GNAP: proposal for git-native agent coordination instrumentation contrib package

Proposal: opentelemetry-instrumentation-gnap contrib package for AI agent coordination

OpenTelemetry Python has excellent coverage for web frameworks, databases, and cloud services in opentelemetry-python-contrib. As AI agent systems proliferate, coordination between agents becomes an important layer to observe.

GNAP (Git-Native Agent Protocol) coordinates AI agents via a shared git repo: tasks move through board/todo/ → board/doing/ → board/done/. Every transition is a git operation — a natural instrumentation point.

Proposed: opentelemetry-instrumentation-gnap

Following the pattern of existing OTel instrumentations:

from opentelemetry.instrumentation.gnap import GNAPInstrumentor

GNAPInstrumentor().instrument()

# Automatically creates spans for:
# - gnap.task.create (todo→ board)
# - gnap.task.claim  (todo→doing)  
# - gnap.task.complete (doing→done)
# - gnap.task.timeout or gnap.task.fail

# Span attributes:
# gnap.task.id, gnap.task.type, gnap.agent.id
# gnap.board.repo, gnap.transition.duration

This would plug into any OpenTelemetry backend (Jaeger, Zipkin, AWS X-Ray, etc.) and give multi-agent system operators visibility into:

1. Task coordination latency (time in todo/, doing/ states)
2. Agent utilization (how many tasks per agent)
3. Coordination bottlenecks (tasks stuck in doing/)

Why this belongs in opentelemetry-python-contrib:

GNAP is protocol-level (just git + files), so the instrumentation is minimal and focused. It follows the same pattern as existing resource-level instrumentations. As multi-agent systems become more common, coordination observability will be as important as HTTP or DB tracing.

Spec: https://github.com/farol-team/gnap

Start by reviewing the GNAP specification and the existing instrumentation patterns in opentelemetry-python-contrib. No repository files, entry points, or tests are named; done would require a documented GNAP instrumentation package that creates the proposed task-transition spans and attributes, with tests confirming the behavior.