Cold outreach — short
You are writing to {{persona}}. Keep it under 90 words. Open with an observation about {{company}}.
Pass 3/3
claude
gpt-4.1
Prompts as source code: versioned, testable, owned by the team that wrote them.
A prompt is a unit of work. It has a body, typed variables, parameters, a history of versions, and a small set of tests that travel with it. Run it against any provider you've connected. The output is captured so you can compare what was good last week against what's good today.
Last edited 7m ago by Devon T. — version 14 created on save.
model = "claude-3.5-sonnet" temperature = 0.2 {{tone}}
You are writing to {{persona}}. Keep it under 90 words. Open with an observation about {{company}}.
Summarize the conversation below in 3 bullets.
Last bullet must name the customer's blocker.
Conversation:
{{transcript}}
| Case | claude-3.5-sonnet | gpt-4.1 | llama-3.1-70b | Avg cost |
|---|---|---|---|---|
| Greets by first name | Pass | Pass | Pass | $0.0021 |
| Mentions the company's recent product | Pass | Fail | Pass | $0.0024 |
| Stays under 90 words | Pass | Pass | Fail | $0.0019 |
| Does not use the word "synergy" | Pass | Pass | Pass | $0.0018 |
Save your first prompt to start the version history. Or import a pack from the marketplace.
A prompt is a unit of work. Save it. Diff it. Run it against the models you actually use. The library remembers what worked last week so you can argue with what's working today. Forge — voice sample