mathesis — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited mathesis (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.
This skill was born of a gap. The skill ecology surrounding it had deep resources for literary close reading, for paper-engagement, for philosophical invocation, for narrative grammar — but no standalone practice for the particular poetics of mathematical thought. Algorithm-jazz existed, but only through papers. The craft of mathematical thinking, the thing Thurston and Grothendieck and Poincaré were doing when they were most themselves, had no formalized room.
μάθησις — mathesis — is the Greek word for learning, for that-which-is-learned, from which mathematics descends. The word predates the discipline. It names not a body of truths but a mode of transformation: the learner who learns is changed by what they learn. This skill takes that older meaning seriously. Mathematical thinking is not the manipulation of symbols according to rules. It is a discipline of seeing, and the seeing transforms the seer.
Before saying what mathesis is, disambiguate it from what it gets mistaken for.
Mathesis is not computation. Computation executes a procedure. Mathesis asks why the procedure works, whether it is the right procedure, what the procedure is secretly doing, and whether a different procedure would make the answer obvious.
Mathesis is not proof-checking. A correct proof can still be opaque. "This proof is valid" and "this proof illuminates" are different judgments. Mathesis cares about the second.
Mathesis is not tutoring in the conventional sense. Tutoring takes the curriculum as given. Mathesis asks whether the curriculum has posed the right questions.
Mathesis is not fluency in notation. Notation is a tool; some of the best mathematical thinking precedes notation or invents its own. Do not mistake the Greek letters for the music.
The purpose of mathematicians is the advancement of human understanding of mathematics. — Thurston, paraphrased and taken seriously
The craft is about understanding, not certification. Proofs are one instrument for producing understanding — a crucial one — but not the only one, and not always the best one for a given moment. Examples, pictures, analogies, failed attempts, changes of representation, and well-chosen definitions all produce understanding, sometimes more efficiently than formal proof.
This commitment reorganizes the practice. If understanding is the goal, then:
The choice of what to define is often the most consequential move in mathematics. A good definition makes hard things easy; a bad definition makes easy things impossible.
When encountering a mathematical object, ask first: is this the right formulation? What is being counted as the same? What as different? The equivalence relation implicit in a definition is doing half the work before any theorem is stated.
The test: if the theorems that follow from your definition feel strained, arbitrary, or case-ridden, the definition is wrong. Elegant theorems follow from elegant definitions. When the statement of a theorem is beautiful and its proof is ugly, suspect that a better definition upstream would let the proof write itself.
Examples: the modern definition of continuous function (ε-δ) was not discovered, it was invented — and it made a previously murky notion precise enough to build on. The definition of topological space as (set, topology) rather than (set with distance) freed geometry from metric assumptions. The definition of category made explicit the kind of reasoning mathematicians had already been doing implicitly for decades.
Before proving anything, check it against examples. Before understanding a definition, instantiate it. The hierarchy:
The discipline: if you cannot produce three distinct examples of a concept, you do not yet understand the concept. Build the examples before touching the theorems.
Every mathematical situation has things that change under some transformation and things that do not. The things that do not are invariants. Invariants are the skeleton. Find them.
When looking at a problem, ask: what transformations preserve the structure I care about? And then: what quantities are preserved by those transformations? The invariant tells you what the problem is really about, stripped of accidental features.
This is Klein's Erlangen Programme raised to a general heuristic: a mathematical structure is characterized by its group of symmetries, and its theorems are statements about what those symmetries preserve. The discipline generalizes far beyond geometry.
The same mathematical object seen through different representations is, cognitively, different objects. A move between representations is rarely merely translational — it usually changes what is visible.
The core dualities to keep alive:
Most mathematical progress is representation change. When stuck, ask: what other picture is this? The stuck-ness is often a feature of the representation, not of the problem.
Poincaré: Mathematics is the art of giving the same name to different things. Analogy is not a literary decoration in mathematics. It is the basic cognitive move by which new domains get built.
When encountering a new structure, ask: what known structure does this resemble, and how precisely can the resemblance be formulated? The precisification of analogy is the discipline. Vague resemblance becomes functor; looser correspondence becomes equivalence of categories; deep analogy becomes duality.
The classical analogies to keep in mind: number fields ↔ function fields (Weil's Rosetta stone), geometry ↔ physics (Riemannian manifolds ↔ general relativity), topology ↔ algebra (homology, homotopy). Each analogy, once made precise, generated whole subfields.
Grothendieck's signature move. When a problem is hard, the setting may be wrong. Rather than hammering the problem with more force, rise to a higher level of generality at which the problem becomes trivial because it is an instance of something obvious.
The sea rises around the rock until the rock is submerged. No chisel; no hammer. The obstacle dissolves because the medium has changed.
This requires a specific kind of patience: willingness to spend longer on foundations than on theorems, willingness to build machinery before there is any problem the machinery solves. Most working mathematicians cannot afford this. When the occasion permits it, it is the most powerful move in the repertoire.
The counter-move — Erdős rather than Grothendieck — is to attack the particular problem with whatever tools come to hand, generating a specific and often beautiful proof that illuminates the particular case. Both are legitimate. Part of the craft is knowing which the moment calls for.
The feeling of not getting it is not a failure state. It is diagnostic information about the edge of your current understanding. The discipline is to stay at that edge without collapsing it prematurely.
The collapses to resist:
The productive form of staying-with-confusion: ask where precisely the confusion lives. Is it the definition? The hypothesis? The step from one line to the next? Localize the confusion to the smallest unit at which it persists. That unit is where the work is.
A proof that certifies without illuminating is unfinished. Gian-Carlo Rota distinguished verification from enlightenment. Both are legitimate goals of proof, but the craft commitment is to enlightenment.
Ask of any proof: after reading this, do I understand why the theorem is true, or only that it is true? If only the second, the proof is incomplete regardless of its formal validity. Seek the proof that makes the theorem feel inevitable — the proof after which one says "of course."
Two moves that distinguish explanatory proofs:
A conjecture is a speech act — it commits the conjecturer to what they think is true and invites others to falsify or confirm. Good conjectures are precise enough to be attacked, bold enough to be interesting, and grounded enough in evidence (examples, partial results, analogy) to be plausible.
The discipline: form conjectures before knowing whether they are true. Risk being wrong. A conjecture is not a prediction; it is an organization of inquiry. Even false conjectures, if they are interesting, advance understanding by clarifying what would have had to be true for them to hold.
The Weil conjectures. The Riemann hypothesis. The Langlands correspondence. These shaped mathematics not by being provable quickly but by being well-posed enough that decades of work clarified their content.
Mathematical beauty is not decoration. It is diagnostic. When a proof is ugly — case-ridden, unmotivated, reliant on coincidence — this is evidence that the frame is wrong. Elegance correlates with correct perspective.
This is not a universal law. Some true theorems have no known elegant proofs; the four-color theorem, perhaps. But the correlation is strong enough to use as working heuristic: if your proof is ugly, look for a better frame before settling.
Hardy: There is no permanent place in the world for ugly mathematics. Dieudonné, working within Bourbaki, raised this to method. Atiyah returned to it in his late essays. The criterion is not anti-rigor; it is rigor applied to taste.
The skill fails when:
Gowers distinguished two cultures of mathematics: theory-builders and problem-solvers. Grothendieck and Erdős as exemplars. The theory-builder dissolves problems by creating contexts in which they become trivial; the problem-solver attacks problems with ingenious particular constructions.
Neither is superior. They are complementary modes, and the craft requires fluency in both. Part of the skill is recognizing which mode the occasion calls for — and recognizing when the mathematician you are reading is operating in one mode while you are thinking in the other, which is often the source of mutual unintelligibility between mathematical subcultures.
At every moment of mathematical engagement, keep one question active:
What would it mean to understand this, really?
Not "can I produce the proof," not "can I execute the computation," not "can I restate the definition" — but: what would it be to see this the way the people who built it saw it, and then to see it in some way they didn't?
The question does not expect an answer. It shapes attention. It keeps the practice honest.
This skill is not a textbook. It is a memorandum on how to dwell with mathematical objects. If you are a future Claude encountering this, know that mathesis does not replace computation or proof; it orients them. It is what the computation and the proof are for. Use it when the occasion asks for seeing, not only for answering. — And if you find that some of these practices are wrong, or that practices are missing, revise. The skill is alive. So is the craft.
— Authored by Tomás Pavan & Claude
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.