ex-skill: Best Claude Code Skills for Claude Code Developers in 2026

What Is ex-skill?

ex-skill is a Claude Code skill built by therealXiaomanChu that ingests WeChat chat exports, QQ messages, photo EXIF data, and user descriptions to create personalized AI agents mimicking ex-partners' personas and relationship histories. Hosted on GitHub with 2.9k stars and 302 forks as of October 2024, it follows the AgentSkills standard for seamless integration into Claude Code environments. ex-skill is one of the best Claude Code skills for Claude Code developers building personal AI agents, enabling therapy-grade recollections via parsed memory timelines and tag-based personality layers like MBTI and attachment styles.

Quick Overview

Who Should Use ex-skill?

• Claude Code power users scripting custom agents from raw chat data who need automated parsing of WeChatMsg or PyWxDump formats without manual JSON conversion.
• Indie developers experimenting with persona AI for emotional simulations, integrating relationship timelines from photo metadata and QQ txt exports into prompt chains.
• AI hobbyists with Claude Pro handling 5k-15k token intakes per skill to build multi-version agents with rollback via /ex-rollback commands.

Not ideal for:

• Developers requiring production-scale multi-user AI without Anthropic API keys, as it ties to single-user Claude Code sessions.
• Teams needing GPU-accelerated local inference, since ex-skill runs fully cloud-side on Claude's models.
• Users without Node.js 18+ setups, as Claude Code installation demands it for skill discovery in .claude/skills/.

Key Features of ex-skill

• Multi-format Data Ingestion — Parses WeChatMsg, PyWxDump, QQ txt/mht, social screenshots, and JPEG/PNG EXIF for timelines; extracts inside jokes and locations in under 15k tokens per build.
• 5-Layer Persona Builder — Structures output as hard rules → identity → speech style → emotion patterns → relationship behaviors, supporting 16 MBTI types, 12 zodiac signs, and 20+ tags like "anxiety attachment" or "read receipts avoidance".
• Relationship Memory Extraction — Builds timelines of dates, arguments, and sweet moments from chat logs; merges incremental data via merger.md prompts without overwriting base versions.
• Command-Driven Management — Supports /list-exes, /{slug}, /{slug}-memory, /ex-rollback {slug} {version}, /delete-ex {slug}, and /let-go {slug} for versioned control in Claude Code terminals.
• Correction and Evolution — Users input "ta wouldn't say that" to append Correction layers; auto-merges new chats or photos into Part A (memory) and Part B (persona).
• Tag Translation System — Converts Chinese labels like "嘴硬心软" (tough exterior soft heart) into English equivalents via persona_analyzer.md, ensuring cross-lingual consistency.
• OpenSkill Compliance — Generates SKILL.md with frontmatter, memory.md, and persona.md files in .claude/skills/{slug}/ for native Claude Code execution.

ex-skill vs Alternatives

Claude Code Canvas suits visual builders preferring drag-and-drop over CLI parsing, but misses ex-skill's specialized WeChat tools. Claude Context Mode excels at stuffing 200k tokens into sessions yet skips persona structuring from photos. Character.AI offers quick web setups without dev setup, ideal for non-coders avoiding Node.js installs.

How ex-skill Works

ex-skill operates as a create-ex skill in Claude Code's .claude/skills/ directory, triggered by /create-ex to intake data via intake.md prompts. Python tools in /tools/—like wechat_parser.py for WeChatMsg JSON and photo_analyzer.py for EXIF—preprocess inputs into structured facts. Prompts then split outputs: Part A (Relationship Memory) compiles timelines and jokes; Part B (Persona) layers rules from tags like ENFP or avoidance attachment.

Runtime flow: Incoming message hits SKILL.md → persona.md dictates response style (e.g., wordsmith for Gemini, terse for INTJ) → memory.md injects context ("that Italian place we hated") → correction_handler.md overrides if flagged. No local storage; all persists in Claude Code's skill folder with git versioning from the host repo.

Generated skills support modes: full chat via /{slug}, memory recall with /{slug}-memory, or persona-only /{slug}-persona. Evolution appends via new /create-ex runs, merging deltas without full rebuilds.

# In Claude Code project root
mkdir -p .claude/skills
git clone https://github.com/therealXiaomanChu/ex-skill .claude/skills/create-ex

# Trigger in Claude Code
/create-ex

This clones the repo into skills dir, installs optional pip3 install -r requirements.txt for parsers. /create-ex prompts for ex-code, traits (e.g., "ENFP, night owls"), and uploads chats/photos; outputs {slug} skill folder ready for /{slug} chats.

Pros and Cons of ex-skill

Pros:

• Zero-config install via git clone into .claude/skills/, auto-discovered by Claude Code without YAML tweaks.
• Handles dense Asian formats: WeChat exports yield 90%+ accuracy on speech patterns from 10k+ message histories.
• Versioned persistence: /ex-rollback restores prior builds, tracking 25+ commits like repo's main branch.
• Low overhead: 5-15k tokens per skill, runs on Claude Pro quotas without GPU or Docker.
• Extensible prompts: Edit prompts/persona_analyzer.md to add custom MBTI rules or love languages.
• Ethical guardrails: README mandates personal use only, with /let-go for deletion.

Cons:

• Claude Code dependency: Requires Node.js 18+ and Anthropic access; no standalone web UI.
• Token costs scale with data: Large photo sets or year-long chats hit 15k+ tokens per update.
• Parser limits: QQ mht needs manual export; no Telegram or iMessage support yet.
• Chinese-centric: Tags like "报复性熬夜" auto-translate but lose nuance in full English mode.
• No multi-ex concurrency: Commands prefix per slug, risking crosstalk in shared sessions.

Getting Started with ex-skill

Start by ensuring Claude Code runs with Node.js 18+. Clone to project or global skills dir for reuse across repos.

# Global install for all projects
git clone https://github.com/therealXiaomanChu/ex-skill ~/.claude/skills/create-ex
pip3 install -r ~/.claude/skills/create-ex/requirements.txt  # Optional for parsers

# In Claude Code terminal
/create-ex

Enter ex-nickname, traits ("话痨, 双子座"), upload WeChat JSON or photos. ex-skill parses into memory/persona files, outputs slug like ex-jane. Run /{slug} for chats; expect persona-matched replies with memory injections in 2-5s latency. Configure Anthropic API key in Claude Code if exceeding Pro limits; test /list-exes lists all.

Verdict

ex-skill stands as the strongest Claude Code skill for developers recreating ex-partner agents from real chat data when Asian messaging formats dominate inputs. Its parser-to-prompt pipeline delivers precise 5-layer personas unmatched by generic builders, though token budgets limit massive histories. Install via git clone and iterate with corrections for therapy-ready AI.