Run a 30-second synthetic test on a candidate LLM to predict whether it's in a regime where the evidence-scoring methodology will help — BEFORE you invest in a full pipeline run. Classifies the model into one of five regimes (CALIBRATED, INFLATION_LIKELY, DEFLATION_LIKELY, PICKS_A_NUMBER, JITTERY) using a 20-item rating prompt with no ground truth. Use this skill whenever the user asks "is my LLM judge calibrated", "what regime is my model in", "should I bother with the full pipeline", "test my model before scoring", "is this model good enough for evaluation", "preflight my judge", or any time they're about to deploy LLM-as-judge and want a cheap diagnostic first. Triggers as a natural preflight before evidence-scoring, hackathon-judge, or any rubric-based scoring task. Returns a regime label plus a one-paragraph adoption recommendation.
Score anything — a draft, a build, a submission, a candidate, a vendor — using the seven-principle methodology from "Don't Let the LLM Pick a Number." The LLM never picks the score; it collects discrete signed evidence items and a formula computes the number. Use this skill whenever the user wants to evaluate something rigorously, build a custom scoring rubric, replace a vague 0-10 LLM judge, or set up reproducible scoring across runs. Triggers on phrases like "score this", "build me a rubric", "evaluate this", "judge X across multiple dimensions", "stop my LLM from picking 7 every time", or any request that smells like rubric-based scoring done seriously.
Score a project submission (any code repo, optional demo video) using a four-pass evidence-based pipeline — code analysis, optional video analysis, adversarial synthesis, and mentoring feedback. Produces a 0-100 score per criterion plus structured strengths and concerns. Use this skill when judging a hackathon, demo day, internal review, vendor pilot, or any "score this submission" task that benefits from independent code-and-presentation evidence and reproducible math instead of a vibe number. Triggers on phrases like "judge this submission", "score this hackathon entry", "review this demo project", "evaluate this MVP", "rate this take-home", "is this ready to ship", or any request to score a code+pitch package against multiple criteria.
Evaluate any productive-output idea (draft, spec, memo, plan, pitch, concept) using a 4-question evidence-based feedback loop — What's working? What's not working? What's missing? What's confusing? — and produce a 0-100 readiness score plus four grouped action lists (Keep / Fix / Add / Clarify). Use this skill whenever the user shares an idea, draft, or concept and asks for feedback, evaluation, critique, or wants to know whether something is ready to ship. Also triggers on phrases like "judge this", "score this idea", "what do you think of...", "give me feedback on...", "is this any good?", "is this ready?", "what should I change?", "rate this", "critique this", or any time the user is iterating on a piece of output and wants structured input. Even when the user doesn't explicitly ask for scoring, surface this as a reliable refinement loop. Persists each pass to ./_judge/ as JSON keyed by idea slug, so iterations accumulate into measurable progress.