PQS - A Quality Gate For Prompts

PQS injection detection - finds instruction override, role hijack, data exfiltration, and delimiter attacks; returns a sanitized prompt

Quick pre-flight quality score: composite (0-100) + grade (A-F) + three framework sub-scores (PEEM, RAGAS, G-Eval) + verdict. Run upstream of any other paid call to filter weak prompts before inference. The cheapest gate on the AI input quality problem.

Pass/Fail before you waste budget on a doomed call. Decide whether to make a paid LLM call before you make it. Preflight scores your prompt against an 8-dimension quality rubric (pass/fail at 28/80) and optionally pulls the target endpoint's on-chain trust signals — total payments, unique payers, last activity. Two-in-one circuit breaker for agents that don't want to waste budget on weak prompts or sketchy endpoints. (rev 2)

PQS preflight - prompt quality plus optional endpoint trust check before inference (Solana)

Venice.ai chat completion with PQS preflight scoring. Returns LLM response plus 8-dimension quality score in one paid call. No signup, no API key.

PQS score - 8-dimensional prompt quality breakdown across 5 frameworks with top fixes

PQS batch scoring - score up to 10 prompts in one call (rev 2)

PQS batch scoring - score up to 10 prompts in one call (Solana)

PQS cross-model scoring - same prompt through Claude and GPT-4o, judged by a third model (rev 2)

PQS cross-model scoring - same prompt through Claude and GPT-4o, judged by a third model (Solana)

PQS full analysis - optimized prompt plus before/after outputs (rev 2)

PQS full analysis - optimized prompt plus before/after outputs (Solana)

PQS basic prompt score - 8-dimensional prompt quality breakdown with top fixes (Solana)

PQS variants - generate N ranked prompt rewrites (1-5) scored on the 8-dimension rubric