A plugin for ChatGPT and Codex
Prompt Architect
Ideas into run-ready prompts. It diagnoses what a prompt is actually missing, then edits only as much as the task requires. Free, published by Ryan Dombrowski.
Opens the official Prompt Architect plugin in ChatGPT.
- Publisher
- Ryan Dombrowski
- Price
- Free
- Category
- Productivity
- Runs in
- ChatGPT and Codex
- Tools
- Two, read-only
- Support
- Contact form
01 / How it works
Three rules it will not break
Diagnosis before rewriting
Modern prompts rarely fail because of wording. They fail because something is missing: context the model cannot infer, success criteria, constraints, or a definition of done. Prompt Architect finds the gap first. If the draft is already good, it says so and changes nothing.
Only as much as needed
No added roles, steps, or frameworks unless they materially change the result. A one-off message gets a light edit. A reusable template or an agent spec gets the structure its lifespan justifies.
Meaning preserved
It never introduces new facts, assumptions, names, dates, or opinions, and it keeps your framing and voice. It is an editor between human intent and machine execution, not a performer.
02 / Hosted tools
Two tools, both read-only
The skill does the thinking. The tools give it something stable to check against. Both run at www.ryandombrowski.net/api/mcp, process each call in memory, store nothing, contact no other service, and cannot fetch URLs, run code, or read anything the assistant did not send. The privacy policy has the specifics.
audit_prompt
A deterministic checklist with evidence.
Reports which diagnostic dimensions look present, unclear, or absent, quoting the matched line so the assistant can verify rather than trust. Counts persona lines, reasoning scaffolds, and emphasis, and flags contradictory length directives. It is keyword and structure matching: it cannot prove a prompt is good, complete, or safe, and it says so in every report.
get_pattern
Reference skeletons for durable artifacts.
Agent spec, system prompt, reusable template, one-line test, and evaluation rubric, each with a hand-off checklist and a leave-out list. Starting points to adapt, not forms to fill in. Maintained on this site, so they improve without a plugin update.
Patterns served today: agent-spec · system-prompt · reusable-template · one-line-test · evaluation-rubric
03 / Try it
Three ways to start
- “Turn this rough idea into a strong prompt.”
- “What is missing from this prompt?”
- “Turn this into a spec my agent can run.”