os-improvement-loop

A multi-agent OS improvement loop coordinates an ORCHESTRATOR with INNER and PEER agents through a standardized Triple-Loop protocol. It runs evaluation cycles, maintains memory, and writes retrospective reports to ensure measurable system upgrades.

The Triple-Loop protocol is a two-tier coordination mechanism combining outer improvement orchestration with inner evaluation. It uses an improvement ledger, memory manager, and surveys to gate KEEP or DISCARD decisions across multiple agents.

You track friction events and memory by maintaining an improvement ledger and per-cycle retrospectives within the loop. When friction escalates, the system triggers a Full Loop retrospective to persist improvements and tune cross-cycle memory.

Yes, the loop supports ORCHESTRATOR, PEER_AGENT, and INNER_AGENT roles in a single repeatable protocol. This allows you to run parallel audits of OS components, collect KEEP or DISCARD results, and maintain isolation across agents.

The best way to gate decisions is by using strict evaluation baselines via an eval runner. The loop enforces gating through standardized surveys and eval baselines to ensure only validated improvements are kept and persisted to memory.