ICML Workflow
Use this as the router for the ICML depth pack. It chooses the next skill and keeps policy,
evidence, artifact, and public-record risks visible.
ICML is a conference, not a journal: it has no standing editor-in-chief and no article-processing
charge. The rotating leadership is the per-edition General Chair and Program Chairs (2026 General
Chair: Tong Zhang; Program Chairs: Dudik, Jaggi, Agarwal, Li, verified 2026-06-22), and the cost
model is registration fees, not APCs — PMLR proceedings are open-access with no author fee.
Conference organizers rotate yearly, so re-check the current CFP and committees page rather than
carrying a name forward.
Stage map
- Fit and track:
icml-topic-selection.
- Drafting:
icml-writing-style, icml-related-work, icml-experiments.
- Artifact and supplement:
icml-artifact-evaluation, icml-reproducibility, icml-supplementary.
- Pre-submission audit:
icml-submission.
- Review diagnosis:
icml-review-process.
- Rebuttal and discussion:
icml-author-response.
- Accepted-paper work:
icml-camera-ready.
Working principles
- Start with the current ICML official pages, not last year's memory.
- Maintain a table mapping claims to main-paper evidence, appendix evidence, code/data, and
limitations.
- Keep anonymous review artifacts and public release artifacts as separate states.
- Treat reciprocal reviewing, reviewer LLM policy, impact statements, lay summaries, and public
original-submission release as first-class workflow items.
- Decide early whether a paper is really an ICML main-track paper or better suited to a position
paper, workshop, NeurIPS, ICLR, AISTATS, TMLR, JMLR, or a domain venue.
Deadline cadence (verify against the current CFP)
ICML runs an abstract deadline before the full-paper deadline, and the 2026 cycle freezes the author
list at the abstract stage. Treat the cadence as: abstract and author list locked, then full PDF and
supplement, then reviews, then the response window, then one discussion round, then decisions, then
camera-ready and the PMLR-bound proceedings. Never hard-code the exact dates here; reopen the current
year's CFP because they move annually.
Stage-to-risk router
| Stage |
Dominant ICML risk |
First skill to open |
| Fit |
Wrong track (main vs Position Papers) |
icml-topic-selection |
| Drafting |
Claim outruns evidence in 8 pages |
icml-writing-style, icml-experiments |
| Pre-submission |
Anonymity or page-limit desk reject |
icml-submission |
| Response |
One-round rebuttal misallocated |
icml-author-response |
| Accepted |
Public-record exposure on OpenReview |
icml-camera-ready |
Worked routing pass
A team has an adaptive-optimizer paper with a convergence proof and ImageNet-scale benchmarks. The
router sends them first to topic selection (confirm main-track, not Position Papers), then to
experiments and writing-style for the tuned-baseline and compute-disclosure gaps, then to submission
for the OpenReview and format-checker audit. Each handoff records the official page to reopen so the
plan never relies on last year's remembered rules. After decisions, an accepted paper flows to
camera-ready with its public-record state in mind, while a rejected one loops back to topic selection
or reroute, carrying the reviewer-quotable issue into the next venue choice.
Output format
[Current stage] fit / drafting / pre-submission / review / response / camera-ready / reroute
[Next skill] <ICML skill>
[Official source to open] <CFP/Author Instructions/FAQ/OpenReview/LLM policy>
[Blocking gap] <policy/evidence/artifact/writing/public-record>
[Next action] <concrete task>
Source: brycewang-stanford/Awesome-Journal-Skills → ICML-Skills/skills/icml-workflow/SKILL.md
1---2name: icml-workflow3description: Use when planning or sequencing an ICML manuscript workflow end to end - from topic selection and drafting through OpenReview submission, double-blind review, author response, camera-ready, PMLR publication, public original-submission release, and rerouting decisions. Use when you need the next ICML skill, the official page to reopen, or the blocking gap for the current stage.4---5
6
7# ICML Workflow
8
9Use this as the router for the ICML depth pack. It chooses the next skill and keeps policy,
10evidence, artifact, and public-record risks visible.
11
12ICML is a conference, not a journal: it has no standing editor-in-chief and no article-processing
13charge. The rotating leadership is the per-edition General Chair and Program Chairs (2026 General
14Chair: Tong Zhang; Program Chairs: Dudik, Jaggi, Agarwal, Li, verified 2026-06-22), and the cost
15model is registration fees, not APCs — PMLR proceedings are open-access with no author fee.
16Conference organizers rotate yearly, so re-check the current CFP and committees page rather than
17carrying a name forward.
18
19## Stage map
20
211. Fit and track: `icml-topic-selection`.
222. Drafting: `icml-writing-style`, `icml-related-work`, `icml-experiments`.
233. Artifact and supplement: `icml-artifact-evaluation`, `icml-reproducibility`, `icml-supplementary`.
244. Pre-submission audit: `icml-submission`.
255. Review diagnosis: `icml-review-process`.
266. Rebuttal and discussion: `icml-author-response`.
277. Accepted-paper work: `icml-camera-ready`.
28
29## Working principles
30
31- Start with the current ICML official pages, not last year's memory.
32- Maintain a table mapping claims to main-paper evidence, appendix evidence, code/data, and
33 limitations.
34- Keep anonymous review artifacts and public release artifacts as separate states.
35- Treat reciprocal reviewing, reviewer LLM policy, impact statements, lay summaries, and public
36 original-submission release as first-class workflow items.
37- Decide early whether a paper is really an ICML main-track paper or better suited to a position
38 paper, workshop, NeurIPS, ICLR, AISTATS, TMLR, JMLR, or a domain venue.
39
40## Deadline cadence (verify against the current CFP)
41
42ICML runs an abstract deadline before the full-paper deadline, and the 2026 cycle freezes the author
43list at the abstract stage. Treat the cadence as: abstract and author list locked, then full PDF and
44supplement, then reviews, then the response window, then one discussion round, then decisions, then
45camera-ready and the PMLR-bound proceedings. Never hard-code the exact dates here; reopen the current
46year's CFP because they move annually.
47
48## Stage-to-risk router
49
50| Stage | Dominant ICML risk | First skill to open |
51| --- | --- | --- |
52| Fit | Wrong track (main vs Position Papers) | `icml-topic-selection` |
53| Drafting | Claim outruns evidence in 8 pages | `icml-writing-style`, `icml-experiments` |
54| Pre-submission | Anonymity or page-limit desk reject | `icml-submission` |
55| Response | One-round rebuttal misallocated | `icml-author-response` |
56| Accepted | Public-record exposure on OpenReview | `icml-camera-ready` |
57
58## Worked routing pass
59
60A team has an adaptive-optimizer paper with a convergence proof and ImageNet-scale benchmarks. The
61router sends them first to topic selection (confirm main-track, not Position Papers), then to
62experiments and writing-style for the tuned-baseline and compute-disclosure gaps, then to submission
63for the OpenReview and format-checker audit. Each handoff records the official page to reopen so the
64plan never relies on last year's remembered rules. After decisions, an accepted paper flows to
65camera-ready with its public-record state in mind, while a rejected one loops back to topic selection
66or reroute, carrying the reviewer-quotable issue into the next venue choice.
67
68## Output format
69
70```text
71[Current stage] fit / drafting / pre-submission / review / response / camera-ready / reroute
72[Next skill] <ICML skill>
73[Official source to open] <CFP/Author Instructions/FAQ/OpenReview/LLM policy>
74[Blocking gap] <policy/evidence/artifact/writing/public-record>
75[Next action] <concrete task>
76```
77
78---
79
80**Source:** [`brycewang-stanford/Awesome-Journal-Skills`](https://github.com/brycewang-stanford/Awesome-Journal-Skills) → `ICML-Skills/skills/icml-workflow/SKILL.md`
thedixitjain