<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This project, ex-skill, enables users to create AI personas mimicking past romantic partners by distilling digital footprints into interactive "Skills." It targets individuals seeking to preserve memories or simulate interactions with an ex-girlfriend, offering a unique tool for digital reminiscence by generating an AI that communicates in a specific individual's tone and style.

How It Works

The core approach synthesizes user-provided data—chat logs (WeChat, iMessage, SMS), social media exports, photos (EXIF), and manual descriptions—into a structured AI "Skill." This skill comprises "Common Memories" (relationship timeline, habits, emotional patterns) and a "Persona" defined by a five-layer structure (rules, identity, expression, logic, behavior). The system processes messages by assessing mood via Persona, enriching with Memories, and outputting in the target's tone. An evolution mechanism allows continuous improvement via data merging and corrections, supported by version management.

Quick Start & Requirements

• Primary install / run command: Clone the repo into specific Claude Code or OpenClaw directories (e.g., ~/.claude/skills/create-ex).
• Non-default prerequisites and dependencies: Optional Python dependencies can be installed via pip3 install -r requirements.txt.
• Usage: Initiate via Claude Code commands like /create-ex, then invoke with /{slug}. Management commands exist for listing, accessing components, rollbacks, and deletion.
• Supported data sources: WeChat, iMessage, SMS, photos (EXIF), Weibo, Douban, Xiaohongshu, Instagram, PDFs/images, direct text.
• Links: INSTALL.md, README_EN.md.

Highlighted Details

• The project's stated purpose is to "distill memories into a Skill, not to get back together, but to remember."
• Features a sophisticated 5-layer structure for defining the AI's persona.
• Includes an evolution mechanism for skill improvement via data merging and conversational correction.
• Supports nuanced personality/behavioral tags (e.g., "loves to act cute," "anxious attachment").
• All data processing is performed locally.

Maintenance & Community

No specific maintainer details, community channels (Discord/Slack), or sponsorships are mentioned in the README.

Licensing & Compatibility

Released under the MIT License, permitting broad commercial and non-commercial use, modification, and distribution.

Limitations & Caveats

AI Skill fidelity depends directly on input data quality; manual descriptions alone may yield less accurate results. Photo analysis is limited to EXIF metadata extraction, not image content processing.