Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
71 lines (52 loc) · 3.54 KB

CONTRIBUTING.md

File metadata and controls

71 lines (52 loc) · 3.54 KB
Raw

Contributing to the prompt library

This document covers what each prompt entry should contain and the workflow for adding new ones.

Prompt entry schema

Every prompt added to the library should include the following fields.

Field Required? Description
Title Yes Short, descriptive name (e.g., "NDA first-pass triage").
Category Yes One of: Legal, Finance, Operations, Product, Research, Comms, Cross-functional.
Use case Yes When to use this prompt — one or two sentences.
Intended audience Yes Who this is for (e.g., legal team, ops, anyone).
Recommended model Yes e.g., Claude Opus for analysis; Sonnet for drafting; Haiku for high-volume classification.
Inputs required Yes What the user must supply (document, list, context) before running.
Prompt text Yes The actual prompt, verbatim. Use placeholders like {{document}} for variable inputs.
Example output Yes Abbreviated example so contributors can see what "good" looks like.
Author / owner Yes Who created or maintains this prompt.
Last updated Yes Date of most recent edit.
Status Yes Draft | Approved | Deprecated.
Sensitivity Yes L1 | L2 | L3 | L4 — see sensitivity-framework.md.
Tags Yes Free-form tags for search (e.g., contracts, NDA, summarization, EU).
Notes Optional Caveats, gotchas, links to related prompts.

Optional fields (add as needed)

  • Connectors / tools used — e.g., Gmail, Google Drive, Notion, Slack.
  • Skill or plugin it pairs with — if you have one.
  • Version history — useful for high-value prompts you iterate on.

How to add a prompt

  1. Pick a category section. Use Cross-functional if it doesn't fit cleanly.
  2. Copy the entry template from prompt-library-template.md and fill it in.
  3. Set Status to Draft.
  4. Ask one colleague (ideally outside your function) to test the prompt and confirm it produces useful output.
  5. Once validated, change Status to Approved and announce it in your team channel.
  6. Revisit your prompts quarterly. Update Last updated; move stale prompts to the Archive.

Promotion criteria (Draft → Approved)

A prompt is ready to be promoted when:

  • It has run successfully on at least two different inputs.
  • The example output reflects the kind of result a teammate should actually expect.
  • The sensitivity tier is set and matches the data the prompt is designed to consume.
  • A second person (not the author) has tested it and confirmed it works.

Deprecation

Prompts get retired when:

  • A better version exists in the library.
  • The underlying workflow has changed (e.g., new tool, new template).
  • The model or tool it depends on is no longer available.

When deprecating, move the entry to the Archive with a one-line note explaining why and (if relevant) which prompt replaces it.

Style

  • Prompts should be verbatim — what you'd actually paste into the model.
  • Use placeholders ({{thing}}) for variable inputs rather than hard-coding examples.
  • Keep example outputs short — three to five lines is usually enough to communicate the shape.
  • Cite the model version you tested with if behavior is model-sensitive.

Code of conduct

  • Never include real confidential data in example inputs or outputs. Use synthetic or fully redacted placeholders.
  • Credit prompts you adapt from other teams in your repo's Notes field.
  • If a prompt depends on a paid tool or specific account, say so in Notes.