agentleFS
LeonardSEOLeonardSEO/ambitious-math-research/skills/proof-audit/SKILL.md

proof-audit

Skill0 starsChanged 33 days agoMITView on GitHub

Audit a supplied mathematical proof or candidate lemma for decisive logical gaps, circular assumptions, quantifier errors, and repairable steps. Use for focused proof checking; originality review and general proof search are separate tasks.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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:

    Use a fresh-context agent for a complex independent reconstruction when delegation is available and permitted.

    line 18
  2. Distinctive:

    Do not claim the entire manuscript has been checked when only a bridge lemma was inspected.

    line 28
  3. Distinctive:

    Use the audit template when delegating or drafting an audit prompt.

    line 30

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Proof Audit

Assess the actual argument against the exact theorem. Read enough of the proof and its dependencies to reconstruct the decisive step. Preserve the difference between a proof defect and a publication or verification boundary.

Focus the challenge

  • Recover the statement, definitions, hypotheses, quantifiers, and conclusion. Check that reductions preserve them and do not silently restrict the problem.
  • Identify the load-bearing inference. Derive it independently or supply a concrete obstruction. Check equivalent-conjecture assumptions, sign and normalization errors, missing domains, boundary cases, multiplicities, and the interchange of limits, sums, or derivatives when relevant.
  • For uniform estimates, track every constant and parameter. For operators, check the actual domain and whether the claimed selfadjointness is proved. For a finite computation, check error bounds and what is certified beyond sample points.
  • Give each substantive finding an exact location, the failed implication, and a counterexample or repair obligation. Vague unease, lack of peer review, and lack of Lean formalization are not mathematical findings.
  • Distinguish false statements from arguments that have not established a true statement. A stronger hypothesis or a restricted conclusion can be a useful repair, but must be labeled as a changed theorem.

Use a fresh-context agent for a complex independent reconstruction when delegation is available and permitted. Supply the statement and raw proof, not the preferred conclusion. Do not require different model brands or claim that independent agents are independent mathematical authorities.

Verdict

Report the exact checked range and one of:

  • Proof established in the checked scope: a complete argument has been reconstructed and no substantive gap remains in that range. Say which hypotheses apply and what can now be used.
  • Candidate with a specific gap: identify the inference still requiring proof and its consequences for the claim.
  • Claim refuted as stated: provide the contradiction or counterexample and a possible restricted replacement.

Do not claim the entire manuscript has been checked when only a bridge lemma was inspected. A passing compiler, checksum, wording test, or agent vote is not proof verification. An accepted lemma can be used immediately; reopen it only for a concrete new reason.

Use the audit template when delegating or drafting an audit prompt. If repairs are not requested, return findings without changing the manuscript.

For a supplied Lean artifact or a requested formal-verification claim, use lean-verification to compare the actual target and its transitive dependencies. Lack of formalization alone remains distinct from a substantive proof defect.

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.