wtf is specjudge?

SpecJudge is a command line tool that reads the specification documents from a Spec Driven Development project and recommends which AI model actually fits the job, before you spend any tokens implementing it. The idea is to make that model choice at the moment it is cheapest to get wrong: right after you have written specs, but before any code exists. Running specjudge against a project folder produces a ranked comparison table showing each candidate model's fit rating, its price, and the date that price was recorded. Every model is judged not by raw power but by how well it fits the actual demands of your project: a rating of good means the model is capable enough and is the sweet spot, overkill means you are paying for capability the project will never use, fair means it falls somewhat short, and poor means it clearly is not capable enough. Price is shown for every model, but it only breaks ties between models that fit equally well. It never wins over a better fitting but pricier option, since choosing something incapable is treated as the most expensive mistake of all. A local, small model running through Ollama acts as the judge that estimates how demanding the project is across several dimensions. That estimate is then compared against a community maintained catalog of models and their declared capabilities using plain, readable rules, so every recommendation comes with a reason instead of being a black box. Because the judge model runs entirely on your own machine, your specs and business logic never leave it, and the optional browser report is a self contained file that loads nothing from the network. The tool is installed with a package manager called uv, requires at least one local model pulled through Ollama, and is used with a single command against any project path. It also reports clearly when it does not have enough information to make a confident recommendation, rather than guessing. The project explicitly invites contributions that need no coding at all, since the list of models and their prices is kept in a separate, human readable data file that anyone can update with a pull request.