Prompt Vault Brief

# Product Brief — `prompt-vault`

## 1. One sentence

Prompt Vault is where I store and reuse my best prompts so I never rewrite a good prompt from scratch again.

## 2. The committable core

Find the right prompt and reuse it in seconds. Saving is table stakes underneath this — the product is judged on retrieval, not capture.

## 3. Non-goals

- Prompt pipelines / multi-step chaining / AI-platform orchestration — explicitly deferred to a later version (this is a vault, not an execution engine).
- Running prompts in-app — no API calls to OpenAI/Claude/etc.; the user copies the prompt out into their own tool.
- Team features — no sharing, collaboration, public libraries, or marketplace.
- AI that writes or improves prompts for the user — no generation or auto-optimization.

## 4. User & moment

- **User:** Technical daily-AI user — covers both the power user (uses Claude/ChatGPT daily, has 20+ prompts they keep rewriting, lives in the browser) and the builder/developer (engineering reusable prompts with known outputs). They share one moment, so they are one user for design purposes.
- **Moment:** Mid-task retrieval — "I'm in the middle of work, I need that prompt right now, find it in 3 seconds and paste it." Designs for speed, search, keyboard, minimal friction. Not library-curation as the primary moment. This moment is desktop/web behavior — see role split below.
- **Surface roles (split):** Desktop/web is **primary** — authoring, organization, heavy search, versioning. Mobile is a **companion**, not a shrunken desktop app — fast capture and emergency retrieval only. Do not design for feature parity across surfaces.

## 5. Where it breaks

The scale / findability cliff. Once the vault holds 200+ prompts, "find it in 3 seconds" gets hard — search returns near-duplicates, tags rot, and it's unclear which version actually worked. The core promise breaks exactly when the user has enough prompts to need it. Build effort goes here.

## 6. The one great thing

Find-by-result retrieval: search surfaces candidate prompts, and the saved output example lets the user recognize the right one at a glance. Search narrowing + output-as-recognizer are one integrated mechanism, and it must be flawless. Everything else can be merely good.

## Decisions & overrides

- User kept **version history with diff between versions** in v1 despite it adding surface to a retrieval-first product. Justified by real iteration on saved prompts over time.
- User named the "one great thing" as a combined search + output-as-recognizer mechanism rather than picking one; recorded as a single integrated retrieval experience (search is the gate, output example is the recognizer).
- Product spans phone and webapp per the original request; product-type decision deferred to the spec stage. User confirmed a **split-role** model (desktop/web primary, mobile companion) rather than unified feature parity.

## Tone / aesthetic direction (optional)

Not specified — spec stage to propose.