create-parents

Overview of create-parents skill

What create-parents does

The create-parents skill helps you turn memories, chat logs, photos, and voice notes into a reusable parent persona skill for Claude Code. Its real job is not just “character generation,” but building a structured memory layer plus speaking style so the result feels closer to how a parent actually cares, repeats phrases, and responds over time.

Who should install this create-parents skill

Best fit: people doing personal memory preservation, emotionally realistic prompt writing, or family-role simulation inside Claude Code. The create-parents skill is especially useful if you want more than a one-off prompt and need a repeatable workflow that creates files under ./parents/{slug}/ and supports later updates, listing, rollback, and deletion.

What makes it different from a normal prompt

A generic prompt can imitate “a caring mom” or “strict father,” but this repo adds a guided intake flow, analyzers for memory and persona, and tool-based parsing for exported chat data. For create-parents for Prompt Writing, the main differentiator is structure: it separates relationship memory from behavioral persona, which usually produces more stable and less cliché outputs than asking the model to “act like my parent.”

Main adoption caveats

This is built for Claude Code, not a generic chat UI. It assumes a skill-install location, tool access such as Read, Write, Edit, and Bash, and enough personal source material to ground the result. Also note that SKILL.md references parsers like qq_parser.py, social_parser.py, and photo_analyzer.py, but the visible repo snapshot mainly includes wechat_parser.py, so your safest path is to rely on WeChat exports and text-based memory input first.

How to Use create-parents skill

Install create-parents in Claude Code

This repo’s README.md shows clone-based install, not npx. From your git repo root:

mkdir -p .claude/skills
git clone https://github.com/xiaoheizi8/parents-skills .claude/skills/create-parents

Or install globally:

git clone https://github.com/xiaoheizi8/parents-skills ~/.claude/skills/create-parents

Then invoke it in Claude Code with /create-parents, /create-mom, /create-dad, or a natural-language request. For create-parents install decisions, the key check is whether you are actually working in Claude Code and can allow filesystem plus shell actions.

Give the skill the right inputs

The minimum input is small: a name, a short summary, and one line about communication style. Better results come from adding raw material such as:

• exported WeChat chats
• voice-note transcripts
• photos with context
• your own written memories

Strong input example:

• Name: Dad
• Basic info: Retired teacher, blunt but soft-hearted, avoids direct praise
• Communication style: Always asks if I ate, worries about money, says "Don't stay up late"

Why this works: the create-parents usage flow can map concrete habits into catchphrases, care patterns, and scene memories. Vague input like “kind and loving” gives flatter output.

Follow the best reading path before first run

Read these files first:

1. README.md for install location and commands
2. SKILL.md for triggers, output paths, and management commands
3. prompts/intake.md for the exact onboarding questions
4. prompts/memory_analyzer.md and prompts/persona_analyzer.md for what evidence the skill tries to extract
5. tools/wechat_parser.py if your main source is WeChat exports

This reading order matters because the create-parents guide is prompt-driven: once you understand the intake and analyzers, you can prepare cleaner source material before invoking it.

Use a practical create-parents workflow

A good workflow is:

1. Start with /create-parents
2. Answer the 3 intake questions with concrete details
3. Import one source type first, ideally chats or written memories
4. Review the generated parent memory and persona separately
5. Test with a short conversation
6. Correct mismatches and re-run in update mode

If your goal is create-parents for Prompt Writing, treat the first version as a draft. Ask: “What would they never say?”, “What phrase is missing?”, and “What do they always ask about first?” Those corrections usually improve realism more than adding long biographies.

create-parents skill FAQ

Is create-parents beginner-friendly?

Yes, if you already use Claude Code. The intake is simple, and the repo includes prompt files that guide the process. The harder part is preparing personal material and understanding where generated files will be written. If you do not use Claude Code or do not want tool-based file creation, this is probably not the right create-parents skill for you.

When is this better than ordinary prompting?

Use create-parents when you want consistency across sessions, updateable memory, and a reusable persona file rather than a single emotional roleplay. If you only need a one-time “write a message in my mom’s tone,” a normal prompt is faster. The skill becomes more valuable as your source material and need for iteration grow.

What are the boundaries and privacy concerns?

The repo is framed for personal memory and emotional communication, not surveillance or privacy-invasive use. You should only import material you have the right to use. Also, the output is an approximation based on available evidence; it should not be treated as a factual or therapeutic substitute for the real person.

Does the repo fully support every source type mentioned?

Not visibly. SKILL.md references multiple parsers, but the repo preview clearly shows tools/wechat_parser.py plus version and skill management scripts. So the safest create-parents usage path is text memories plus WeChat data, then manual refinement.

How to Improve create-parents skill

Give create-parents more behavioral evidence, not more adjectives

The biggest quality jump comes from patterns, not praise words. Instead of “my mom is warm,” provide:

• first question she asks on calls
• topics she repeats
• how she shows worry
• exact sayings
• what she does when you visit or leave

This helps the skill build rules from observable behavior, which improves response realism and reduces generic sentiment.

Fix the most common failure modes

Typical weak outputs come from:

• overly abstract input
• mixing your interpretation with direct evidence
• no examples of speech habits
• too much backstory but no recurring scenes

If the result feels “nice but not them,” ask the model to revise three things only: forbidden phrases, must-include catchphrases, and default concern topics. That is usually more effective than a total rewrite.

Iterate after the first draft

The repo supports an evolution path, so use it. After testing the generated parent skill, add corrections like:

• “They would never use long emotional paragraphs.”
• “They always send short messages, not polished prose.”
• “They care about sleep first, career second.”
• “Replace abstract comfort with practical reminders.”

This kind of update sharpens the create-parents skill without throwing away the memory structure already created.

Improve create-parents for Prompt Writing

If you are using create-parents for Prompt Writing, define both voice and boundaries. Good additions include:

• dialect or wording preferences
• response length
• whether affection is direct or implied
• what topics the parent avoids
• how they react to bad news versus good news

The best outputs usually come from balancing two layers: factual memories in prompts/memory_* style and behavioral constraints in prompts/persona_* style. That combination is the core strength of create-parents.