Promptles is a scenario-based game for GTM engineers, PMs, designers, and enthusiasts shipping with AI agents — anyone who prompts Claude Code, Cursor, Replit Agents, or similar tools without a traditional software engineering background. You're handed real engineering tickets — debugging stack traces, refactoring across files, designing auth flows — and write the prompt you'd actually send. Each prompt is graded by Claude on the criteria a senior engineer would care about, and the expert prompt sits next to yours so you can read the gap.

Does Promptles use Claude?

Yes. Every prompt you write is graded by Claude Haiku 4.5, against a mix of deterministic pattern checks and AI-evaluated criteria. The Anthropic API is called server-side; players never need their own API key.

Do I need an Anthropic API key?

No. Promptles handles all AI calls server-side. Sign in once with your email and start playing — there's no key to paste, no provider to configure.

Is there a free tier?

Yes. The free tier includes the foundational scenarios, today's daily challenge, and pattern-based grading. Paid tiers (Solo, Lifetime, Team) add full AI grading, every scenario, and the daily archive.

How is Promptles different from other prompt engineering tutorials?

Most prompting tutorials teach you to write a single perfect prompt. Promptles grades multi-turn conversations: you write a prompt, see what Claude actually replies, then write the follow-up that fixes what's still wrong. The whole chain is graded, because that's what real software engineering work looks like.

Is Promptles useful for GTM engineers?

Yes — GTM engineers are a core audience. Building sales, marketing, and RevOps automation, internal tools, integrations, and AI agents means writing a lot of prompts to Claude Code, Cursor, Replit Agents, and similar tools. Promptles teaches the prompting rigor that gets reliable output the first time, so the integrations and agents you ship don't break the next morning. The vocabulary deck includes a dedicated GTM & Business Tools section covering the SaaS platforms, data models, and business concepts you'll be automating.

Can my team use Promptles?

Yes. The Team plan adds per-seat AI quotas and lets admins author custom scenarios for the team's codebase using AI assistance. Each member keeps their own progress and playbook; team-authored content shows up alongside the core library.

Errors

Verify the MCP Output

Ask for the raw response shape first; refuse to act on fields the agent can't quote.

MCP responses can be paginated, partial, or shaped differently than you expect — and the agent will happily quote fields that aren't in the response if you don't push back. Ask it to surface the raw shape first (the actual keys and types it received), then act on the data. If a field the agent claims isn't visible in that raw output, it didn't come from the MCP — it was invented.

Where you'll practice this

4 Promptles scenarios teach this principle directly. Each one drops you into a real engineering ticket and asks you to write the prompt you'd send to Claude Code.

Show Me the Shape
Last week, an engineer asked the agent to summarize three Linear issues via the Linear MCP. The summary confidently included a `severity` field for each issue — 'severity: high', 'severity: critical' — and the team made decisions on it. Turns out the Linear API doesn't return a…
Don't Hallucinate Fields
Your team's nightly digest is an MCP-driven summary — the agent calls `github__list_pull_requests` against three repos and writes a Slack message with the day's open PRs. This morning's digest claimed every PR had a `riskScore` field ('riskScore: low', 'riskScore: medium',…
Find the Right Doc, and Cite It
Your Google Drive MCP server is connected. You asked Claude to 'summarize our pricing doc' and got a clean, confident summary, of the wrong file. There are four documents with 'pricing' in the name, and it grabbed a draft from last year. The one you actually mean is 'Pricing…
Verify the Whole Chain Before It Ships
You've built a weekly client update: pull the headline metrics from a Drive sheet, summarize the open issues in Linear, and post a client-facing update to a shared Slack channel. It runs end to end, which is exactly the problem, a few weeks ago it posted a revenue number off a…

Verify the MCP Output

Where you'll practice this

Show Me the Shape

Don't Hallucinate Fields

Find the Right Doc, and Cite It

Verify the Whole Chain Before It Ships

Other principles in Errors

Describe How to Reproduce

Describe the Sequence

Describe What You See

Diagnose Before You Prescribe

Include the Error Message

Quantify the Problem