Prompt-Template Library for AI/LLM Workflows in *Zed*

[zhengma021]

Background & Problem Statement

When working with AI assistants (e.g., Claude) inside Zed, I frequently encounter the following friction points:

• Without explicitly instructing “do not use OOP”, the assistant often defaults to an object-oriented design (classes, inheritance, parent/child relationships) even when a flat functional or composition-based approach would be far more appropriate.
• If I do not pre-specify “first provide a plan, then wait for my review, and only afterwards write the code”, the assistant often jumps straight into modifying code. This leads to sub-optimal results, extra rework, and inefficient back-and-forth.
• I end up writing the same kind of prompt instructions repeatedly (e.g., “avoid OOP”, “give me a plan before coding”, “use FP style”), which is tiring and error-prone.
• It would be far more efficient if Zed offered a Prompt Template Library: pre-configured templates like FP first, Planning first, No-OOP style, which I could select rather than rewrite each time.

Proposed Feature

I propose that Zed introduce (or support via extension) a Prompt Template Library and selection workflow, with the following capabilities:

1. Prompt Template Library

• A built-in (or user-configurable) library of common prompt templates. Example templates might include:
  • “Function-First / FP‐first – emphasise functional style, avoid class/inheritance.”
  • “Planning‐First – first output a high-level plan → wait for review → then code.”
  • “Explain-Then-Code – provide explanation/context before generating code.”
  • “No‐OOP – avoid object-oriented patterns; favour flat/functional/record-based structures.”
• Users can create, edit, delete, duplicate templates.
• Templates have metadata: tags (e.g., “FP”, “No-OOP”, “ReviewFirst”), description, optional variables (e.g., language, target audience, style guidelines).

2. Scope / Dimension Classification

Templates may be applied on different scopes/dimensions:

• Project dimension – For a given project (or workspace) you may designate a default template set or default style (e.g., for this project we always use “FP-first”).
• Model dimension – For different LLM backends (e.g., Claude, GPT-4, Zeta) you may designate different default templates.
• Request dimension – For a specific prompt/request inside the session, you may choose a template (or combine templates) just before sending.