proof-readability — Claude Code skill: readable math proofs
proof-readability edits a math proof that already passed verification so a reader can follow it in one go: reordered steps, added lead-ins, fixed references. The claims stay as they are.
no license ★ 240 updated 2026-09-18
npx skills add franklee16/academic-research-skills --skill proof-readability -a claude-codeWhat proof-readability does
It runs late in the chain. The proof exists and has already been accepted, and this skill goes over the write-up: move steps around, add a lead-in, repair references and typos. Nothing in the argument changes. Claims are not strengthened, weakened or quietly simplified.
The work is split into six layers, and the first two take the whole proof at once. One covers layout: the statement and a sketch sit in the main text, the formal argument moves to an appendix, each lemma gets a sentence about what it says and why it is there, and an assembled proof ends with an explicit wrap-up. The other covers navigation: the opening line names the method, a long proof is broken into named steps, and an announced count of cases has to match what follows. Layers three through six are local edits.
A real gap does not get patched. When a step cannot be derived from the stated assumptions, editing stops and that step goes back to the prover, because a silent fix would undermine the verification that already happened. Inside a co-math project the skill is wired into the pipeline: project-coordinator opens a W{NNN}-readability-{slug} workstream after approval, prover carries it out in readability mode, and paper-reviewer checks only that the content survived. Scope is the approved proofs named in instructions.md.
What it helps with
- turn a verified proof into something a reader can follow in one sitting
- merge two near-identical derivations into one lemma used twice
- give a long proof named steps and a closing recap
- check that the case count announced at the start matches the body
- restate a lemma so it reads on its own outside the surrounding prose
Example prompts for the agent
- Polish the proof of Theorem 2 — presentation only, don't touch the math.
- Run a readability pass over the proofs in Appendix C.
- Rewrite Lemma 4 so it can be read on its own.
How to install proof-readability
1Skills CLI (Vercel)
npx skills add franklee16/academic-research-skills --skill proof-readability -a claude-code2Manually: put the files in place
~/.claude/skills/proof-readability/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/proof-readability/
1Skills CLI (Vercel)
npx skills add franklee16/academic-research-skills --skill proof-readability -a codex2Manually: put the files in place
.agents/skills/proof-readability/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add franklee16/academic-research-skills --skill proof-readability -a cursor2Manually: put the files in place
.cursor/skills/proof-readability/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install franklee16/academic-research-skills theory-skills/theorist-toolbox-master/skills/proof-readability2Manually: put the files in place
.github/skills/proof-readability/SKILL.md1Skills CLI (Vercel)
npx skills add franklee16/academic-research-skills --skill proof-readability -a gemini-cli2Platform CLI
gemini skills install https://github.com/franklee16/academic-research-skills --path theory-skills/theorist-toolbox-master/skills/proof-readabilityManually: put the files in place
.devin/skills/proof-readability/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about proof-readability
How do I install proof-readability in Claude Code?
Run npx skills add franklee16/academic-research-skills --skill proof-readability -a claude-code. The skill lands at ~/.claude/skills/proof-readability/SKILL.md.
What does proof-readability do?
It reworks how a checked proof reads: step order, lead-ins, references, notation. The claims and the arguments behind them stay put.
What happens if the pass turns up a mistake?
Editing that proof stops, the suspect step is described, and it goes back to the prover. Patching the math quietly would defeat the verification already done.