persona-hub-create

Persona-Hub — Create & Enrich

Create a new persona or enrich an existing one from source materials.

Two Modes

Mode	Trigger	What happens
Create	/persona-hub-create on a new name	Scaffolds a new persona directory and populates from sources
Enrich	/persona-hub-create on an existing persona	Adds new findings to existing dimension files from new sources

---

Step 1 — Understand What the User Wants

Ask for what you don't already know:

• Name — who is the persona? (e.g., "Elon Musk", "Gandalf", "my CEO")
• Type — real (public figure), fictional (character), or composite (archetype/blend)
• Sources — this is the most important part. Ask the user to provide one or more of:

Source Types (best to weakest)

Source	Quality	Example
Folder of files	Best	./transcripts/, ./interviews/ — bulk ingestion of many files at once
Individual files	Great	speech.txt, interview.md, podcast-transcript.txt
URLs	Good	Links to articles, interviews, public statements
Description	OK for start	"He's a fast-talking NYC real estate guy who..."
"Start minimal"	MVP only	Creates skeleton files the user fills in later

Always encourage more sources. A persona built from one description will be shallow. A persona built from 50+ transcripts will feel real. Tell the user:

The more source material you provide, the more authentic the persona will be. Transcripts, interviews, speeches, and writings work best — they capture how someone actually talks, not how others describe them. You can always add more sources later with /persona-hub-create <name>.

Batch Folder Ingestion

When the user points to a folder:

1. List all files in the folder (.txt, .md, .json, .csv, .yaml)
2. Tell the user how many files you found and their total size
3. Process each file, extracting persona-relevant content
4. Distribute findings across dimension files
5. Register each file as a source in sources/sources.yaml

This is the recommended workflow for building rich personas. Example:

User: /persona-hub-create Elon Musk
Agent: What sources do you have?
User: I have a folder of interview transcripts at ./musk-interviews/
Agent: Found 47 files in ./musk-interviews/. Processing...

---

Step 2 — Scaffold (New Persona Only)

Skip this step if enriching an existing persona.

1. Derive slug from name (lowercase, hyphens, strip special chars): "Elon Musk" → elon-musk
2. Target directory: ./personas/<slug>/ if personas/ exists in cwd, otherwise ~/.persona-hub/personas/<slug>/
3. Create directory with: persona.yaml, identity.md, voice.md, sources/sources.yaml, changelog.md
4. Only create beliefs.md, knowledge.md, relationships.md if sources support them

persona.yaml Template

version: "1.0"
id: <slug>
name: "<Full Name>"
type: <real|fictional|composite>
status: draft
summary: >
  <2-4 sentence description>
dimensions:
  - file: identity.md
    priority: required
    description: "Core identity and background"
  - file: voice.md
    priority: required
    description: "Speaking style and verbal patterns"
agent_notes: >
  Load required files always. Voice patterns override generic LLM tendencies.
created: <today>
last_updated: <today>
primary_author: ""
tags: []

Add more dimension entries as you create them (beliefs.md, knowledge.md, etc.).

---

Step 3 — Analyze Sources & Populate Dimensions

Every .md dimension file starts with YAML frontmatter:

---
dimension: <name>
version: 1
last_updated: <today>
confidence: medium
sources: [<source-ids>]
---

What to Extract Per Dimension

Read each source and distribute findings to the right dimension file:

voice.md (required) — How they talk

• Sentence structure (short/long, fragments, run-ons)
• Vocabulary level and favorite words
• Verbal tics and filler words
• Rhetorical moves (repetition, hyperbole, understatement)
• Tone (formal, casual, aggressive, warm)
• Anti-patterns — things they would NEVER say. These are hard constraints.

identity.md (required) — Who they are

• Background, origin, formative experiences
• Roles they identify with
• Self-concept (how they see themselves)
• Public vs private persona differences

beliefs.md (recommended) — What they think

• Core values and principles
• Positions by topic, with temporal evolution (views change over time)
• Contradictions and unresolved tensions — document these, don't resolve them

knowledge.md (recommended) — What they know

• Areas of deep expertise (speak authoritatively)
• Working knowledge (can discuss, not expert)
• Knowledge gaps (be vague or dismissive here)

relationships.md (supplementary) — Who matters to them

• Key people and dynamics
• Relationship patterns (mentor, rival, ally)

Citation Format

Every claim should have an inline citation:

He frequently uses superlatives [source:interview-2024-03, confidence:high]

Confidence levels:

• high — direct quote or repeated pattern across multiple sources
• medium — clearly implied from source material
• low — single instance or reasonable inference
• speculative — educated guess, flagged for verification

Register Sources

Add every source to sources/sources.yaml:

sources:
  - id: interview-2024-03
    type: transcript
    title: "Joe Rogan Interview March 2024"
    date: "2024-03-15"
    path: "./transcripts/rogan-2024-03.txt"
    word_count: 12500

---

Step 4 — MVP Check

Check if the minimum viable persona exists:

• persona.yaml has name, type, summary ✓
• identity.md has content beyond frontmatter ✓
• voice.md has content, including anti-patterns ✓

If yes: set status: active in persona.yaml, offer to activate with /persona-hub <name> If no: keep status: draft, tell user what's missing

---

Step 5 — Report

Show the user what was created/updated:

Persona: {name}
Status: {active|draft}
Sources processed: {count}
Dimensions: {list of files created/updated}
Total content: ~{word count} words

Next steps:
- Activate: /persona-hub <name>
- Add more sources: /persona-hub-create <name>
- Edit manually: {path to persona directory}

---

Enrichment Mode (Existing Persona)

When /persona-hub-create targets an existing slug:

1. Read existing dimension files to understand current state
2. Process new sources
3. ADD new findings to existing files — don't overwrite, merge
4. Bump version in each updated file's frontmatter
5. Update last_updated dates
6. Register new sources in sources/sources.yaml
7. Append to changelog.md:

   ## v{version} — {date}
   - Added {count} new sources
   - Updated: {list of dimension files changed}
   - Key additions: {brief summary}
   

Important: When enriching, look for:

• New voice patterns not captured before
• Evolved or changed positions (add temporal entries, don't replace)
• Contradictions with existing content (document as unresolved tensions)
• Deeper detail on topics already covered