Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when converting an accepted CoRL paper into its final form — the 9-page camera-ready main text with the extra page for review feedback, the final-mode corl LaTeX template, appendix merged into one PDF, de-anonymization, external hosting for videos because PMLR accepts none, and the October deadline into the PMLR volume.
.claude/skills/brycewang-stanford-corl-camera-ready/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 70% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 54% | 0% |
Acceptance converts the paper from a review object into a permanent PMLR record — alongside, at CoRL, the public release of its reviews and rebuttal. The camera-ready pass is therefore three jobs: absorb what the review process demanded, de-anonymize cleanly, and stand up the external artifact surface that the proceedings pipeline cannot host. The 2026 rules below were read from corl.org author instructions on 2026-07-08; the deadline was October 12, 2026, 23:59 AoE.
| Dimension | Submission | Camera-ready | |---|---|---| | Main-text pages | 8 (Limitations included) | 9 — the added page exists to accommodate review feedback | | Template mode | anonymous review mode | \usepackage[final]{corl_2026} | | Identity | double-anonymous | authors, affiliations, acknowledgments restored | | Appendix | same PDF, optional | included at the end of the camera-ready PDF, not a separate file | | Video | supplementary upload (≤ 250 MB) | not accepted by PMLR — host externally, link from the main text / project page | | Audience | reviewers + AC | permanent open-access record, reviews attached |
The extra page is explicitly tied to review feedback, and your reviews will be published next to the paper — so the highest-value use is closing the loop visibly:
("we will add the 5-seed table," "we will scope the claim to tasks A–B").
able to diff your promises against the PDF.
into §5; an expanded Limitations section; the evaluation-protocol detail reviewers had to ask for.
were never reviewed do not belong in the archival version.
personnel thanks (acknowledgments remain outside the page count).
work 12]" where that is the honest phrasing.
after those URLs exist and are populated (corl-artifact-evaluation).
notes, \anonymize{} macros, and the anonymous-repo README inside your ZIPs.
Because PMLR will not carry video — the field's core evidence medium — every accepted CoRL paper effectively ships a project page. Build it to archival standards before the PDF freezes:
uncut per-task evaluation clips that back your tables.
corl-artifact-evaluation; DOI-stamped archives for anything citable.
takes your evidence with it; prefer organization-level or archival hosting.
bash# Final-PDF gate, run on the actual upload candidate grep -c 'usepackage\[final\]{corl_2026}' main.tex # exactly 1 pdftotext final.pdf - | head -40 # author block present & correct pdftotext final.pdf - | grep -nEi 'anonym|under review|paper #' # expect no hits pdfinfo final.pdf | grep -i pages # 9pp body + refs + appendix merged pdftotext final.pdf - | grep -oE 'https?://[^ ]+' | sort -u # every URL: live, permanent, deliberate
what you submit to the camera-ready pipeline; a typo here propagates to citation indexes for good. Reconcile against every coauthor's preferred name form in writing before upload.
(see corl-related-work on the year-offset trap) — your own paper may be cited with either year; you cannot control it, but your project page can display the canonical BibTeX.
(待核实); recent anchors are v270 (2024) and v305 (2025).
November 9). Registration and presentation obligations for accepted papers — who must register, in-person requirements, poster/talk formats — were not verified in this pack (待核实): read the acceptance email and the year-site's attendance pages as controlling.
will be delivered to an audience that can read your reviews.
text[ ] All rebuttal commitments landed and locatable in the PDF [ ] 9-page body; Limitations expanded, not amputated [ ] [final] template mode; identity restored; scaffolding scrubbed [ ] Appendix merged at end of the single PDF [ ] Project page live: video, code, data, canonical BibTeX [ ] Every printed URL durable and populated [ ] Coauthor sign-off on names/affiliations/order, in writing [ ] Uploaded before Oct 12 23:59 AoE (2026); confirmation archived [ ] Registration/attendance obligations confirmed from the acceptance email [待核实 in this pack]
Camera-ready mechanics are re-issued with each cycle's acceptance emails and at https://www.corl.org/contributions/instruction-for-authors — that text, not this snapshot, controls.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 27,409 | 25,601 | -7% | 1 | 1 | 0% | 3,532 | 4,457 | +26% | 0 | 0 | — |
case-02 | fail→pass | 20,643 | 21,917 | +6% | 1 | 1 | 0% | 4,137 | 4,380 | +6% | 0 | 0 | — |
case-03 | fail→pass | 22,963 | 24,990 | +9% | 1 | 1 | 0% | 3,743 | 4,773 | +28% | 0 | 0 | — |
case-04 | fail→pass | 15,736 | 13,596 | -14% | 1 | 1 | 0% | 1,648 | 2,795 | +70% | 0 | 0 | — |
case-05 | fail→fail | 20,905 | 25,636 | +23% | 1 | 1 | 0% | 2,517 | 4,488 | +78% | 0 | 0 | — |
case-06 | fail→pass | 17,622 | 14,081 | -20% | 1 | 1 | 0% | 1,859 | 2,868 | +54% | 0 | 0 | — |
case-07 | pass→pass | 13,169 | 10,626 | -19% | 1 | 1 | 0% | 1,250 | 2,328 | +86% | 0 | 0 | — |
case-08 | pass→pass | 19,137 | 14,347 | -25% | 1 | 1 | 0% | 2,039 | 2,916 | +43% | 0 | 0 | — |
case-09 | pass→pass | 14,513 | 11,336 | -22% | 1 | 1 | 0% | 1,541 | 2,497 | +62% | 0 | 0 | — |
case-10 | fail→fail | 18,929 | 20,204 | +7% | 1 | 1 | 0% | 3,051 | 3,891 | +28% | 0 | 0 | — |
case-11 | pass→pass | 14,495 | 12,419 | -14% | 1 | 1 | 0% | 1,590 | 2,620 | +65% | 0 | 0 | — |
case-12 | pass→pass | 16,076 | 13,521 | -16% | 1 | 1 | 0% | 1,717 | 2,803 | +63% | 0 | 0 | — |
case-13 | fail→pass | 11,365 | 9,794 | -14% | 1 | 1 | 0% | 1,707 | 2,313 | +36% | 0 | 0 | — |
case-14 | pass→pass | 18,774 | 7,710 | -59% | 1 | 1 | 0% | 1,682 | 2,634 | +57% | 0 | 0 | — |
case-15 | fail→pass | 22,474 | 6,653 | -70% | 1 | 1 | 0% | 2,083 | 2,301 | +10% | 0 | 0 | — |
case-16 | fail→pass | 17,478 | 10,897 | -38% | 1 | 1 | 0% | 1,899 | 2,477 | +30% | 0 | 0 | — |
case-17 | fail→pass | 15,515 | 9,368 | -40% | 1 | 1 | 0% | 1,660 | 2,070 | +25% | 0 | 0 | — |
case-18 | fail→pass | 13,948 | 7,718 | -45% | 1 | 1 | 0% | 1,457 | 1,833 | +26% | 0 | 0 | — |
case-19 | fail→pass | 20,947 | 17,884 | -15% | 1 | 1 | 0% | 2,038 | 3,330 | +63% | 0 | 0 | — |
case-20 | fail→fail | 19,448 | 22,434 | +15% | 1 | 1 | 0% | 2,936 | 3,912 | +33% | 0 | 0 | — |
case-21 | pass→pass | 13,417 | 8,348 | -38% | 1 | 1 | 0% | 1,183 | 2,006 | +70% | 0 | 0 | — |
case-22 | pass→pass | 16,541 | 11,511 | -30% | 1 | 1 | 0% | 1,700 | 2,394 | +41% | 0 | 0 | — |
case-23 | pass→pass | 19,721 | 13,106 | -34% | 1 | 1 | 0% | 1,811 | 2,613 | +44% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +48 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.