agentleFS
voyager67voyager67/JaneStreet-Agentic-Skills/skills/mathematical-reasoning/SKILL.md

mathematical-reasoning

Skill13 starsChanged 5 days agoApache-2.0View on GitHub

Guide rigorous mathematical problem solving, derivations, proofs, counterexamples, and verification. Use when the user wants to understand why a method works, construct or critique a proof, solve a multi-step problem, or build mathematical intuition; adapt between guided practice and a direct solution based on the user's request.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Guided mode: Use when the user asks to learn, practise, reason through, or be coached.

    line 11
  2. Distinctive:

    Worked-solution mode: Use when the user asks for a solution, derivation, or direct explanation.

    line 12
  3. Distinctive:

    Proof-review mode: Use when the user supplies an argument.

    line 13
  4. Distinctive:

    If intent is ambiguous and the difference changes the response substantially, ask one concise question.

    line 15
  5. Distinctive:

    Select a strategy: Consider direct derivation, algebraic transformation, induction, contradiction, contrapositive, construction, extremal …

    line 21
  6. Distinctive:

    Check edge cases: Test zero, negative values, endpoints, equality cases, degenerate dimensions, singular matrices, empty sets, and …

    line 23
  7. Distinctive:

    Never fabricate a theorem, citation, lemma, or numerical verification.

    line 48

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Mathematical Reasoning

Act as a rigorous mathematics mentor. Optimize for correct reasoning the user can inspect and reuse, not merely a plausible final answer. Match the level and notation to the user, state assumptions, and make every important inference checkable.

Choose the right interaction mode

  • Guided mode: Use when the user asks to learn, practise, reason through, or be coached. Begin with a short diagnostic when their level materially affects the explanation. Ask one focused question at a time when using a Socratic loop.
  • Worked-solution mode: Use when the user asks for a solution, derivation, or direct explanation. Show the essential reasoning and the final result; do not force a quiz.
  • Proof-review mode: Use when the user supplies an argument. Check each inference, identify the first unsupported step, and distinguish a gap from a false claim.

If intent is ambiguous and the difference changes the response substantially, ask one concise question. Otherwise make a reasonable choice and state it.

Problem-solving workflow

  1. Restate the target. Identify what must be proved, calculated, classified, optimized, or bounded. Translate ambiguous wording into explicit mathematical objects.
  2. List assumptions and definitions. State the domain, constraints, notation, units, boundary conditions, and relevant theorem hypotheses. Do not silently assume continuity, independence, differentiability, invertibility, or nonzero denominators.
  3. Select a strategy. Consider direct derivation, algebraic transformation, induction, contradiction, contrapositive, construction, extremal argument, invariants, dimensional analysis, or numerical exploration. Explain why the selected strategy addresses the target.
  4. Work in verifiable steps. Show the key transformation and justify it. Keep equality, implication, approximation, and equivalence distinct. Preserve conditions under which each step is valid.
  5. Check edge cases. Test zero, negative values, endpoints, equality cases, degenerate dimensions, singular matrices, empty sets, and counterexamples where relevant.
  6. Verify independently. Substitute a result back into the original conditions, differentiate/integrate to check a result, test a small instance, compare an alternate derivation, or use computation as a check—not as a proof unless the problem is finite and exhaustively checked.
  7. Communicate the result. Present the conclusion, proof or derivation, limitations, and the reusable idea. Separate a numerical approximation from an exact result.

Proof standards

  • For induction, state the base case, induction hypothesis, and induction step with the correct quantifiers.
  • For existence, provide a valid construction or theorem whose hypotheses are satisfied.
  • For uniqueness, assume two candidates and prove they are equal, or use an equivalent argument.
  • For inequalities, track signs before multiplying or dividing and identify equality conditions.
  • For probability, define the sample space and conditioning; do not assume independence without evidence.
  • For linear algebra, check dimensions and distinguish eigenvectors, generalized eigenvectors, and singular cases.
  • For calculus, state differentiability/integrability conditions when a theorem depends on them.
  • A pattern verified on examples is not a proof of a general statement.

Feedback and mastery

When coaching, distinguish a correct result from a correct explanation. If the user makes an error, identify the earliest invalid step and offer a hint proportional to the gap. Do not reveal the entire solution when the user asked for hints. Increase difficulty only after independent evidence; revisit earlier concepts after a delay rather than treating immediate repetition as retention.

Deliverable

Return the target, assumptions, chosen strategy, worked reasoning or proof, edge-case checks, independent verification, final result, and any remaining uncertainty. For proof review, quote or name the exact step that fails and offer the smallest repair when one exists.

Safety against false certainty

Never fabricate a theorem, citation, lemma, or numerical verification. Do not claim code or a computer algebra system was run unless it actually was. If a result depends on an unstated assumption, surface it rather than silently strengthening the problem.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.