Evals

Evaluation suites for prompts, skills, and agent workflows—regression gates before changes reach production traffic.

Marketing capability page: exemplar.dev/harness/evals .

Why it matters

Shipping a new system prompt or skill without regression checks is how “it worked in the demo” becomes production incidents. Evals turn harness changes into reviewable gates.

What Exemplar delivers

• Scenario-based evals for triage, change proposals, and tool selection
• Regression suites tied to prompt and skill versions
• Human-in-the-loop review queues for edge cases
• Score trends across model and harness upgrades

How teams use it

Attach eval suites to prompt and skill versions. Block promotion when scores regress. Use review queues for ambiguous cases before traffic shifts.

Capability checklist

• Scenario-based evals for agent workflows
• Regression suites tied to prompt/skill versions
• Human review queues for edge cases
• Score trends across upgrades