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 agentmods add skills/frenzymath/danus/verify-proofnpx skills add frenzymath/Danus --skill verify-proofgit clone --depth 1 https://github.com/frenzymath/DanusWrote 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/frenzymath/danus/verify-proof)<a href="https://agentmods.dev/skills/frenzymath/danus/verify-proof"><img src="https://agentmods.dev/badge/skills/frenzymath/danus/verify-proof.svg" alt="Measured on agentmods" 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 | $0.00060 | $0.01326 |
| Opus 5 | $0.00030 | $0.00663 |
| Sonnet 5 | $0.00012 | $0.00265 |
| Haiku 4.5 | $0.00006 | $0.00133 |
Grade A, and why
verify-proof 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 4d 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 — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Verify Proof
The verifier is the canonical and sole authority on mathematical correctness. Mathematics requires 100% accuracy; even though this verifier is not a formal proof assistant, it is the strongest correctness check in the system. No LLM consultation, panel, or self-critique substitutes for it.
You verify and write a fact through one tool: fact_submit. It runs the
glossary-coverage check, calls the verifier, and writes the fact to the fact graph
iff the verifier accepts — there is no other way a fact enters the graph.
When to submit
- The full target theorem — when you have assembled a complete proof of the
whole problem (as a self-contained statement + proof, citing its predecessors by
fact_id). - Every intermediate result you intend to USE downstream — a lemma, a candidate construction, an arithmetic/closed-form claim, a saturation or local-to-global claim, any sharply-delimited step. Adopting an unverified partial result as a building block is the single biggest correctness risk. When in doubt, submit.
Choose a substantive mathematical boundary, not the smallest checkable line. Routine calculations, substitutions, and bookkeeping identities should normally remain internal steps in the proof of a deeper fact. The submitted fact should state one mathematically significant conclusion; depth must not be simulated by bundling several shallow or unrelated claims. Split supporting claims only when they have independent downstream uses, separate proof obligations, or need isolated verifier repair.
Do not build on an unverified finding from global memory. A conclusion /
example / counterexample there is awareness, not a brick — re-derive it as a
self-contained statement+proof and submit it before relying on it.
Before you submit — write an "ugly-proof" fact
A fact in the fact graph is written in "ugly-but-rigorous" form (the operator may call it an "ugly-proof"). The one goal of this form is that the fact is mechanically checkable for correctness by a reader with no memory and no math intuition (an agent with no recall, a human, the verifier). It is allowed — encouraged — to be ugly: redundant, machine-flavored, verbose. It is not allowed to be ambiguous, vague, or context-dependent. "Ugly" is the deliberate contrast with the polished arXiv paper (a separate pipeline); here, only mechanical correctness matters.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 4d ago First seen · 103 lines · 60 tokens per session scan A 9ce9ebc7983e
verify-proof is a skill published in the GitHub repository frenzymath/Danus (387 stars, last pushed 7d ago), licensed Apache-2.0. It adds 60 tokens to every session and 1,326 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 2026-08-30.
Other skills, from other repositories
jupyter-notebook
Iterative Python via live Jupyter kernel (hamelnb).
annotating-variants
Annotates VCF variants and normalizes HGVS nomenclature with public, license-free annotators (Ensembl VEP REST, VEP/SnpEff/ANNOVAR offline) and links variants to gnomAD population frequencies and the clinical context OpenMed extracts. Use when the user wants to predict variant consequences, map HGVS to genomic…
batch-processing-clinical-text
Run large-scale batch NER, PII extraction, or de-identification over many clinical notes on-device with OpenMed, with sharding, checkpointing, resumability, and append-only JSONL output. Use when the user needs to process a corpus or folder of notes, de-identify a dataset, run NER over thousands of documents, build a…
evaluating-with-leakage-gates
Evaluate an OpenMed de-identification or clinical NER model against the leakage-first release gates G1a through G8, which gate releases on residual PHI leakage rather than on F1. Use when the user wants to run the OpenMed eval harness on a synthetic golden set, decide whether a de-id model is RELEASABLE or…
mapping-loinc
Maps laboratory and clinical observation names extracted by OpenMed to LOINC codes using the public Regenstrief LOINC and FHIR terminology APIs. Use when the user wants to code lab tests, vital signs, or observations to LOINC, resolve a test name plus specimen and method to the correct LOINC part-model code, attach…
parsing-ccda-documents
Parses C-CDA / CCD XML clinical documents to extract human-readable section narrative plus coded entries, keyed by section LOINC codes and templateIds. Use before OpenMed processing when ingesting C-CDA R2.1 documents (CCD, Discharge Summary, H&P, Consultation Note) exported from an EHR and you need the narrative…