agentsclimarketplace

Popl supplementary

Skill brycewang-stanford/Awesome-Journal-Skills/POPL-Skills/skills/popl-supplementary

Use when splitting a POPL paper across the 25-page body, the proof appendix, and anonymous supplementary material — deciding which proofs and auxiliary judgments leave the text, packaging proof scripts without identity leaks under full double-blind, and keeping every artifact self-consistent with the submission PDF.From its SKILL.md

Install
npx -y skills add brycewang-stanford/Awesome-Journal-Skills --skill popl-supplementary

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

3.5 KB, 774 tokens by cl100k_base, as published. Nobody here has run it

POPL Supplementary Material

A POPL submission is really three documents: 25 pages of text (bibliography excluded) that must carry the whole argument, an appendix of full proofs and auxiliary definitions, and — usually — an anonymized proof development or prototype. Reviewers are typically expected to judge the paper from the body alone, so the split is an argumentation decision, not a storage decision. Format and anonymity rules per the POPL 2027 call, read 2026-07-08; confirm the current cycle's supplement wording before uploading.

What lives where

ContentBody (25 pp)AppendixAnonymous artifact
Main definitions, typing/semantics rules actually discussedyesmirrored in fullformalized
Main theorem statements + proof sketchesyesfull proofschecked statements
Auxiliary lemmas, weakening/substitution boilerplatenoyesyes
Full figure of every judgment (all rules)representative rules onlycomplete figuresource of truth
Extended examples, failed design alternativesone motivating exampleyestest files
Proof scripts, build instructionsnonoyes, with README

Two disciplines make the split safe:

  • The body must stand alone. A reviewer who never opens the appendix should still believe the theorem plausible from the sketch: state the invariant, the hard case, and why it goes through. "Proof in appendix" after an unexplained claim reads as a gap.
  • Sketch and proof must not disagree. The classic incident: the body's sketch describes induction on typing derivations while the appendix inducts on evaluation steps because the proof changed. Reviewers who notice stop trusting both.

Anonymizing a proof development

Full double-blind covers everything you upload. Proof repositories are leaky: _CoqProject paths with usernames, lakefile package names matching a public GitHub project, author headers auto-inserted by editors, and .git directories with full commit history. Build the archive from an export, never from a working tree:

git archive --format=tar.gz -o /tmp/supp.tar.gz HEAD          # no .git, no untracked junk
tar tzf /tmp/supp.tar.gz | grep -iE '\.git|/home/|users/|TODO|AUTHORS' && echo LEAK
grep -rInE '(Copyright|Author|@[a-z]+\.(edu|org|fr|de))' \
  --include='*.v' --include='*.lean' --include='*.agda' extracted/ | head

Also rename the development if its public name is googleable to your group, and strip institutional CI configuration.

Version-lock the trio

  • Tag the exact commit that generated the submitted PDF, appendix, and archive; the author response will need to quote them line-precisely months later.
  • If the appendix is a separate PDF, give it the same section numbering scheme as the body so "App. C.2" resolves unambiguously.
  • Late theorem renumbering must propagate to the correspondence table and README — do it with a script, not by hand, in deadline week.

Output format

[Split audit] <claims whose evidence sits only outside the body>
[Sketch-proof consistency] <mismatches found>
[Anonymity scan] clean / leaks listed
[Version lock] <tag/commit for PDF + appendix + archive>
[Upload set] <files, sizes, formats for HotCRP>

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.