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 vince-gonzalez/gonzalgo --skill lean-axiom-provenancegit clone --depth 1 https://github.com/vince-gonzalez/gonzalgoWrote 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/vince-gonzalez/gonzalgo/lean-axiom-provenance)<a href="https://agentmods.dev/skills/vince-gonzalez/gonzalgo/lean-axiom-provenance"><img src="https://agentmods.dev/badge/skills/vince-gonzalez/gonzalgo/lean-axiom-provenance/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/vince-gonzalez/gonzalgo/lean-axiom-provenance"><img src="https://agentmods.dev/badge/skills/vince-gonzalez/gonzalgo/lean-axiom-provenance.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00203 | $0.02579 |
| Opus 5 | $0.00102 | $0.01290 |
| Sonnet 5 | $0.00041 | $0.00516 |
| Haiku 4.5 | $0.00020 | $0.00258 |
Grade A, and why
lean-axiom-provenance 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 12d 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 — 227 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Lean axiom provenance
#print axioms tells you a theorem depends on Classical.choice. It does not
tell you which step introduced it, whether the theorem's own statement required
it, or whether a constructive route exists. This skill answers those, over a
whole environment rather than one declaration at a time.
The tool produces facts. Reading them without overclaiming is the hard part and most of this file is about that.
Setup
pip install gonzalgo
Analysis runs against a dump: a declaration graph written out of a built Lean environment. Generate it once, then ask it many questions.
gonzalgo lean-files ./scripts # writes the Lean extractors
cd my-lean-project
lake build # the environment must already be built
lake env lean scripts/Split.lean # -> mathlib_split.tsv
gonzalgo check mathlib_split.tsv # verify it contains proof terms
Run check and do not skip it. ConstantInfo.value? returns none for
theorems unless it is passed allowOpaque := true. An extractor written the
obvious way silently reads every theorem's proof as empty, and then every
statement-versus-proof figure is not imprecise, it is measuring statements and
labelling them proofs. check is what catches that.
A dump of Mathlib takes about three minutes and roughly 600 MB. For a small project it is seconds.
If the environment is large, lake env lean needs -D maxErrors=4000 on the
command line. set_option maxErrors inside a file is ignored, and without the
flag Lean stops at 100 errors long before the last declaration.
The questions
What does this project rest on?
gonzalgo trust mathlib_split.tsv
gonzalgo trust mathlib_split.tsv --fail-on-trust # non-zero exit, for CI
Every theorem reaching sorryAx anywhere upstream, and everything carrying
Lean.ofReduceBool / Lean.ofReduceNat, which is what native_decide emits —
those results were obtained by compiling and running code and believing the
answer, so the compiler and runtime are trusted rather than the kernel. Lean
warns once, on the line where the sorry was typed; it says nothing about a
theorem three files later that uses that lemma and is therefore also not proved.
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.
- 12d ago First seen · 227 lines · 203 tokens per session scan A a2a4492e035d
lean-axiom-provenance is a skill published in the GitHub repository vince-gonzalez/gonzalgo (2 stars, last pushed yesterday), licensed Apache-2.0. It adds 203 tokens to every session and 2,579 once invoked, about $0.0010 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 2026-08-31.
Other skills, from other repositories
automatic-differentiation-advanced
Extend Sounio's automatic differentiation capabilities to higher‑order derivatives, Hessians, differentiation through control flow, and GPU‑accelerated gradient computation, enabling state‑of‑the‑art scientific machine learning.
epistemic-uncertainty-quantification
Extend Sounio's epistemic types with advanced uncertainty quantification methods, such as confidence intervals, non‑Gaussian distributions, and Dempster‑Shafer evidence combination, positioning the language at the state‑of‑the‑art in measurement science.
formal-verification-epistemic
Integrate formal verification tools (SMT solvers, Lean) to prove properties of epistemic programs, such as uncertainty bounds, confidence guarantees, and provenance integrity, establishing Sounio as a language with mathematically verified scientific claims.
Multimodal Alignment
Align speech, text, image, or video signals for multimodal benchmarks.
Academic Writing
Rewrite research prose to be clear, concise, precise, and appropriately cautious.
Citation Checking
Check whether citations support empirical claims and flag unsupported statements.