About
Why PromptMarket
Ask an LLM to “write me a system prompt for a tool-using agent” and you already know what comes back: plausible-sounding, but missing the refusal boundaries, ambiguity handling, and output contracts that actually hold up once real traffic hits it. We kept re-deriving the same diagnostic questions under deadline pressure — system prompt review, failure diagnosis, injection hardening — so we wrote them down properly instead.
Three packs came out of it, one for each stage of the loop: Baseline Presets to start from, Redline Prompts to audit against your own traces, Press Check to grade the output at scale.
The packs are drafted and reviewed with AI models and curated by the author. We'd rather you know that up front.
Nothing here is a first draft
Every Redline prompt goes through the same four stages: draft against a category brief, refine against one question — would a professional pay for this — score on specificity and uniqueness, then adversarial review by three separate AI reviewer passes briefed to argue against including it. Baseline Presets and Press Check add mechanical checks on top: every JSON and JSONL file has to parse and validate before a pack is built.
6 of 56 Redline drafts didn't survive review
A Redline prompt needs at least two "include" votes from three AI reviewer passes to ship. Ties and uncertainty default to rejection — six drafted prompts were cut and replaced before the pack went out the door.
Built for engineers, not prototypers
This is written for developers and small teams already running LLM-powered agents in production — maintaining something with real users and real failure modes, not experimenting with a first prototype.