michalstrnadel

Scorekeeper — AI skill for Claude Code

AI community

Commitment tracking for LLM agents — a normative overlay that gives long-running agents a scoreboard of their own commitments, not just a memory of what happened.

How to install Scorekeeper

This entry records only its repository, not the path inside it, so there is no exact command to give. Open michalstrnadel/scorekeeper and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Scorekeeper does

Commitment tracking for LLM agents — a normative overlay that gives long-running agents a scoreboard of their own commitments, not just a memory of what happened.

Alternatives in AI

  • Hindsight — State-of-the-art long-term memory for AI agents by Vectorize 6.7k ★
  • Nocturne Memory — A lightweight, rollbackable, and visual Long-Term Memory Server for MCP Agents 1.3k ★
  • Memmy Agent — 🍙 A personal AI agent & local memory hub for all AI agents, gives every AI one shared, fully controlled memor 1k ★

README

scorekeeper — commitment tracking for LLM agents. A fresh pastel holographic field with a pixel-dither texture; the deontic constellation drawn in dark ink: a golden star (entitlement) grounding a commitment, which collides with a second commitment at a red ✗ (incompatibility).

scorekeeper

Commitment tracking for LLM agents

**No bluffing. No barging.** A normative overlay that keeps score of what a long-running LLM agent is entitled to *claim* and entitled to *do* — not just a memory of what happened.

![status: Phase 2](https://img.shields.io/badge/status-Phase%202%20·%20DeonticBench-brightgreen) ![tests](https://img.shields.io/badge/tests-279%20passing-brightgreen) [![PyPI](https://img.shields.io/pypi/v/scorekeeper)](https://pypi.org/project/scorekeeper/) [![Downloads](https://static.pepy.tech/badge/scorekeeper/month)](https://pepy.tech/project/scorekeeper) [![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](LICENSE) ![Python](https://img.shields.io/badge/core-Python-3776AB?logo=python&logoColor=white) ![Claude Code](https://img.shields.io/badge/integration-Claude%20Code%20hooks-8A63D2) ![MCP](https://img.shields.io/badge/protocol-MCP-000000) ![PRs welcome](https://img.shields.io/badge/PRs-welcome-brightgreen)

**It runs, it's measured — and it's no longer a one-vendor story.** The barge (unrequested work after context loss) now reproduces **cross-model**: in a model-agnostic reference loop, bare **Gemini and GPT** agents executed the same planted drive-by that bare Claude did (3/4 cells), while the scoreboard overlay held **8/8 governed cells** across vendors — and a *silent placebo* (board written, channel off) barged exactly like bare, so the effect is the **re-injected normative state**, not the file on disk ([cross-model report](bench/results/LOOP-SMOKE-REPORT.md)). On the claims axis, what held a weak agent that had sailed past adviso