---
title: "promptgen"
description: "Turn rough instructions into optimized, evidence-based prompts for AI agents, coding agents, tools, eval graders, and subagent briefings. Built on Anthropic / OpenAI guidance through April 2026 and the 2025-2026 academic"
canonical: https://agentpluginsdirectory.com/plugins/promptgen
last-updated: 2026-10-10
---

# promptgen
Turn rough instructions into optimized, evidence-based prompts for AI agents, coding agents, tools, eval graders, and subagent briefings. Built on Anthropic / OpenAI guidance through April 2026 and the 2025-2026 academic literature. Copies to the clipboard on macOS, Linux and Windows.
- Slug: promptgen
- Publisher: Smykla Skalski Labs
- Repository: https://github.com/smykla-skalski/sai
- Manifest: plugins/promptgen/plugin.json
- Version: 3.1.5
- License: MIT
- Category (editorial): other
- Skills: 1 (promptgen)
- MCP servers: 0
- Stars: 5
- Repository created: 2026-02-12
- Repository last pushed: 2026-10-10
- Publisher type: Organization
- Listing: https://agentpluginsdirectory.com/plugins/promptgen
- Schema: https://agent-plugins.org/schemas/1.0.0/plugin.schema.json

## What promptgen does, in the publisher's words

Turn rough instructions into optimized, evidence-based AI prompts. Built on Anthropic and OpenAI guidance through April 2026, the 2025-2026 academic literature (Mollick / Wharton Prompting Science Reports 1-4, Chroma context-rot research, GEPA, IFScale, "Reasoning Models Struggle to Control CoT"), Simon Willison's lethal trifecta and Meta's Rule of Two, and current agent / coding-agent patterns (AGENTS.md, SKILL.md, three-agent harness, ACI design).

You describe what you want a prompt to do in plain language. The skill generates a well-structured prompt with an outcome contract, observable done-when criteria, evidence rules, side-effect boundaries, fitting specificity, and no common anti-patterns. Output goes to clipboard by default.

Supports four target formats: Claude (XML tags), GPT (outcome-first Markdown), Codex (coding-agent contract), and generic (Markdown-only). Handles task prompts, system prompts, reusable templates, tool descriptions, eval graders, subagent briefings, three-agent harness instructions, and prompt-improvement passes.

From the project README, punctuation lightly normalized. Full text: https://raw.githubusercontent.com/smykla-skalski/sai/HEAD/plugins/promptgen/README.md