Prompt Optimizer

About the project

An open-source AI agent skill that turns natural-language development requests into inspectable, model-adapted execution contracts behind a human approval gate before any file is modified. Published on GitHub under the MIT license at v1.0.0.

Our approach

213 official technical documentation pages were crawled and distilled into four offline Markdown reference packs with zero external network dependencies. Incoming prompts are routed across three gears: read-only questions bypass overhead, /small locks single-target surgical edits, and multi-file work reads project handoff documents before compiling a runtime-specific contract across Claude 5.5/4.x, Gemini, GPT/Codex and local Qwen on MLX.