Show HN: Promptometer – feedback on your AI Agent system prompts (arpx.space)

Promptometer is a small tool/service that gives actionable feedback on system prompts used to steer AI agents. When you submit a system prompt it’s judged by a locally run LLM (GPT-OSS via Ollama) against best practices distilled from Anthropic’s research on context engineering. The output highlights strengths (e.g., specificity and clear expectations) and suggests concrete fixes—examples shown include adding target-audience context and specifying output format—while scoring clarity, token efficiency, structure, and how well the prompt respects a model’s attention budget. This matters for builders turning experiments into production: system prompts are the single biggest lever on agent behavior, and subtle prompt design choices affect reliability, cost and generalizability. Using an LLM-as-judge approach encodes modern heuristics into rapid feedback loops while keeping data local—Ollama is used to run GPT-OSS and the project asserts prompts aren’t stored. Beyond the tool, the team offers consultancy to help productionize agents (prompt engineering to full architecture), positioning Promptometer as both a diagnostics utility and a pathway for companies wanting to scale reliable AI workflows.