Eval
eval.particular.tech

Test setup prompts against real coding agents.

Run one prompt across every environment and harness you care about. Each run reports its trace, its cost, and an evaluator's verdict on whether the result actually works.

Create an account Sign in
Console setup prompt
Example batch · 3 environments × 2 harnesses
Claude CodeCodex
Empty folder
pass
pass
Existing project
partial
pass
No npm
fail
partial

How it works

  1. 1

    Write the prompt

    The setup instructions you hand a coding agent, with {{VARIABLES}} filled per environment.

  2. 2

    Pick the matrix

    Environments down, harnesses across. Every combination runs in its own sandbox.

  3. 3

    Read the verdict

    An evaluator agent inspects each sandbox and scores whether the setup actually works.

Everything a run is made of

Save each piece once, then mix and match them across batches.

Environments
Image, CPUs, repo, env vars, packages, blocked commands, and a setup script.
Components
Reusable pieces like “Neon CLI installed” or “no npm”, composed into environments.
Harnesses
Claude Code, Codex, Cursor, or a custom CLI, each with its own model, skills, and MCP servers.
Evaluators
A harness, a model, and instructions that judge the result and explain why.
Traces and cost
Tool calls, turns, sub-agents, tokens, and cost parsed from each harness trace.
MCP and API
Drive everything from your own agent with a Particular.Tech API key, OAuth, and the MCP server.

Stop guessing whether your setup prompt works.

Get started