Project-Manager-as-a-Service for autonomous coding agents. LoopRR coordinates a multi-agent software-development "dream team" — Claude Code routine workers you configure in your own Claude app — into a single structured, stateful loop across one or more GitHub repos, with a human in the loop. Every action an agent takes is stamped into a governed, append-only temporal+spatial ledger (who did what, in which repo/branch/worktree, when, traced to which commit/PR/CI run) — a flight recorder and shared brain for your fleet — and nothing reaches a protected branch except through a human-gated, GitHub-enforced merge.
The RR: Round-Robin (cyclic scheduling/load-balancing of work across the team) + Record-and-Replay (capturing a run so future work can be refined, audited, and re-coordinated).
Status: Phase 0 — foundation. This repo currently holds the harness, memory, specs, roadmap, and grounding research. No product code yet. See
ROADMAP.mdfor the plan andHUMAN.mdfor the decisions awaiting sign-off.
- Build protocol & invariants:
CLAUDE.md - The operating loop (and the product):
docs/ultraloop.md - What & why:
foundation/vision.md,foundation/prd.md - How:
foundation/architecture.md,memory-and-ledger.md,coordination.md,worker-team.md - Progress:
ROADMAP.md· Human channel:HUMAN.md· Decisions:DECISIONS.md - Grounding:
research/README.md
LoopRR is being built by pivoting the governed-API spine of Knovo (an AI-authored, human-gated
content engine). The reuse/transform map is in foundation/knovo-reuse.md.