UPDATE 2, May 12, 2026: Architectural fix shipped in one day

Nafiul surprised me. He had told me the architectural change would take a week or more. Less than 24 hours later he emailed a video walkthrough of a working implementation called Global Variables. The "build once, reuse everywhere" behavior I described in the original review as missing is now real on the web app and in the Chrome extension. Variables persist through Improve Prompt and resolve at the moment a saved prompt is copied.

The MCP server still exposes only its original five operations (improve, refine, shorten, enhance, get_refinement_questions) and does not yet expose saved prompts or Global Variables, so MCP resolution is not yet on the table. I'm not penalizing for that here. Given how quickly Nafiul moved on the harder architectural piece, I'm confident we can work through MCP directly when the time is right.

The product I described in my original review is not the product I have today. Genuinely impressed, and looking forward to continuing to work with Nafiul on what's left.

------------------------

UPDATE 1, May 12, 2026: Raising from 2 stars to 5 stars

Nafiul reached out after the original review and we began working together directly. He shipped a partial fix over the weekend and committed to addressing every item I raised. His pace and responsiveness convinced me to raise the rating ahead of full delivery. Update 2 above validated that bet within hours.

------------------------

I bought Tier 3 ($199) after the founder's pre-purchase response on the AppSumo questions board described "save prompts with reusable context slots... build once, reuse everywhere." Combined with the Tier 3 listing of unlimited context slots and unlimited saved prompts, I expected that variables inside saved prompts would resolve at the moment the saved prompt was used.

After testing, that was not the case. Context slots only resolved at prompt-improvement time, not at saved-prompt-execution time. Copying a saved prompt copied the literal {slot_name} string. There was no resolution mechanism in the web app, in the copy-paste workflow, or via MCP. The MCP server exposed only enhance, improve, refine, shorten, and get_refinement_questions, with no slot resolution at the boundary.

In practice this meant that every time a context slot's content changed, every saved prompt that depended on it had to be manually rebuilt by re-running prompt improvement. That is the opposite of build-once-reuse-everywhere. It also produces inconsistent results because the improvement step uses a non-deterministic LLM.

Smaller issues at the time:

- Context slot content capped at 3000 characters, undocumented (I hit it twice on a real use case in my first two slots).
- Context slot picker had no search bar and was not alphabetized. Same issue on the category dropdown when saving a prompt.
- Model selector targeted platforms (ChatGPT, Gemini, Claude) rather than specific model versions.
- Improve Prompts output was usable when a specific model was selected but generic and occasionally hallucinatory on "Any" (one run added "Montessori expert" framing to a context that was explicitly not Montessori).
- Structural headers (ROLE, CONTEXT, INSTRUCTION) appeared in English even when the prompt body was in French, and output sometimes started in English then switched to French mid-response.

Why two stars instead of one: the underlying product idea was right and the founder was responsive in the Q&A section. The Improve Prompts feature did work, just unevenly.

What I asked for to change the review: context slots actually functioning as reusable across saved prompts (especially via MCP), search and sort on the slot picker, removal of the 3000-character cap, and naming specific models in the model selector. Updates 1 and 2 above describe what has shipped and what remains.