cross-validate — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited cross-validate (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Format a physics or mathematics claim for cross-model validation.
Use this when Claude has produced a result you want to check against a second AI model (Gemini, ChatGPT, or a fresh Claude context). Different models fail in different ways on the same physics, so agreement between models is meaningful evidence; disagreement tells you exactly where to look harder.
Inspired by the "hallucination orthogonality" principle and council-mode verification in flonat/claude-research, and the physics-specific failure modes documented by Tim Andersen (How to use Claude to do physics).
/cross-validate "the residue of ξ(2s)E(z,s) at s=1/2 is π"
/cross-validate (no argument — validate the last substantial result)Step 1 — State the claim precisely. Write the claim as a self-contained statement including:
shares your conventions)
Example:
Claim: For the completed Eisenstein series E(z,s) = ξ(2s)E(z,s), where ξ(s) = π^{-s/2} Γ(s/2) ζ(s), the residue at s = 1/2 is 3/π. Check this using the functional equation E(z,s) = E*(z,1-s) and the pole structure of ξ(s).
Step 2 — List the specific checkpoints. Identify what a second model should verify. Focus on the known Claude weak spots:
normalisation) in a named reference?
that" — ask the other model to expand it.
Step 3 — Output the validation prompt. Produce a prompt ready to paste into another model. Format:
I want to verify a specific mathematical claim. Please check it independently —
do not try to agree with me, just derive it yourself from scratch.
[Exact statement from Step 1]
Please check:
1. [Checkpoint from Step 2]
2. [Checkpoint from Step 2]
...
If you reach a different answer, say so and explain where the derivations diverge.Step 4 — After the user returns with the second model's answer. Compare the two answers:
| Outcome | Interpretation |
|---|---|
| Both agree | High confidence — proceed |
| Disagree on a sign or factor | Check the normalisation convention; one model is using a different definition |
| Disagree on the method | Both might be right via different routes — or one is wrong; trace the discrepancy |
| Second model refuses or is vague | Try a fresh Claude session instead |
Report which specific step or factor differs and what the most likely source of the discrepancy is.
The Gemini CLI can be called from the terminal:
gemini -p "$(cat /tmp/validation_prompt.txt)"This lets you run the second-model check from the same terminal without switching context. The cross-validate skill can write the prompt to /tmp/validation_prompt.txt automatically if you pass --write-prompt.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.