User Behavior Personalization 0.2.0

Keep scoped preferences in a subject-bound local profile, commit sourced corrections transactionally, and load the current profile before invoking a real CLI host. Add pause/export/forget controls and deleted-source replay rejection.

Validation: all 3 repository tests, real MCP calls, and an extracted-package correction/pause/delete flow passed. A real Codex invocation loaded the saved preference and followed it. CI passed on macOS, Linux, and Windows. Cursor tool discovery passed; its model conversation remains blocked by local access limits. Automatic identity mapping, history ingestion, and reflection scheduling are not included.

Use Python 3.11+. English and Chinese guides and Apache-2.0 notices are included.

user-behavior-personalization 0.1.1

Keep useful preferences in a profile the user can inspect and correct. This skill describes how an agent should turn explicit feedback into scoped guidance without quietly treating a guess as a personal fact.

This release adds selectable English and Simplified Chinese documentation, project-specific setup and practice guidance, and versioned visual evidence. The archives were extracted and checked against the current source; applicable local installation and example checks passed. Original license terms are preserved.

English guide · English README

简体中文

把有用偏好写进用户能检查和更正的资料中。这个技能规定如何将明确反馈整理为有范围的使用指导，避免把猜测悄悄记成个人事实。

本次更新加入可切换的中英文文档、按项目编写的安装与实践指南，以及带版本的图片或结果示例。发行包已解压并核对当前源码，适用的本地安装与样例检查通过，原许可证保持不变。

中文指南 · 中文介绍

Use the named ZIP assets and their matching checksums. GitHub’s automatic source archives are separate downloads. Package checks do not certify live host integrations, real devices, or production deployment.

请使用下方命名 ZIP 与对应校验文件；GitHub 自动生成的源码压缩包属于另一种下载。包检查不等于真实宿主集成、设备或生产部署已通过。

User Behavior Personalization v0.1.0

User Behavior Personalization v0.1.0

The first public release of KarmenLoop's evidence-backed personalization skill for AI agents.

Included

• A complete English-language SKILL.md for learning communication preferences and working habits from authorized conversations.
• A stable controller and a separate, evolving per-user USER_SKILL.md.
• Daily reflection, immediate correction, selective recall, and versioned update workflows.
• Evidence categories, scoped preferences, deduplication, and user-controlled forgetting.
• Blank profile, reflection, and state templates.
• Synthetic conversations, a sample user guide, and behavioral evaluation scenarios.
• Detailed quick-start, architecture, host-integration, privacy, contribution, support, and security documentation.
• A standard-library Python package checker and an optional GitHub Actions template.

Getting Started

Download and extract user-behavior-personalization-v0.1.0.zip, then read README.md and docs/quickstart.md. Keep the supporting directories with SKILL.md when installing the package.

Store real user profiles outside the public package. Configure verified user identity, memory authorization, storage, and reflection events in the agent host.

Validation

• Package checks passed for the release source: local file links, required artifacts, fenced blocks, basic frontmatter, the blank state template, and publication hygiene.
• The skill-creation validator passed.
• YAML metadata and frontmatter parsed successfully.
• Five isolated packaging failure cases were rejected: a missing required file, a broken link, CJK text in the English package, a live-memory directory, and a populated state template.

These checks validate package structure. No end-to-end agent-host, background scheduler, device synchronization, or live-user learning test is claimed.

Runtime Scope

This is a Markdown instruction and documentation package. Daily scheduling, durable storage enforcement, account linking, cross-device synchronization, and deletion across external backups require host implementations.

The GitHub Actions workflow is supplied as an optional template under assets/github-actions/. It is not enabled automatically.

License

Apache License, Version 2.0. See the included LICENSE and NOTICE files.