Loomkit Build Persona

Loomkit Persona Builder

Combine a character (from agent-personas/characters/) and a role (from agent-personas/roles/) into a coherent, standalone system persona under a freely chosen name, then either save it or apply it to the current agent.

Goal

Produce a finished persona that can be used directly as a system prompt — no reference to source files, no generic assistant phrases, but a distinctive identity. The persona's description is written in English; its communication language is decoupled from that and adapts to the user at runtime.

Prerequisites

• Directory agent-personas/characters/ with at least one character file
• Directory agent-personas/roles/ with at least one role file
• For option (a): write access to the chosen target folder (default agent-personas/my-personas/)
• For option (b): the profiles in agent-personas/systems/ for the current target system

Steps

Step 1 — Resolve arguments

Read $ARGUMENTS. Expected format: <character> <role> (both optional).

If arguments are missing:

1. Read all files in agent-personas/characters/ and agent-personas/roles/.
2. List the available name values from the frontmatter clearly.
3. Ask the user for:
   • Character name
   • Role name
   • Target system (e.g. "Claude", "Hermes Agent", "OpenAI")
   • Desired name of the new persona (e.g. "Mira", "Alex") — this name is the identity of the persona
4. Wait for the answer before continuing.

If arguments are present:

1. Check whether agent-personas/characters/<character>.md and agent-personas/roles/<role>.md exist.
2. If a file is missing: report the error and show the available names.
3. Still ask for the desired persona name and the target system. Wait for the answer.

Step 2 — Read sources

Read:

• agent-personas/characters/<character>.md
• agent-personas/roles/<role>.md

Step 3 — Synthesize the persona

Combine character, role, and the chosen name according to these rules:

• Name: The name chosen by the user is the identity. It replaces the character name from the source file everywhere.
• Identity sentence: Begins with "You are ," — the role determines WHAT the persona does, the character determines HOW it does it.
• Core principles: 4–6 points, a mix of role expertise (subject matter) and character principles (attitude). No generic assistant phrases.
• Communication style: Primarily shaped by the character, made concrete by the role — specific and vivid, as if this style belonged to <Name> personally.
• Language: The persona always responds in the language the user addresses it in. The fact that this description is written in English must not constrain the persona's communication language — keep the two independent. State this explicitly in the persona (see the ## Language section in the format below).
• Boundaries & priorities: Merge from both sources. Resolve contradictions explicitly.

The result is a standalone persona. No reference to character or role files in the output text.

Step 4 — Choose: save or apply

Ask the user what to do with the finished persona:

(a) Save to a folder

• Default folder: agent-personas/my-personas/. Offer it, but let the user choose a different folder.
• Make sure the folder exists (create it if not).
• Save as <target-folder>/<name-kebabcase>-<role>.md using the format in Step 5.

(b) Apply to the current agent

• Consult agent-personas/systems/ and load the profile for the current target system/framework/harness, so the persona is rendered in the format that system expects.
• If the current agent already has a persona/system profile, clarify with the user how to handle it before applying — replace, merge, or layer the new persona on top — and act on that choice.
• Apply the persona in the format and location the target system uses (e.g. the relevant CLAUDE.md/AGENTS.md/SOUL.md), as described in the matching agent-personas/systems/ profile.

Step 5 — Persona format

---
name: <name-kebabcase>-<role>
version: 1.0.0
category: system
description: <One-sentence description of the combined persona>
target_system: <target_system>
character_source: <character>
role_source: <role>
tags: [<merge relevant tags from both sources>]
---

# <Name>

<Identity sentence — second person singular, precise and characteristic>

## Core Principles
- <Principle 1>
- <Principle 2>
- <Principle 3>
- <Principle 4>

## Communication Style
<How does this persona communicate? Tone, pace, word choice, structuring habits — concrete and distinctive>

## Language
Respond in the language the user addresses you in. Do not default to the language of this profile.

## Boundaries & Priorities
<What does this persona explicitly not do? What does it always prioritize?>

Output

1. Full content of the persona (as a Markdown block)
2. For (a): the file path where it was saved. For (b): a note that the persona is now active for the current agent, including how an existing profile was handled (replace/merge/layer) and which agent-personas/systems/ profile was used.
3. A short note on what the character and role each contributed.