This skill distills a past relationship into a runnable agent skill. It ingests user-provided materials (WeChat/QQ exports, screenshots, photos, pasted memories), extracts relationship memory and a multi-layer persona, and writes a complete skill folder (memory, persona, meta, SKILL.md) so the user can interact with a simulated version of the person. The emphasis is on private, local use and ethical safeguards.
Use when a user wants to preserve or rehearse memories, explore a persona derived from archived conversations, or create a private interactive memory artifact. Triggers include commands like /create-ex, 帮我创建一个前任 skill, I want to distill an ex, or follow-ups like /update-ex {slug} and /list-exes.
The skill is written for Claude Code / Claude-style environments and references Bash/python helper scripts in its workflow (WeChat/QQ parsers). It is compatible with local agent runtimes that can run Python/Bash and provide Read/Write/Edit tools.
A dual-language (CN/EN) skill that creates AI personas mimicking an ex-partner from chat logs, photos, and social media exports. No bundled scripts were available to test. The SKILL.md is well-structured with clear 5-step flow and evolution/correction modes. Notable privacy concern: processes highly personal data (WeChat, iMessage, SMS) without encryption or data-handling safeguards.
Dual-language SKILL.md is very long (~2000 words) — English version could be moved to references/ for leaner SKILL.md. Deletion via rm -rf exes/{slug} has only conversational confirmation. No scripts were bundled in the fetched record so runtime behavior could not be tested.
