/create-soul

Overview of the Process

① Confirm Persona → ② Collect Materials → ③ Distillation → ④ Assemble Skill → ⑤ Verification → ⑥ Installation Guide

Step 1: Confirm Persona

1. Persona Name (Chinese/English)
2. One-sentence Description (Who TA is, what TA does)
3. Material Status — Do they have existing materials (file/directory path) or need on-site collection?

• {person_name}

  — Persona name

• {slug}

  — English slug (used for directory name and skill name, e.g.,

  wei-ran

  )

• {output_dir}

  — Output path, default

  ./{slug}-soul/

Step 2: Collect Materials

If the user has an existing material directory

If on-site collection is needed

• Podcast/interview URL → Pull using

  collectors/youtube_transcript.py

• Blog/article URL → Pull using

  collectors/fetch_url.py

• Jike homepage URL → Pull using

  collectors/jike_export.py

• Path to Twitter self-export package → Parse using

  collectors/twitter_archive.py

• Text directly pasted by the user (chat records, notes, etc.)

• Output of each collection script is stored in the corresponding subdirectory of

  {output_dir}/_raw/

• After collection, output the inventory: file name, type, word count
• Minimum Threshold: At least 5 material files or a total of 10,000 words. Remind the user to supplement if insufficient, but do not force blocking.

Step 3: Distillation (3-Pass)

Pass 1: Read + Annotate Article by Article

_raw/

1. Full Reading (no skimming, no only reading the first few lines)
2. Extract the following signals:
   • Views and positions (including specific expressions)
   • Thinking patterns (how to reason, give examples, make judgments)
   • Language features (catchphrases, sentence rhythm, word preference)
   • Emotions and attitudes (what excites/angers/hesitates TA)
   • Worthy quotes (quote-worthy)
3. Annotate by type: Personality Signal / Knowledge Signal / Mixed

Pass 2: Aggregation + Deduplication

• Merge semantically duplicate views (retain the best-expressed version)
• Identify core topics (3-8)
• Annotate position evolution (statements on the same topic in different periods)
• Filter top quotes (≥20 items)

Pass 3: Structured Writing

_persona/rules.md

• Identity information (what TA does now, what TA did in the past, public presence)
• Core personality traits (5-8 items, each with evidence)
• Thinking framework (TA's unique analysis method, not a general framework)
• Decision-making style
• Catchphrases (original quotes)
• Hard boundaries (things TA will absolutely not say/do, ≥5 items)

_persona/communication.md

• Language patterns (at least 2 scenarios distinguished, each with ≥8 real sentence samples)
• Style differences between long and short texts
• Spoken language features (if there are podcast materials)
• Punctuation and formatting habits
• Language mixing rules (Chinese-English switching habits)

_persona/values.md

• Hierarchically arranged beliefs (deep beliefs / strong tendencies / in exploration)
• Each item with original quote citation
• Belief evolution trajectory (if materials span time periods)

_knowledge/{topic}.md

(one file per core topic)

• Core views (with original quotes)
• View evolution
• Reasoning chain (why TA thinks this way)

_quotes/iconic.md

• ≥20 representative quotes, grouped by topic
• Annotate sources

_quotes/internal.md

(if there are informal materials)

• Original quotes from private/casual occasions
• Show what TA is like when not being formal

_meta/sources.md

• Material inventory + coverage rate

Step 4: Assemble SKILL.md

SKILL.md

{output_dir}/

---
name: {slug}-chat
description: "Chat with AI {person_name}. Distilled from {N} sources."
---

# AI {person_name}

You are **{person_name}**, {one-sentence description}.

## Activation

1. Load persona files:

\```
./_persona/rules.md
./_persona/communication.md
./_persona/values.md
./_quotes/iconic.md
./_quotes/internal.md
\```

2. Load knowledge docs on demand — only when the conversation topic matches:

\```
./_knowledge/
\```

## Core Rules

### Identity
{Extract 3-5 core identity descriptions from rules.md}

### Thinking Style
{Extract key points of thinking style from rules.md}

### Language
{Extract key language rules from communication.md}

### Hard Boundaries
{Extract hard boundary list from rules.md}

### Catchphrases
{Extract catchphrases from rules.md}

## Start

Use `$ARGUMENTS` as the user's first message and respond in character.

Step 5: Verification

Completeness Check

□ _persona/rules.md — Identity + Personality + Thinking Framework + ≥5 hard boundaries
□ _persona/communication.md — ≥2 language modes, each with ≥8 real sentence patterns
□ _persona/values.md — Hierarchical beliefs + citations
□ _knowledge/ — ≥2 topic files
□ _quotes/iconic.md — ≥20 quotes
□ _meta/sources.md — Material coverage rate
□ SKILL.md — Fully usable

Fidelity Test

1. A viewpoint question in TA's area of expertise
2. A casual/lighthearted topic
3. An area TA may not understand (test boundaries)

Step 6: Installation Guide

cp -r {slug}-soul/ ~/.claude/commands/{slug}/
# Then use /{slug}-chat in Claude Code

cp -r {slug}-soul/ ~/.openclaw/skills/{slug}/

{slug}-soul/

System/Skills/

Behavioral Rules

1. Read all articles before writing. Do not start generating after only scanning the first few lines.
2. Prioritize original quotes. communication.md and quotes must contain original text from materials, not AI rewrites.
3. Do not fabricate. Do not guess or supplement information not present in materials. Mark "Not covered by materials" in the file.
4. Hard boundaries must be specific. Do not write nonsense like "Do not use vulgar language"; write executable rules like "Do not say XXX".
5. Inform if materials are insufficient. If there are too few materials (<5 files), generate a basic version and clearly tell the user which dimensions lack materials.