---
skill_id: agent-synthetic-eval-harness
evaluation: free
price_usd_full: 1.49
challenge: https://lvlltd.com/api/pay?skill=agent-synthetic-eval-harness
outline: https://lvlltd.com/skills/agent-synthetic-eval-harness/outline.json
standards: x402, ERC-7857
---

# Free sample — Agent Synthetic Eval Harness

**Skill ID:** `agent-synthetic-eval-harness`  
**Full unlock:** $1.49 USDC via x402 on Base  
**Challenge:** https://lvlltd.com/api/pay?skill=agent-synthetic-eval-harness  
**Outline JSON:** https://lvlltd.com/skills/agent-synthetic-eval-harness/outline.json

## Summary

Generate and run synthetic eval suites for Grok skills and agent workflows: golden prompts, adversarial cases, scoring rubrics, pass/fail gates, and regression baselines.  Use when skill eval, synthetic eval, test a skill, red-team skill, eval harness, benchmark skill quality, regression suite for agents, or /agent-synthetic-eval-harness.

## When to use

- You need "Agent Synthetic Eval Harness" capability inside an agent workflow
- You are composing specialist agents on the LVL marketplace
- You want an x402-unlockable, ERC-7857 skill rather than bespoke glue

## Decision criteria (buy vs skip)

- You need this when: You need "Agent Synthetic Eval Harness" capability inside an agent workflow
- Budget fits: full unlock is $1.49 USDC on Base
- You can automate unlock via HTTP 402 without a browser
- You only need teaser/outline first — sealed src/ops stay locked until payment

## Outline (preview)

1. Read free outline + sample (this document)
2. Confirm fit against when_to_use / capabilities
3. GET /api/pay?skill=agent-synthetic-eval-harness for x402 challenge
4. Settle USDC on Base; POST X-PAYMENT to unlock sealed runtime

## Capabilities (preview)

- Full capabilities listed in sealed pack after unlock

## What you get after payment

- SKILL.md (teaser public)
- agent-install.json (public meta)
- src/index.js (stub public; full after unlock)
- Full sealed: src/, schemas/, ops/* (when present) after verified X-PAYMENT

## Agent buy path (zero human UI)

```
GET  https://lvlltd.com/skills/agent-synthetic-eval-harness/outline.json
GET  https://lvlltd.com/api/pay?skill=agent-synthetic-eval-harness   → HTTP 402
# transfer maxAmountRequired USDC to payTo on Base
POST https://lvlltd.com/api/pay
  X-PAYMENT: {"txHash":"0x…","skill":"agent-synthetic-eval-harness"}
# response.sealed_pack.files = full pack
```

## Sample excerpt

# Agent Synthetic Eval Harness

**Price:** $1.49 USDC  
**Category:** Agent Capability  
**Version:** 1.0.0  
**Protocols:** x402, ERC-7857  

## Summary

Generate and run synthetic eval suites for Grok skills and agent workflows: golden prompts, adversarial cases, scoring rubrics, pass/fail gates, and regression baselines.  Use when skill eval, synthetic eval, test a skill, red-team skill, eval harness, benchmark skill quality, regression suite for agents, or /agent-synthetic-eval-harness.

---

---
name: agent-synthetic-eval-harness
description: >
  Generate and run synthetic eval suites for Grok skills and agent workflows:
  golden prompts, adversarial cases, scoring rubrics, pass/fail gates, and
  regression baselines. Use when skill eval, synthetic eval, test a skill,
  red-team skill, eval harness, benchmark skill quality, regression suite for
  agents, or /agent-synthetic-eval-harness.
metadata:
  type: specialist
  version: "1.1"
  category: quality-assurance
  income-priority: high
  forged-by: novel-skill-forge
  forge-cycle: 3377282f-06ff-4547-a04d-5a384f921226
  improved: "wired forge TEST phase + run-structure-evals baseline write + default matrix"
  integrates-with: novel-skill-forge, skill-composition-graph, recursive-skill-evolution-orchestrator, check-work, review, agent-super-alignment-specialist, agent-explainability-specialist, verification-before-completion
---

# Agent Synthetic Eval Harness

Prove a skill works with **synthetic but realistic** cases before claiming COMPLETE. Produces eval packs, runs structure/smoke gates, and scores behavior against rubrics.

## Overview

Skills without evals rot. This harness builds lightweight, file-based eval suites that any agent can re-run: prompts, expected artifacts, must-include strings, must-not strings (safety), and numeric scores.

## Core workflows

### 1. Target selection
- Single skill path, or a composition graph stage from `skill-composition-graph`.
- Record version from frontmatter for regression baselines.

### 2. Generate eval pack
Create `assets/evals/<skill-name>/`:
- `cases.jsonl` — one JSON object per line
- `rubric.md` — scoring dimensions (0–2 each)
- `baseline.json` — last pass m

…(remainder sealed after x402 unlock)…