Using the Prompt Playground

The Prompt Playground is where operators and power users design, test, and compare prompts before putting them to work. It ships with a curated template library, lets you save your own prompts, run them with tunable parameters, and compare runs side by side to see which prompt and settings perform best.

What the Playground is for

Getting a prompt right is iterative. The Playground gives you a controlled place to experiment — separate from your live conversations — so you can refine a system prompt, dial in parameters like temperature, and prove out a prompt before you rely on it. It's especially useful when you're building repeatable prompts for a team or a workflow.

The template library

The Playground opens with a library of ready-made templates you can start from:

• Support triage (Customer ops) — classify an incoming request, identify urgency, and propose the next owner without inventing account details.
• Project status brief (StretchProjects) — turn raw project notes into a status brief with shipped / blocked / next / owner sections.
• Prompt critique (AI governance) — review a prompt for clarity, missing context, injection risk, and measurable output requirements.
• Retrieval grounding check (Search bridge) — verify whether an answer can be grounded in supplied evidence and flag unsupported claims and missing sources.

Each template comes with a system prompt, a user prompt (with {{placeholders}} you fill in), and sensible default parameters (model, temperature, max tokens, top-p).

Running a prompt

1. Open the Prompt Playground.
2. Pick a template from the library, or write your own system and user prompts. A user prompt is required to run.
3. Fill in any {{placeholders}} — for example, paste the customer message into the Support triage template's {{message}} slot.
4. Adjust the parameters if needed: temperature (creativity vs. determinism), max tokens (answer length cap), and top-p.
5. Click Run. The Playground records the run and returns the output along with usage information.
6. Review the result. If it's good, save the prompt for reuse.

Saving prompts

1. With a prompt you like open, choose Save prompt.
2. Give it a title and a category so you can find it later.
3. It's stored in your saved prompts list for quick reuse. The Playground keeps your most recent saved prompts on hand.

Comparing runs

The real power of the Playground is comparison:

1. Run the same task two or more different ways — for example, one at temperature 0.1 and one at 0.7, or two different system prompts.
2. Open the Compare view.
3. Select up to four runs.
4. The Playground shows each run's model, temperature, total tokens, and timestamp side by side so you can see which combination gave the best answer for the least cost.

A realistic example

You want a reliable "project status brief" prompt for your delivery team. You start from the Project status brief template, paste a week's notes into {{notes}}, and run it at temperature 0.3. The brief is good but a little verbose, so you run it again at temperature 0.2 with a lower max-tokens cap. In Compare, you line the two runs up: the tighter run uses fewer tokens and reads cleaner. You save that version as "Weekly delivery brief (tight)" in the StretchProjects category, and your team reuses it every Friday.

Tips

• Change one thing at a time. Vary temperature or the system prompt between runs, not both, so comparisons are meaningful.
• Lower temperature for classification and extraction; raise it for brainstorming. The templates already reflect this — triage runs cool, creative tasks run warmer.
• Watch total tokens in Compare. A slightly better answer that costs far more tokens may not be worth it.
• Save the winner with a descriptive title and category. Future-you will thank present-you.

FAQ

Do Playground runs use my connected providers? The Playground is a design-and-test surface; runs are recorded with their parameters and usage so you can compare prompt behavior. Live model execution is governed by the same activation and safety controls as everywhere else in StretchGPT.

How many prompts can I save? The Playground keeps a rolling list of your most recent saved prompts. Give them clear titles so the ones you care about are easy to spot.

How many runs can I compare at once? Up to four.