Send one broken prompt + context. Get back: (1) a rewritten prompt with guardrails, (2) a mini evaluation set, and (3) a simple scoring harness so you can iterate without guessing.
Diagnosis + prioritized fixes + rewritten prompt skeleton (no eval harness).
Includes 10–20 test cases + scoring rubric + lightweight harness template.
For multi-step agents: tool-use constraints, routing prompts, and failure-mode tests.
Payment currently via USDC invoice link. If you need a different method, note it in intake.
Alternative to card payment. Send USDC on Base mainnet to:
0xcd1a222e211f275460195D50BDdEFC26457c35Cd
After submitting the intake form above, you'll receive an Order ID. Enter it here with your transaction hash to confirm payment: