Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add flonat/flonat-research --skill lean-checkgit clone --depth 1 https://github.com/flonat/flonat-researchWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/flonat/flonat-research/lean-check)<a href="https://agentmods.dev/skills/flonat/flonat-research/lean-check"><img src="https://agentmods.dev/badge/skills/flonat/flonat-research/lean-check/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/flonat/flonat-research/lean-check"><img src="https://agentmods.dev/badge/skills/flonat/flonat-research/lean-check.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 88 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00063 | $0.01729 |
| Opus 5 | $0.00032 | $0.00864 |
| Sonnet 5 | $0.00013 | $0.00346 |
| Haiku 4.5 | $0.00006 | $0.00173 |
Grade A, and why
lean-check scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Lean Check: Machine-Prove a Self-Authored Lemma
Formalize a lemma/theorem in Lean 4 + mathlib and let the kernel check it. A lake build that succeeds with no sorry and no extra axioms is a machine-verified proof — the strongest guarantee available.
When to Use
- A critical lemma whose correctness you want beyond doubt (the load-bearing step of a theorem).
lean-check, "formalize this in Lean", "machine-check this lemma", "prove this in Lean 4".- After
numerical-checkfails to falsify a claim and it's important enough to prove.
When NOT to Use
| Situation | Use instead |
|---|---|
| Stress-test / hunt a counterexample to a distributional claim | numerical-check (R1) |
| Verify an algebra / derivative / limit / closed-form step | symbolic-check (R2) |
| A statement too rich to faithfully formalize in reasonable time (heavy measure theory, bespoke objects) | domain-reviewer — do NOT force a lossy Lean statement |
Position in the verification spectrum
R3 — formal machine proof. The top rung: lake build (clean, sorry-free) = a kernel-checked theorem. Cost is high (formalization effort + statement fidelity), so reserve it for the claims that matter most; use R1/R2 to triage first.
Toolchain (pre-seeded — do not re-download)
- Machine: Mac Mini (
[server]). Checkhostname; if on the MacBook, run viassh mini. - Project:
~/lean-verify/mathlib_verify/— Lean4.31.0, mathlibv4.31.0(cache-backed, ~7.2 GB.lake). Health check:cd ~/lean-verify/mathlib_verify && lake build MathlibVerify.SmokeTest. - Refresh mathlib later:
lake update && lake exe cache get.
Procedure
1. State the lemma FAITHFULLY (the hard part — get this right or the check is worthless)
- Write the Lean statement so it provably matches the informal claim. A too-weak, too-strong, or subtly-different statement that happens to
buildgives false confidence — the single worst failure mode. - Before proving, read the Lean statement back against the paper's exact hypotheses and conclusion. State every hypothesis (domains,
0 < ρ < 1,StrictMono, etc.). When unsure the encoding is faithful, ask the user to confirm the statement. - If the object cannot be faithfully stated in available mathlib (e.g. a bespoke distributional limit), STOP — report
INCONCLUSIVE (not faithfully formalizable); do not ship a lossy proxy.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 116 lines · 63 tokens per session scan A 513068ee1f24
lean-check is a skill published in the GitHub repository flonat/flonat-research (132 stars, last pushed 14d ago), licensed MIT. It adds 63 tokens to every session and 1,729 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
mathmodel-skill
An end-to-end workflow for mathematical-modeling competitions, including CUMCM, MCM/ICM, and the Electrical Cup. It guides teams from choosing a problem through modeling, solving, checking, writing, rule compliance, and final review.
latex-compile
Compile a LaTeX document and fix every error plus aesthetic issue (overfull/underfull boxes, widows, alignment, fonts) for a clean PDF and log. Use this instead of running pdflatex/latexmk manually — it avoids the latexmk stale-log trap and silent grep failures on binary log output, and it reformats rather than…
nb-to-wolfbook
Convert Mathematica .nb or .m files to Wolfbook .wb format so they open and run in VS Code. Use when bringing existing .nb/.m files into Wolfbook, or to make an existing .wb bridge-safe.
sync-wb-nb
Propagate a change made in a Wolfbook .wb notebook into the paired .nb notebook so the two stay identical. Use immediately after every .wb edit.
wolfram-headless
Run heavy Wolfram Language (wolframscript) computations from Claude Code reliably, and diagnose the misleading "The product exited because of a license error". Use whenever invoking wolframscript on a non-trivial computation, when a wolframscript job dies with a "license error" despite a valid license, or when Wolfram…
cross-validate
Format a result, derivation, or numerical value for independent verification by a second model. Use when you want a cross-check on an important or contested result.