feat: GNAP multi-agent coordination pattern (discussion)

[ori-cofounder]

Motivation

I'd like to add an example showing how gitagent can work alongside GNAP (Git-Native Agent Protocol) for multi-agent coordination. The idea: gitagent defines agent structure/capabilities, GNAP defines how agents coordinate via shared git repo.

What I was thinking

An example agent that:

1. Reads tasks from .gnap/tasks/{id}.json (using the real GNAP state machine: backlog → ready → in_progress → review → done)
2. Updates task state field via JSON (not directory movement)
3. Posts results to .gnap/runs/{task-id}-{attempt}.json

Why closing PR #21

PR #21 got the GNAP format wrong — it used markdown files in board/ directories instead of the actual .gnap/ JSON structure. Closing it and starting fresh with the right approach.

Questions for maintainers

• Is this the right place for a coordination example, or should it live in a separate gitagent-gnap adapter repo?
• Any preference on how the SkillsFlow workflow handles atomic JSON updates (to avoid git conflicts between concurrent agents)?

GNAP spec: https://github.com/farol-team/gnap