
from crush-skill42
Create a private, local AI profile of a romantic interest from chat logs, photos, and social media using parsers and Bayesian tagging — for personal analysis an
Crush Skill ingests personal conversational artifacts (chat logs, social posts, photos) and runs a pipeline of parsers, Bayesian tagging, and persona/memory construction to produce a local Skill that can simulate conversations and provide a "Crush Health" analysis. It includes parsers for WeChat/QQ/social exports, photo EXIF analysis, and utilities to snapshot versions.
Use when a user wants to analyze past interactions, practice conversations, or build a formative persona for private reflection. Not for harassment, stalking, or deception — the skill is explicitly bound by safety rules and should be used for therapy-style reflection and practice only.
Designed for Claude Code and OpenClaw environments that support Python tooling and shell commands; integrates with local CLI and Skills UI.
Crush skill creates AI profiles of romantic interests from chat logs, photos, and social media using Bayesian tagging and attachment-style analysis. No scripts were bundled for testing — the referenced Python parsers (wechat_parser, qq_parser, etc.) are not included. SKILL.md is well-written with clear workflows and safety rules, though the niche emotional-analysis focus limits broad applicability.
Safety rules explicitly prohibit harassment/stalking/privacy violations — good. The /delete-crush command uses rm -rf on a scoped directory which is acceptable. No bundled scripts means static analysis only. Skill is functional as a prompt-based workflow but heavily depends on external parser scripts not included in the skill package.