This document is the source of truth for what Cooperations currently implements versus what remains experimental.
- CLI commands:
run,status,history - Dry-run routing without requiring model CLIs
- Heuristic role routing for
architect,implementer,reviewer, andnavigator - Per-task sandbox contract for real runs
- Optional task-scoped capability references for plugins and skills
- Local task, handoff, and artifact persistence with sandbox/capability metadata
- Codex CLI as the default runtime adapter on Linux/bash
- Shared workflow event model used by the orchestrator and interfaces
tui: supported secondary interface for interactive workflow execution, with plain fallback output when no controlling TTY is availablegui: exploratory/demo interface built on Gio and enabled explicitly with-tags gio
- RVR and confidence-scoring documents
- Meta-cognitive reasoning and long-context research notes
- Planning documents and design explorations
- Historical benchmark-style and evaluation notes
- Learned or adaptive routing beyond simple heuristics
- Confidence-driven control flow in the default runtime
- Wired RVR execution in the default runtime path
- Benchmark automation or reproducible performance evaluation pipeline
- OpenAI API integration or SDK-based provider runtime
- Implicit sandbox defaults for real tasks
- Production-readiness claims for the GUI or research layers
- Default builds that assume desktop GUI system libraries are present