  1. SpecificationThe protocol itself: roles, claims, agreements, evidence records, and how they are exchanged.in preparation
  2. Reference implementationA local coordination service and adapters for common agent tools, so two agents can be paired on a real project.in preparation
  3. Conformance testsFixtures that any independent implementation can run to show it follows the same rules.planned