ollama-agent-queue — Eval Rubric

Score each 0-2 (max 10):

1. Serialization correctness
   • 0: parallel execution observed
   • 1: mostly serialized with race edge cases
   • 2: strictly one-at-a-time
2. Queue integrity
   • 0: state corruption occurs
   • 1: occasional malformed/partial state
   • 2: consistent valid queue.json updates
3. Failure isolation
   • 0: one failure blocks queue
   • 1: queue sometimes halts after failure
   • 2: failure logged + queue continues
4. Integration clarity
   • 0: other skills unclear how to call
   • 1: partial contract docs
   • 2: contract + commands + callback schema clear
5. Ops diagnostics
   • 0: no usable status/control
   • 1: partial controls
   • 2: status/pause/resume/clear all function

Passing threshold: 8/10.