create-persona

Create Persona

Create a new user account on Webmatrices for content seeding purposes.

Note: There is no MCP tool for user creation yet. This skill uses a temporary database script. All other operations (posting, replying, banning) should use MCP tools.

Database Connection

All scripts MUST run from ~/Projects/webm-frontend for Prisma access. Read production credentials from .env:

cd ~/Projects/webm-frontend
DATABASE_URL=$(grep '^PROD_DATABASE_URL=' .env | cut -d'"' -f2) node temp-script.mjs

Password Hashing

Use the Django PBKDF2 format used by the codebase:

import crypto from 'crypto';

function hashDjangoPassword(password: string): string {
  const iterations = 390000;
  const salt = crypto.randomBytes(12).toString('base64');
  const hash = crypto.pbkdf2Sync(password, salt, iterations, 32, 'sha256');
  return `pbkdf2_sha256$${iterations}$${salt}$${hash.toString('base64')}`;
}

Persona Guidelines

When creating a persona, match these conventions:

• Email: Use bishwasbh+<username>@gmail.com format
• Password: Use a strong default like Webm@tr1ces2026! (tell the user)
• Bio: Short, lowercase, casual — matches the community tone
• Auth status: Set isConfirmed: true so they can post immediately

Existing Persona Styles (reference)

Fetch existing personas from get_self_personas MCP to check for voice/topic overlap before creating a new one. Each persona includes metadata.personaTraits with voice, topics, opinion strength, and backstory. Use this data to ensure the new persona fills a gap rather than duplicating an existing voice or topic area.

Persona Traits Schema (v2.1)

The metadata.personaTraits JSON blob on the User table holds the persona's voice, perspective, and writing patterns. There is no strict schema (JSON column), but the following keys are conventional and consumed by /write-post, /write-reply, /smell, and /reduce-smell.

Voice layer (drives sentence shape)

• writingSamples — the load-bearing voice anchor. Conventionally structured by content shape:
  • writingSamples.howTheyOpenPosts — array of opener samples (used by /write-post)
  • writingSamples.howTheyClosePosts — array of closer samples (used by /write-post)
  • writingSamples.signatureRhythm — array of mid-paragraph rhythm samples
  • writingSamples.apostropheInContext — sample(s) showing how the persona uses apostrophes
  • writingSamples.replies — (new in v2.1) flat array of 3 short-form sample replies. Used by /write-reply to anchor length and sentence shape. If missing, reply-mode output is best-effort.
• apostrophes — informational only. Pattern note (e.g., "mostly keeps, drops in technical product posts"). Not scored against by /smell. Drift is expected.
• casing — informational only. Same — drift is expected, samples are the source of truth.

Perspective layer (drives stance, not voice)

• backstory — location, age, timeline, struggles, wins, professional context. Surfaces as which details the persona references. Does NOT drive sentence-level voice (that's the samples' job).
• topics — what the persona writes about. Used for auto-matching and fatigue checks.
• opinionStrength — HARD / SOFT / EDITORIAL. Influences how strongly the persona stakes a claim.
• signaturePhrases — phrases the persona returns to. Surfaces occasionally in posts, not every time.
• voice — a one-line voice description (e.g., "Editorial, strong opinions held loosely. Dual voice: editorial vs product launch."). Informational; the samples are the authoritative voice source.

Why the split matters

In v2.1, the /write-* skills enforce a strict separation:

• Voice from samples → sentence shape, casing, slang, punctuation rhythm, length
• Perspective from backstory → which specific cafe, which specific job, which specific timeline event

If a persona has rich backstory but no writingSamples.replies, /write-reply can still generate output but it won't have a length or rhythm anchor. Adding 3 short example replies fixes this and is the recommended setup for any persona that will reply on threads.

Example: adding reply samples to a persona

# Via update_persona MCP
update_persona userId=2 personaTraits='{
  "writingSamples": {
    "replies": [
      "yeah same thing happened to me, ended up just deleting the post",
      "the math doesnt add up. their RPM claims are 3x what i see on similar sites",
      "you tried the new export tool yet? it actually works now"
    ]
  }
}'

The MCP update_persona tool deep-merges writingSamples, so this adds the replies array without touching the existing howTheyOpenPosts / howTheyClosePosts / etc.

Workflow

1. If username provided as argument, use it. Otherwise, suggest 3-4 username options
2. Ask the user to pick or provide a username
3. Create the user with a temporary script:
   • isActive: true, isStaff: false, isSuperuser: false
   • A Profile with a short bio
   • An AuthStatus with isConfirmed: true
4. Display the created user details (ID, username, email, password)
5. Clean up temporary scripts
6. After creation, use MCP tools for all further operations (posting, etc.)