Brainforge Knowledge

prompt-guide

5 min read

Guide: Adding and Linking New Prompts Using Langfuse (ELI5, Final Naming Standard)

This guide explains how to:

  • create new prompts in Langfuse
  • name them using the final, human-readable standard
  • link them into the playbook
  • create prompt usage documentation for new use cases

This version replaces all previous naming guidance.

Big Picture (Read This First)

There are three systems, each with one job:

  1. Langfuse

    • Stores prompt logic
    • Handles versioning automatically
    • Is the single source of truth

  2. playbook

    • Documents how humans should use prompts
    • Explains when and why to use each prompt
    • Lives at standards/ within brainforge-platform

  3. Cursor

    • Is where people do the actual work
    • Does NOT store or version prompts
    • Follows the documented prompt behavior

Golden rule: Prompt logic lives ONLY in Langfuse.
The playbook explains how to use it.

Final Prompt Naming Philosophy (IMPORTANT)

Prompt names are for humans, not machines.

Because:

  • Langfuse already handles versioning
  • Prompts are browsed visually
  • Non-technical users must understand names instantly

Therefore:

  • Prompt names should be verbose
  • Prompt names should be readable English
  • Prompt names should not include version numbers
  • Prompt names should not use underscores
  • Prompt names should not rely on technical syntax

Prompt names should read like titles.

Final Prompt Naming Format (Authoritative)

Use this format:

[Domain] Short Descriptive Name

Examples:

  • [SOW] Sales Draft Generator
  • [SOW] Client-Facing Rewrite
  • [PRD] Requirements Generator
  • [PRD] Requirements Refinement
  • [SOP] Step-by-Step Procedure Generator
  • [Review] Clarity Improver
  • [Review] Senior Consultant Rewrite
  • [Meeting] Transcript Summarizer
  • [Meeting] Action Item Extractor
  • [Tickets] Internal Linear Ticket Generator
  • [Tickets] Ticket Grooming Assistant
  • [Zoom Slack Summarizer] Sales Lead Call Summarizer

This matches your real example exactly.

Domain Rules

The bracketed prefix is mandatory.

It tells users:

  • what category the prompt belongs to
  • when to use it
  • where it appears in Langfuse folders

Recommended domains:

  • [SOW]
  • [PRD]
  • [SOP]
  • [Review]
  • [Meeting]
  • [Tickets]
  • [Sales]
  • [Ops]
  • [Zoom Slack Summarizer]

Domains should be:

  • short
  • familiar
  • meaningful to non-technical users

Versioning Rule (Very Important)

DO NOT include version numbers in the prompt name.

Examples of what NOT to do:

  • v1
  • v2
  • version 3
  • final

Langfuse already:

  • versions prompts automatically
  • tracks changes
  • allows rollback

Prompt names must remain stable forever.

When to Create a New Prompt

Create a new prompt when:

  • the responsibility is different
  • the output format is different
  • the audience is different
  • the use case is different

Examples:

  • “Sales Draft Generator” vs “Client-Facing Rewrite”
  • “Transcript Summarizer” vs “Action Item Extractor”

Do NOT create new prompts for minor wording changes. Those are Langfuse version updates.

Part 1: Creating a New Prompt in Langfuse

Step 1 — Define the Responsibility (One Sentence)

Before opening Langfuse, write:

“This prompt is responsible for __________.”

Examples:

  • “Summarizing sales lead calls from Zoom transcripts.”
  • “Generating a first-draft SOW using a template and meeting context.”
  • “Rewriting content in a senior-consultant, client-facing tone.”

If this sentence is unclear, stop.

Step 2 — Choose the Prompt Name

Convert the sentence into a readable title.

Examples:

  • [Meeting] Sales Call Summarizer
  • [SOW] Sales Draft Generator
  • [Review] Senior Consultant Rewrite

Rules:

  • Capitalize words normally
  • No underscores
  • No version numbers
  • No emojis
  • No abbreviations unless universally understood

Step 3 — Create the Prompt in Langfuse

  1. Open Langfuse.
  2. Go to Prompts.
  3. Click “Create Prompt.”
  4. Enter the prompt name exactly as chosen.
  5. Add a short description explaining:
    • what it does
    • when to use it
    • when NOT to use it

This description is for humans.

Step 4 — Write the Prompt Logic

Inside Langfuse:

  • define role and behavior
  • define tone and constraints
  • define expected output
  • define how inputs are used

Keep prompts:

  • focused on one job
  • explicit
  • constrained
  • predictable

Avoid:

  • vague instructions
  • combining unrelated tasks
  • hidden assumptions

Step 5 — Define Inputs Clearly

Explicitly define inputs such as:

  • Template
  • Transcript
  • Meeting Notes
  • Context
  • Constraints

Clear inputs make prompts easier to test and safer to reuse.

Step 6 — Test Thoroughly

Before publishing:

  • test with real data
  • verify accuracy
  • check for hallucinations
  • confirm tone
  • revise until outputs are reliable

Never publish untested prompts.

Step 7 — Publish the Prompt

Once ready:

  • publish the prompt
  • rely on Langfuse’s internal versioning
  • treat the name as permanent

Part 2: Creating Prompt Usage Documentation (REQUIRED)

Every prompt MUST have usage documentation.

If it is not documented, it does not exist.

Step 8 — Choose the Correct Usage File

In the playbook:

  • go to standards/04-prompts/
  • open the domain file:
    • sow-prompts.md
    • prd-prompts.md
    • review-prompts.md
    • meeting-summarization-prompts.md

If none exists:

  • create one
  • name it clearly

Step 9 — Document the Prompt

For each prompt, include:

  1. Prompt name
  2. What it does
  3. When to use it
  4. When NOT to use it
  5. Required inputs
  6. Example Cursor instruction

Do NOT paste the prompt text.

Step 10 — Write Cursor Instructions (Copy-Paste Ready)

Example:

“Use the Langfuse prompt [SOW] Sales Draft Generator. Apply it to the open file using the SOW template and relevant meeting notes. Generate a complete first draft in the current document.”

This is what non-technical users will copy.

Step 11 — Commit and Push

  1. Save the usage doc.
  2. Commit with a clear message: “add usage docs for [SOW] Sales Draft Generator”
  3. Push to GitHub.

Common Mistakes to Avoid

  • Adding version numbers to names
  • Using technical naming conventions
  • Editing prompt logic in markdown
  • Creating prompts without documentation
  • Overloading prompts with multiple responsibilities

Final Rule to Remember

Langfuse defines behavior.
The playbook explains behavior.
Cursor executes behavior.

Readable names + centralized logic = scalable adoption.

Graph View