_model: devlog-entry
---
schema_version: 1
---
title: Making model prompts configurable
---
date: 2026-07-17
---
author: Christopher Chambers
---
summary: Conversation and drafting prompts moved into documented, user-editable JSON files that reload without rebuilding the application.
---
tags: prompts, models, configuration, user-control
---
source_commit: 58e5510445d7ab403d8df0792c29743a933531e0
---
body:

Thinkloom's model behavior should not be a hidden, immutable part of the
application. Conversation and drafting prompts were moved into structured JSON
templates and documented as part of the project configuration.

The application creates user-owned copies and reloads them before each request,
so a technical writer can adjust instructions without rebuilding or restarting
Thinkloom. Validation happens before a request is sent: invalid JSON or missing
template variables produce a recoverable error and leave the writing unchanged.

This work reinforced two of the project's recurring principles—model behavior
should be inspectable, and failure should not silently alter a manuscript.

[View the configurable-prompts commit](https://git.labyricorn.com/Labyricorn/thinkloom-openai-hackathon/commit/58e5510445d7ab403d8df0792c29743a933531e0)