SkillProof tested Claude skills

Research Proof

Forces research and benchmark claims through a frozen-verifier proof ladder instead of confident prose.

Autor: tonyblu331 · tonyblu331/research-proof

Testowano · Działa ★ 9.2/10

Research Proof — Forces research and benchmark claims through a frozen-verifier proof ladder instead of confident prose.

Co robi ten skill

Turns a vague research, benchmark, or causal claim into a structured proof program with fixed Claim / Verifier Boundary / Rejection Gates / Verdict / Proof Ledger headings. Triggers on research reviews, eval design, benchmark or causal claims, and evidence-certainty questions, and forces PROVEN/SUPPORTED/REJECTED/OPEN labeling instead of unqualified confidence.

Raport z testu

Ran a real production-latency claim through both paths: the skill version forced an independent-evaluator check and a randomization rejection-gate the freeform baseline never raised, then correctly downgraded the verdict from an informal 'strongly supported' to the disciplined SUPPORTED label.

Testowano: 2026-07-14 · Claude Code 2.x (agent harness)

Instalacja

git clone https://github.com/tonyblu331/research-proof.git
cd research-proof
mkdir -p ~/.claude/skills
cp -r skills/research-proof ~/.claude/skills/research-proof

Komendy i przykładowe prompty

  • /research-proofForces research and benchmark claims through a frozen-verifier proof ladder instead of confident prose.

Skille uruchamiają się na zwykłe polecenia — bez komend do zapamiętania. Po instalacji aktywują go prompty takie jak te (po angielsku):

  • Prove this latency benchmark claim before I publish it
  • My research says X is faster — verify that properly
  • Check whether this experiment actually supports its claim

Pełny raport →