TL;DR

1. The "prompt store" category is crowded, low-ARPU, and consolidating. PromptBase leads (270k prompts, 450k users, 20% commission) but typical sellers net ~$200–300/mo. Free alternatives from Anthropic, OpenAI, Google and GitHub's 143k-star awesome-chatgpt-prompts repo crush willingness-to-pay for generic prompt packs.
2. The "prompt services" category is hot but commoditising. Upwork reports AI skills +109% YoY. Fiverr $30–$125 gigs, Upwork $35–$300/hr, Toptal $200–$300/hr. But Gartner flags "context engineering" as the 2026 breakout skill — buyers want systems work (RAG, evals, agent orchestration), not clever strings.
3. No incumbent cleanly serves the combined thesis. PromptBase bolted a "Hire" page onto the store but it lacks escrow, SOW workflow, or matching. Snack Prompt was acquired by Spectral Capital (Oct 2025), PromptPerfect is sunsetting Sept 2026, FlowGPT rebranded to Kaon. Shakeout energy.
4. The wedge is services-led, vertical, eval-first — NOT another prompt store. A curated, SOW-driven marketplace with escrow, evals, and versioning built in, wedged into legal / medical / code / compliance where hallucination cost is high and buyers pay for verifiable quality. A small catalogue = lead-gen, not core revenue.
5. Single biggest risk: frontier models are eating prompt-craft. A 9,649-experiment study shows context quality dominates prompt-wording. Skill is repositioning from "prompt engineer" to "context / skill / harness engineer" — the product should follow that shift, not fight it.