paperlint 2.0.0

package/skills/submit-paper/SKILL.md

@@ -0,0 +1,182 @@
+---
+name: submit-paper
+description: End-to-end playbook for submitting a peer-reviewed paper to a double-blind venue (HotCRP workshops/conferences) — from a hardened draft to "ready for review." Covers building and anonymously hosting a reproduction artifact, HotCRP profile + form mechanics, double-blind hygiene, and the authorship/credential angles for an independent researcher. Use when the paper is drafted and reviewed and it's time to actually submit. Compose with pc-panel-review / paper-adversarial-review (harden first), and with the venue data card (references/venues/<venue>.md) for venue specifics.
+allowed-tools: [Read, Write, Edit, Grep, Glob, Bash, WebSearch, WebFetch, Skill]
+---
+
+<!-- vigiles:sha256:6befc6c0f3e910bd compiled from skills/submit-paper/SKILL.md.spec.ts -->
+
+# submit-paper — get a reviewed paper from "done" to "ready for review"
+
+## Run me
+
+🔴 FIRST, before any other step:
+
+```
+node .claude/skills/paper-pipeline/scripts/announce.mjs submit-paper <paper-dir>
+```
+
+An advisory pass cannot be observed failing — silence is both its error state and its normal
+state — so starting is an event, and events get written down.
+
+Assumes the paper is already hardened (run `pc-panel-review` — incl. its venue-fit mode — +
+`paper-adversarial-review` first). This skill is the submission mechanics — the part that bit us with
+avoidable friction the first time. Distilled from the AgenticDev 2026 submission.
+
+## Venue specifics — a DATA card, not a skill
+The venue-specific facts (deadline, page limits, tracks, PC-member conflict list, blind model, format
+quirks, whether a supplementary-upload field exists, the extension target) live as **plain reference
+data files** under `references/venues/<venue>.md`:
+- `references/venues/agenticdev.md` — AgenticDev @ ASE (workshop; ACM DL/IEEE Xplore).
+- `references/venues/aisec.md` — AISec @ ACM CCS (security workshop; ACM DL; harder bar).
+
+**The pattern (follow it for every new venue):** when you fetch a venue's exact submission instructions,
+save them as a new **data file** `references/venues/<venue>.md` (short `title:` frontmatter only — NOT a
+skill with `name:`/`description:`). This preserves the "save the exact venue instructions" rule while
+keeping the skill namespace from proliferating one card per venue. The *mechanics* stay here; the venue
+file is only the facts.
+
+## Publisher specifics — one level ABOVE the venue
+Camera-ready mechanics belong to the **publisher**, not the venue: ACM eRights, the submit-vs-final
+preamble swap, the copyright block, CCS 2012 codes and the mandatory Source-files upload are identical
+for every ACM venue. They live in `references/publishers/<publisher>.md`:
+- `references/publishers/acm.md` — ACM (AgenticDev @ ASE, AISec @ CCS, and every future ACM venue).
+
+The venue card keeps only what is genuinely its own: dates, page limits, its DOI/ISBN, its copyright
+line, its blind model. 🔴 **Filing publisher mechanics under one venue is the mistake this split exists
+to prevent** — it was written into `agenticdev-2026/CAMERA-READY-FORM.md` first, where AISec could
+never find it.
+
+## 0. Pre-flight
+- PDF compiles clean (`pdflatex; bibtex; pdflatex; pdflatex`), ≤ page limit, 0 undefined refs.
+- **De-anon scan the PDF text AND every shipped file** before submitting — author-identifying strings
+  must return zero; identifying the *studied* artifact is fine, identifying *you* is a desk-reject.
+  **MECHANICAL GATE (run it, don't eyeball) — greps the PDF text + every artifact file + inside any
+  `*.zip` for the identity deny-list, exit non-zero on a hit:**
+  ```
+  bash .claude/skills/submit-paper/check-deanon.sh <paper-dir> [pdf=paper.pdf]
+  ```
+  Must PASS (exit 0) for a double-blind submission. Deny-list: `.claude/skills/submit-paper/deanon-denylist.txt`
+  (edit as handles/repos change). This is the sibling of `render-paper/check-render.sh` and guards the
+  single highest reject-vector (an artifact README naming real maintainers/repos is invisible to a
+  paper-only read). Only run it on *double-blind* papers — a de-anonymized single-blind PDF is supposed
+  to carry your name and will (correctly) fail. Canonical deny-list rationale → `paper-pipeline/references/anonymization.md`.
+
+## 1. Build a reproduction artifact (empirical papers)
+The single biggest accept-probability lever for a measurement paper — reviewers reward one they can run.
+In short: a stdlib-only script that *recomputes* every headline number from raw data and **exits
+non-zero on drift**; honest badge/scope (don't claim "raw N runs" for per-arm aggregates); a README
+"what reproduces which number" table + LICENSE; scrubbed of `__pycache__` and fingerprints before it
+ships. Canonical build checklist → `paper-pipeline/references/artifact-checklist.md`.
+
+## 2. Host the artifact ANONYMOUSLY (double-blind)
+Many workshop HotCRPs have **no supplementary-upload field** — check first; if none, host externally and
+`\url{}` the link in the paper's Availability.
+- **OSF (osf.io) — best; self-upload, no repo needed.** Create a **Private** project (title = paper
+  title, no name), upload the files/zip, then **Contributors → View-only Links → Add → check
+  "Anonymize" → Create**. Keep the project private; the anonymized view-only link is what reviewers get.
+  OSF has a REST API + `osfclient` (`OSF_TOKEN`) → **automatable**: with a personal access token you can
+  create the project, upload, and mint the link programmatically. See `osf-artifact-upload` for the API calls; whatever uploader you use must read the token
+  from the `OSF_TOKEN` environment variable and never from a file in the repository. Record each
+  minted view-only link in your own notes and REUSE it — a second link for the same paper is a
+  second thing reviewers can be handed, and only one of them is the one you cited.
+- Alternatives: `anonymous.4open.science` (anonymizes a **GitHub repo** — needs a repo, not file upload;
+  use a *private* source repo so the original isn't Google-findable); a throwaway identity-free account.
+  Avoid Google Drive / Zenodo / Figshare for blind review — they leak the owner name.
+- **VERIFY in an incognito window** (logged out): the link must show the files with contributors =
+  "Anonymous" and your name nowhere. Then put the URL in Availability via `\url{}` and rebuild.
+
+## 3. HotCRP profile (do this once, before the form)
+- **Name = your canonical / passport legal name**, identical on every paper you submit. (For
+  a dossier, name consistency across papers/LinkedIn/letters is load-bearing — see §6.)
+- **Affiliation**: `None` (or `Unaffiliated`) if independent.
+- **Country/region = where you actually reside**, not citizenship — this sidesteps ACM/IEEE
+  sanctions/payment friction for authors from sanctioned countries.
+- **ORCID**: create one (free, 2 min) and reuse it on every paper. It's a permanent ID that ties all
+  your work to one identity *regardless of name spelling* — the fix for any name-variant problem.
+- **Collaborators / other affiliations (COI)**: `None` if you have no recent coauthors/advisors.
+
+## 4. The submission form
+- **Title**: exactly as it should appear in the proceedings (ACM DL) — plain text, no markup.
+- **Submission**: upload **PDF only**. NEVER upload a source zip of the paper directory — sibling files
+  (drafts, READMEs, internal notes) leak identity. Only a scrubbed bundle is safe.
+- **Abstract**: HotCRP makes you paste it into its own box even though it's in the PDF (it indexes it
+  separately). Paste plain text; convert LaTeX math to unicode (`~0.6%`, `≈`).
+- **Authors (anonymous)**: enter your **real name + email** here — this metadata is hidden from
+  reviewers; the *PDF* is what's anonymized. That's how double-blind works; it's correct, not a leak.
+- **ACM corresponding author**: default (first author).
+- **PC conflicts**: check only genuine COIs (past advisor/student, same affiliation, ≤2-yr coauthor).
+  An independent researcher with no ties usually checks **none** — that's honest.
+
+## 5. Submit
+- **Save and submit**, then **mark it "ready for review."** A saved draft alone is NOT evaluated — this
+  is the step people miss.
+- You can edit until the deadline: use **Replace** on the Submission field to swap the PDF for a late
+  fix (e.g. adding the artifact link after you host it). Submit early, upgrade before the deadline.
+
+## 6. Independent-researcher / dossier notes
+- **Name consistency** is the quiet risk: publications, LinkedIn, passport, letters must resolve to one
+  person. ORCID is the tool; pick one surname and use it everywhere from paper #1.
+- Keep the **organizer review-offer email** + any PC/reviewer confirmation — that's the *judging*
+  criterion evidence (separate from authorship).
+- **Space submissions out** — don't cluster all papers in the month before an external
+  filing — the FILING-SPACING rule `plan-paper-timeline` applies. A burst right before a filing reads
+  as manufactured rather than sustained, and that is an observed, documented refusal ground.
+- After acceptance: plan the **camera-ready de-anonymization** and an **extension to a higher-prestige
+  venue** (workshop → conference, ≥30% new) as a second, stronger publication.
+
+## 7. Freeze the version you actually uploaded
+
+🔴 **The moment the portal accepts the file, download it back and commit it.** The rendered PDF is the
+only artifact git cannot reconstruct — it was built outside the repo and lived on the portal.
+
+```
+<paper-dir>/versions/<ISO-date>-submitted.pdf
+```
+
+and in `PIPELINE-STATUS.md`, on the Submit row, the source that produced it: the literal word
+**`commit <hash>`**. Stages are `submitted` / `camera-ready` / `arxiv`; each one you *declare* in
+PIPELINE-STATUS needs its own frozen file.
+
+**This is enforced, not advised** — `checkFrozenVersions()` in `.claude/hooks/paper-lint.mjs` reads the
+declared stages and reports a paper that declared one without freezing it. The convention itself is
+documented per-paper in `versions/README.md`.
+
+**Why it is a step and not hygiene:** `agenticdev-2026` kept a folder for exactly this for a month and
+the file inside was the wrong build (340 952 B vs the real 352 357 B). It surfaced only when a reviewer
+cited `L. 166` and there was nothing to resolve the line against. A folder without a checker is a
+convention with no reader.
+
+## 8. Record the verdict
+
+🔴 LAST step, once the deliverable exists:
+
+```
+node .claude/skills/paper-pipeline/scripts/ledger.mjs record submit-paper <paper-dir> FINDING <count> <report-path>
+node .claude/skills/paper-pipeline/scripts/ledger.mjs record submit-paper <paper-dir> ABSTAINED <reason> "<one line>"
+```
+
+**FINDING** — `check-deanon.sh` exited non-zero, `access` is not green, the PDF is over the limit, or
+the ready-for-review click never happened; `<count>` is the number of blocking items and
+`<report-path>` names them. Always `--blocking` — every item here stops the submission.
+**ABSTAINED** — `no-witness`: uploaded **and** marked ready for review, with the venue card or the
+submission id in the note. `blocked`: the portal account is not through moderation yet.
+
+🔴 **A saved draft must never be recorded as done — and it now cannot be.** A saved-but-not-submitted
+paper is the one failure this skill's provenance names by hand, and it was invisible to every other
+check in the pipeline. There is no constructor for "submitted"; there is only the absence of blocking
+items, plus a submission id in the note that a human can go and look up. The id IS the witness, and
+recording `no-witness` without one is the same lie in a new vocabulary.
+
+## Compose with
+- Harden first: `pc-panel-review` (incl. its venue-fit mode), `paper-adversarial-review`, `render-paper`.
+- Publisher mechanics (camera-ready): `references/publishers/<publisher>.md`.
+- Venue specifics: the matching data card `references/venues/<venue>.md` (e.g.
+  `references/venues/agenticdev.md`) for deadlines, page limits, PC-member conflict lists, whether
+  supplementary upload exists, and the extension target.
+
+## Provenance
+Written from the AgenticDev 2026 @ ASE submission (2026-07-13): PC-panel-reviewed, self-checking
+artifact hosted on an OSF anonymized view-only link (no supplementary field at the venue), submitted #20
+ready-for-review. The friction points here — no supplementary slot, 4open needing a repo, the abstract
+re-paste, "ready for review" being a separate step, the PDF-only rule — are the ones that cost time.

package/skills/submit-paper/SKILL.md.spec.ts

@@ -0,0 +1,199 @@
+// Compiled to SKILL.md by `vigiles compile`. Edit THIS file, never the markdown.
+//
+// Adopted 2026-08-17 (batch 3). Body carried over VERBATIM so the compiled diff shows
+// only what the compiler adds. No `disallowedTools` fence yet — the field landed on
+// `SkillSpec` in vigiles branch `claude/skill-disallowed-tools` and is not in a release
+// this repo installs, so writing one here would not compile.
+import { experimental_skill } from "vigiles/spec";
+
+export default experimental_skill({
+  name: "submit-paper",
+  description:
+    'End-to-end playbook for submitting a peer-reviewed paper to a double-blind venue (HotCRP workshops/conferences) — from a hardened draft to "ready for review." Covers building and anonymously hosting a reproduction artifact, HotCRP profile + form mechanics, double-blind hygiene, and the authorship/credential angles for an independent researcher. Use when the paper is drafted and reviewed and it\'s time to actually submit. Compose with pc-panel-review / paper-adversarial-review (harden first), and with the venue data card (references/venues/<venue>.md) for venue specifics.',
+  tools: [
+    "Read",
+    "Write",
+    "Edit",
+    "Grep",
+    "Glob",
+    "Bash",
+    "WebSearch",
+    "WebFetch",
+    "Skill",
+  ],
+  body: `
+# submit-paper — get a reviewed paper from "done" to "ready for review"
+
+## Run me
+
+🔴 FIRST, before any other step:
+
+\`\`\`
+node .claude/skills/paper-pipeline/scripts/announce.mjs submit-paper <paper-dir>
+\`\`\`
+
+An advisory pass cannot be observed failing — silence is both its error state and its normal
+state — so starting is an event, and events get written down.
+
+Assumes the paper is already hardened (run \`pc-panel-review\` — incl. its venue-fit mode — +
+\`paper-adversarial-review\` first). This skill is the submission mechanics — the part that bit us with
+avoidable friction the first time. Distilled from the AgenticDev 2026 submission.
+
+## Venue specifics — a DATA card, not a skill
+The venue-specific facts (deadline, page limits, tracks, PC-member conflict list, blind model, format
+quirks, whether a supplementary-upload field exists, the extension target) live as **plain reference
+data files** under \`references/venues/<venue>.md\`:
+- \`references/venues/agenticdev.md\` — AgenticDev @ ASE (workshop; ACM DL/IEEE Xplore).
+- \`references/venues/aisec.md\` — AISec @ ACM CCS (security workshop; ACM DL; harder bar).
+
+**The pattern (follow it for every new venue):** when you fetch a venue's exact submission instructions,
+save them as a new **data file** \`references/venues/<venue>.md\` (short \`title:\` frontmatter only — NOT a
+skill with \`name:\`/\`description:\`). This preserves the "save the exact venue instructions" rule while
+keeping the skill namespace from proliferating one card per venue. The *mechanics* stay here; the venue
+file is only the facts.
+
+## Publisher specifics — one level ABOVE the venue
+Camera-ready mechanics belong to the **publisher**, not the venue: ACM eRights, the submit-vs-final
+preamble swap, the copyright block, CCS 2012 codes and the mandatory Source-files upload are identical
+for every ACM venue. They live in \`references/publishers/<publisher>.md\`:
+- \`references/publishers/acm.md\` — ACM (AgenticDev @ ASE, AISec @ CCS, and every future ACM venue).
+
+The venue card keeps only what is genuinely its own: dates, page limits, its DOI/ISBN, its copyright
+line, its blind model. 🔴 **Filing publisher mechanics under one venue is the mistake this split exists
+to prevent** — it was written into \`agenticdev-2026/CAMERA-READY-FORM.md\` first, where AISec could
+never find it.
+
+## 0. Pre-flight
+- PDF compiles clean (\`pdflatex; bibtex; pdflatex; pdflatex\`), ≤ page limit, 0 undefined refs.
+- **De-anon scan the PDF text AND every shipped file** before submitting — author-identifying strings
+  must return zero; identifying the *studied* artifact is fine, identifying *you* is a desk-reject.
+  **MECHANICAL GATE (run it, don't eyeball) — greps the PDF text + every artifact file + inside any
+  \`*.zip\` for the identity deny-list, exit non-zero on a hit:**
+  \`\`\`
+  bash .claude/skills/submit-paper/check-deanon.sh <paper-dir> [pdf=paper.pdf]
+  \`\`\`
+  Must PASS (exit 0) for a double-blind submission. Deny-list: \`.claude/skills/submit-paper/deanon-denylist.txt\`
+  (edit as handles/repos change). This is the sibling of \`render-paper/check-render.sh\` and guards the
+  single highest reject-vector (an artifact README naming real maintainers/repos is invisible to a
+  paper-only read). Only run it on *double-blind* papers — a de-anonymized single-blind PDF is supposed
+  to carry your name and will (correctly) fail. Canonical deny-list rationale → \`paper-pipeline/references/anonymization.md\`.
+
+## 1. Build a reproduction artifact (empirical papers)
+The single biggest accept-probability lever for a measurement paper — reviewers reward one they can run.
+In short: a stdlib-only script that *recomputes* every headline number from raw data and **exits
+non-zero on drift**; honest badge/scope (don't claim "raw N runs" for per-arm aggregates); a README
+"what reproduces which number" table + LICENSE; scrubbed of \`__pycache__\` and fingerprints before it
+ships. Canonical build checklist → \`paper-pipeline/references/artifact-checklist.md\`.
+
+## 2. Host the artifact ANONYMOUSLY (double-blind)
+Many workshop HotCRPs have **no supplementary-upload field** — check first; if none, host externally and
+\`\\url{}\` the link in the paper's Availability.
+- **OSF (osf.io) — best; self-upload, no repo needed.** Create a **Private** project (title = paper
+  title, no name), upload the files/zip, then **Contributors → View-only Links → Add → check
+  "Anonymize" → Create**. Keep the project private; the anonymized view-only link is what reviewers get.
+  OSF has a REST API + \`osfclient\` (\`OSF_TOKEN\`) → **automatable**: with a personal access token you can
+  create the project, upload, and mint the link programmatically. See \`osf-artifact-upload\` for the API calls; whatever uploader you use must read the token
+  from the \`OSF_TOKEN\` environment variable and never from a file in the repository. Record each
+  minted view-only link in your own notes and REUSE it — a second link for the same paper is a
+  second thing reviewers can be handed, and only one of them is the one you cited.
+- Alternatives: \`anonymous.4open.science\` (anonymizes a **GitHub repo** — needs a repo, not file upload;
+  use a *private* source repo so the original isn't Google-findable); a throwaway identity-free account.
+  Avoid Google Drive / Zenodo / Figshare for blind review — they leak the owner name.
+- **VERIFY in an incognito window** (logged out): the link must show the files with contributors =
+  "Anonymous" and your name nowhere. Then put the URL in Availability via \`\\url{}\` and rebuild.
+
+## 3. HotCRP profile (do this once, before the form)
+- **Name = your canonical / passport legal name**, identical on every paper you submit. (For
+  a dossier, name consistency across papers/LinkedIn/letters is load-bearing — see §6.)
+- **Affiliation**: \`None\` (or \`Unaffiliated\`) if independent.
+- **Country/region = where you actually reside**, not citizenship — this sidesteps ACM/IEEE
+  sanctions/payment friction for authors from sanctioned countries.
+- **ORCID**: create one (free, 2 min) and reuse it on every paper. It's a permanent ID that ties all
+  your work to one identity *regardless of name spelling* — the fix for any name-variant problem.
+- **Collaborators / other affiliations (COI)**: \`None\` if you have no recent coauthors/advisors.
+
+## 4. The submission form
+- **Title**: exactly as it should appear in the proceedings (ACM DL) — plain text, no markup.
+- **Submission**: upload **PDF only**. NEVER upload a source zip of the paper directory — sibling files
+  (drafts, READMEs, internal notes) leak identity. Only a scrubbed bundle is safe.
+- **Abstract**: HotCRP makes you paste it into its own box even though it's in the PDF (it indexes it
+  separately). Paste plain text; convert LaTeX math to unicode (\`~0.6%\`, \`≈\`).
+- **Authors (anonymous)**: enter your **real name + email** here — this metadata is hidden from
+  reviewers; the *PDF* is what's anonymized. That's how double-blind works; it's correct, not a leak.
+- **ACM corresponding author**: default (first author).
+- **PC conflicts**: check only genuine COIs (past advisor/student, same affiliation, ≤2-yr coauthor).
+  An independent researcher with no ties usually checks **none** — that's honest.
+
+## 5. Submit
+- **Save and submit**, then **mark it "ready for review."** A saved draft alone is NOT evaluated — this
+  is the step people miss.
+- You can edit until the deadline: use **Replace** on the Submission field to swap the PDF for a late
+  fix (e.g. adding the artifact link after you host it). Submit early, upgrade before the deadline.
+
+## 6. Independent-researcher / dossier notes
+- **Name consistency** is the quiet risk: publications, LinkedIn, passport, letters must resolve to one
+  person. ORCID is the tool; pick one surname and use it everywhere from paper #1.
+- Keep the **organizer review-offer email** + any PC/reviewer confirmation — that's the *judging*
+  criterion evidence (separate from authorship).
+- **Space submissions out** — don't cluster all papers in the month before an external
+  filing — the FILING-SPACING rule \`plan-paper-timeline\` applies. A burst right before a filing reads
+  as manufactured rather than sustained, and that is an observed, documented refusal ground.
+- After acceptance: plan the **camera-ready de-anonymization** and an **extension to a higher-prestige
+  venue** (workshop → conference, ≥30% new) as a second, stronger publication.
+
+## 7. Freeze the version you actually uploaded
+
+🔴 **The moment the portal accepts the file, download it back and commit it.** The rendered PDF is the
+only artifact git cannot reconstruct — it was built outside the repo and lived on the portal.
+
+\`\`\`
+<paper-dir>/versions/<ISO-date>-submitted.pdf
+\`\`\`
+
+and in \`PIPELINE-STATUS.md\`, on the Submit row, the source that produced it: the literal word
+**\`commit <hash>\`**. Stages are \`submitted\` / \`camera-ready\` / \`arxiv\`; each one you *declare* in
+PIPELINE-STATUS needs its own frozen file.
+
+**This is enforced, not advised** — \`checkFrozenVersions()\` in \`.claude/hooks/paper-lint.mjs\` reads the
+declared stages and reports a paper that declared one without freezing it. The convention itself is
+documented per-paper in \`versions/README.md\`.
+
+**Why it is a step and not hygiene:** \`agenticdev-2026\` kept a folder for exactly this for a month and
+the file inside was the wrong build (340 952 B vs the real 352 357 B). It surfaced only when a reviewer
+cited \`L. 166\` and there was nothing to resolve the line against. A folder without a checker is a
+convention with no reader.
+
+## 8. Record the verdict
+
+🔴 LAST step, once the deliverable exists:
+
+\`\`\`
+node .claude/skills/paper-pipeline/scripts/ledger.mjs record submit-paper <paper-dir> FINDING <count> <report-path>
+node .claude/skills/paper-pipeline/scripts/ledger.mjs record submit-paper <paper-dir> ABSTAINED <reason> "<one line>"
+\`\`\`
+
+**FINDING** — \`check-deanon.sh\` exited non-zero, \`access\` is not green, the PDF is over the limit, or
+the ready-for-review click never happened; \`<count>\` is the number of blocking items and
+\`<report-path>\` names them. Always \`--blocking\` — every item here stops the submission.
+**ABSTAINED** — \`no-witness\`: uploaded **and** marked ready for review, with the venue card or the
+submission id in the note. \`blocked\`: the portal account is not through moderation yet.
+
+🔴 **A saved draft must never be recorded as done — and it now cannot be.** A saved-but-not-submitted
+paper is the one failure this skill's provenance names by hand, and it was invisible to every other
+check in the pipeline. There is no constructor for "submitted"; there is only the absence of blocking
+items, plus a submission id in the note that a human can go and look up. The id IS the witness, and
+recording \`no-witness\` without one is the same lie in a new vocabulary.
+
+## Compose with
+- Harden first: \`pc-panel-review\` (incl. its venue-fit mode), \`paper-adversarial-review\`, \`render-paper\`.
+- Publisher mechanics (camera-ready): \`references/publishers/<publisher>.md\`.
+- Venue specifics: the matching data card \`references/venues/<venue>.md\` (e.g.
+  \`references/venues/agenticdev.md\`) for deadlines, page limits, PC-member conflict lists, whether
+  supplementary upload exists, and the extension target.
+
+## Provenance
+Written from the AgenticDev 2026 @ ASE submission (2026-07-13): PC-panel-reviewed, self-checking
+artifact hosted on an OSF anonymized view-only link (no supplementary field at the venue), submitted #20
+ready-for-review. The friction points here — no supplementary slot, 4open needing a repo, the abstract
+re-paste, "ready for review" being a separate step, the PDF-only rule — are the ones that cost time.`,
+});

package/skills/submit-paper/check-deanon.sh

@@ -0,0 +1,149 @@
+#!/usr/bin/env bash
+# check-deanon.sh — mechanical double-blind de-anonymization gate.
+#
+# WHY: for a double-blind submission, an author-identifying string leaking into
+# the PDF *or any shipped artifact file* is the single highest reject-vector, and
+# it's invisible to a paper-only read (e.g. an artifact README that names real
+# maintainers/repos). This greps a deny-list across the compiled PDF's TEXT and
+# every artifact file (including inside artifact.zip) so "anonymized" is a checked
+# fact. Sibling to render-paper/check-render.sh. Spirit: prose isn't policy.
+#
+# Usage:  check-deanon.sh <paper-dir> [pdf=paper.pdf] [denylist=<skill>/deanon-denylist.txt]
+#   It auto-scans, in <paper-dir>: the PDF text, everything under artifact/ (if present),
+#   and the contents of any *.zip (artifact bundles). Pass extra paths after the
+#   denylist arg to scan additional dirs/zips.
+#
+# Exit: 0 = clean (no identifiers found); 1 = leak(s) found; 2 = usage/tool error.
+# NOTE: run this ONLY on double-blind submissions. A single-blind/de-anonymized
+# paper is SUPPOSED to carry the author name — this gate would (correctly) fail it.
+
+set -uo pipefail
+
+DIR="${1:?usage: check-deanon.sh <paper-dir> [pdf=paper.pdf] [denylist=...] [extra-paths...]}"
+PDF="${2:-paper.pdf}"
+SKILL_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+DENY="${3:-$SKILL_DIR/deanon-denylist.txt}"
+shift $(( $# >= 3 ? 3 : $# ))
+EXTRA=("$@")
+
+[ -d "$DIR" ]  || { echo "FAIL: no dir $DIR"; exit 2; }
+[ -f "$DENY" ] || { echo "FAIL: no deny-list $DENY"; exit 2; }
+
+# Build the active pattern (strip comments/blanks).
+PAT="$(grep -vE '^\s*(#|$)' "$DENY" || true)"
+[ -n "$PAT" ] || { echo "FAIL: deny-list $DENY has no patterns"; exit 2; }
+
+TMP="$(mktemp -d)"
+trap 'rm -rf "$TMP"' EXIT
+
+echo "== check-deanon: $DIR (pdf=$PDF, $(printf '%s\n' "$PAT" | wc -l | tr -d ' ') patterns) =="
+
+hits=0
+
+scan() { # scan <label> <file-or-textfile>
+  local label="$1" file="$2"
+  [ -f "$file" ] || return 0
+  local m
+  # -I skips binary files. REQUIRED since 2026-08-07, when the artifact branch stopped
+  # filtering by extension: without it, grep prints "Binary file X matches" for any .so,
+  # .map or image whose bytes happen to contain the pattern, and that line would be counted
+  # as a LEAK. A gate that cries wolf on a shared object gets muted, and a muted deanon
+  # gate is the same as no deanon gate.
+  m="$(grep -inEI "$PAT" "$file" 2>/dev/null || true)"
+  if [ -n "$m" ]; then
+    while IFS= read -r line; do
+      echo "  LEAK  $label:$line"
+      hits=$((hits+1))
+    done <<< "$m"
+  fi
+}
+
+# 1. PDF text (via pymupdf — same dep render-paper uses)
+if [ -f "$DIR/$PDF" ]; then
+  if python3 - "$DIR/$PDF" "$TMP/pdf.txt" <<'PY' 2>/dev/null
+import sys
+try:
+    import fitz
+except Exception:
+    sys.exit(3)
+d = fitz.open(sys.argv[1])
+open(sys.argv[2], "w").write("".join(p.get_text() for p in d))
+PY
+  then scan "PDF[$PDF]" "$TMP/pdf.txt"
+  else echo "  WARN: could not extract PDF text (pymupdf missing?) — PDF not scanned"; fi
+else
+  echo "  WARN: $DIR/$PDF not found — PDF not scanned"
+fi
+
+# 2. Every file under artifact/ (README, scripts, data) — text files only
+if [ -d "$DIR/artifact" ]; then
+  while IFS= read -r f; do
+    scan "artifact/${f#"$DIR"/artifact/}" "$f"
+  # 🔴 NO EXTENSION FILTER, and that is the fix, not an oversight. Until 2026-08-07 this
+  # branch carried an allow-list (*.md *.py *.txt *.json *.js *.ts *.sh *.cfg *.csv *.tex
+  # LICENSE README*) and SILENTLY SKIPPED everything else — while printing "PASS: no
+  # author-identifying strings found. Safe for double-blind." about files it never opened.
+  #
+  # Measured, one fixture, identical bytes, three placements:
+  #   artifact/config.yml (loose)     -> exit 0, the affirmative safety sentence
+  #   config.yml inside artifact.zip  -> exit 1, LEAK reported
+  #   config.yml via an extra path    -> exit 1, LEAK reported
+  #
+  # The other two branches never filtered — the zip branch decodes every member, the
+  # extra-path branch is a bare `find -type f` — so the allow-list was an oversight, and
+  # the header's own contract says "every artifact file". Unlisted extensions that are
+  # ordinary in a real artifact: .yml/.yaml, .toml, Dockerfile, Makefile, .ipynb,
+  # .env.example, and .bib — a BibTeX file under artifact/ carrying the author's own name
+  # is the sharpest version of the miss.
+  #
+  # `grep -I` in scan() skips binary files, which is what the zip branch already relies on,
+  # so dropping the filter costs a little time and no correctness.
+  done < <(find "$DIR/artifact" -type f 2>/dev/null)
+fi
+
+# 3. Contents of any *.zip in the paper dir (and any extra zip paths) — the sneaky one
+scan_zip() { # scan_zip <zip>
+  local zip="$1"
+  [ -f "$zip" ] || return 0
+  python3 - "$zip" "$TMP/zipdump" <<'PY' 2>/dev/null || return 0
+import sys, zipfile, os
+z, out = sys.argv[1], sys.argv[2]
+os.makedirs(out, exist_ok=True)
+zf = zipfile.ZipFile(z)
+for n in zf.namelist():
+    if n.endswith("/"): continue
+    try:
+        data = zf.read(n)
+        text = data.decode("utf-8", "ignore")
+    except Exception:
+        continue
+    safe = n.replace("/", "__")
+    open(os.path.join(out, safe), "w").write(text)
+PY
+  if [ -d "$TMP/zipdump" ]; then
+    for f in "$TMP/zipdump"/*; do
+      [ -f "$f" ] || continue
+      scan "$(basename "$zip")!$(basename "$f" | sed 's/__/\//g')" "$f"
+    done
+    rm -rf "$TMP/zipdump"
+  fi
+}
+while IFS= read -r z; do scan_zip "$z"; done < <(find "$DIR" -maxdepth 2 -name '*.zip' -type f 2>/dev/null)
+
+# 4. Extra caller-supplied paths (dirs or zips)
+for p in "${EXTRA[@]:-}"; do
+  [ -z "$p" ] && continue
+  if [ -d "$p" ]; then
+    while IFS= read -r f; do scan "extra:${f}" "$f"; done < <(find "$p" -type f 2>/dev/null)
+  elif [[ "$p" == *.zip ]]; then scan_zip "$p"
+  elif [ -f "$p" ]; then scan "extra:$p" "$p"; fi
+done
+
+echo "-- summary: $hits identifier hit(s) across PDF + artifact + zip"
+if [ "$hits" -eq 0 ]; then
+  echo "PASS: no author-identifying strings found. Safe for double-blind."
+  exit 0
+else
+  echo "FAIL: de-anonymization leak(s) above — scrub before submitting double-blind."
+  exit 1
+fi

package/skills/submit-paper/references/publishers/acm.md

@@ -0,0 +1,92 @@
+---
+title: "ACM — camera-ready mechanics, shared across ALL ACM venues"
+created: 2026-08-24
+updated: 2026-08-24
+tags: [publisher, acm, camera-ready, erights, ccs]
+---
+
+# ACM: what camera-ready requires
+
+> This is the **publisher** level, not the venue level. Identical for AgenticDev@ASE, AISec@CCS,
+> and any other ACM conference. The venue card (`../venues/<venue>.md`) holds only what's its own:
+> dates, page limit, its own DOI/ISBN, blind model.
+>
+> Captured 2026-08-24 from the live AgenticDev #20 HotCRP form — field by field, not from memory.
+
+## 🔴 Order is mandatory, and the first step breaks everything else
+
+**eRights is filled in FIRST and requires the FINAL title and author list.** So any decision about
+a title change has to be made **before** it — otherwise the copyright block gets issued under the
+old name.
+
+1. **ACM eRights** — a form from ACM, arrives as a link in HotCRP. The final version isn't
+   accepted without it.
+2. Update **Title** on the portal if it changed (it still holds the submission version).
+3. Convert the preamble to camera-ready (table below).
+4. Insert the copyright block that eRights returned.
+5. Build the PDF → run the portal's format checker.
+6. Upload the **Final version** (PDF) **and Source files** — both are mandatory.
+7. References: easiest is uploading `.bbl` + `.aux`.
+8. Fill in **ACM keywords** and **CCS codes**.
+9. `Save and resubmit`.
+
+## Preamble: submission mode ≠ camera-ready
+
+The typical submission preamble is **wrong** for the final. What to change:
+
+| in the submission                       | in camera-ready        | why                                                        |
+| --------------------------------------- | ---------------------- | ---------------------------------------------------------- |
+| `\settopmatter{printacmref=false}`      | remove                 | the ACM reference line has to print                        |
+| `\setcopyright{none}`                   | the value from eRights | otherwise the block is wrong                               |
+| `\pagestyle{plain}`                     | remove                 | the form explicitly says **"Do not include page numbers"** |
+| the `review` option on `\documentclass` | remove                 | it adds line numbers                                       |
+
+Plus: **all fonts must be embedded** in the PDF.
+
+The **copyright block** goes in the **bottom-left corner of the first page**, its exact form
+depends on the choice made in eRights:
+
+```
+<Venue> '<YY>, <dates>, <city>, <country>
+© <year>
+ACM ISBN <ISBN>
+https://doi.org/<DOI>
+```
+
+⚠️ **DOI and ISBN are assigned in advance** and are visible on the form before upload — there's no
+need to wait for them or make them up.
+
+## Portal fields
+
+| Field                  | Req. | What matters                                                                                                          |
+| ---------------------- | ---- | --------------------------------------------------------------------------------------------------------------------- |
+| **ACM eRights**        | ✱    | before everything else; final title + authors                                                                         |
+| **Title**              | ✱    | "exactly as it should appear in the ACM Digital Library"                                                              |
+| **Final version**      | ✱    | PDF, up to 600 MB                                                                                                     |
+| **Abstract**           | ✱    | markdown + LaTeX math, has a Preview                                                                                  |
+| **Source files**       | ✱    | **ALL** build files: `.tex`, `.bib`, figures. Best as a single `.zip`                                                 |
+| **ACM keywords**       | —    | one **per line**                                                                                                      |
+| **CCS**                | —    | terms from **2012 CCS** via the CCS Browser: Assign concepts → View your CCS Concept → Generate Code → paste the XML  |
+| **References**         | —    | upload `.bbl` + `.aux`, or as text one per line                                                                       |
+| **Supplements**        | —    | an **integral** part of the work (proof, online appendix)                                                             |
+| **Auxiliary material** | —    | NOT integral, but still lands in ACM DL: **dataset/code**. ACM does not claim copyright, but gets distribution rights |
+
+## What to decide deliberately
+
+- **Auxiliary material.** Whether to also put the artifact in ACM DL alongside OSF/GitHub. Upside —
+  permanence and an extra findability point; downside — ACM gets distribution rights. Decide per
+  paper, not per venue.
+- **APC.** If the venue charges one — pay only **after** ACM's author kit, which has the amount and
+  the process.
+
+## The frozen version
+
+Camera-ready is a separate **stage** in the sense of `versions/` (see `checkFrozenVersions()` in
+`paper-lint.mjs`): the uploaded PDF is stored as `versions/<date>-camera-ready.pdf`, and the commit
+of its source is recorded in `PIPELINE-STATUS.md` as `commit <hash>`.
+
+**Why this isn't a formality:** the arXiv posting and the extension paper get diffed against
+camera-ready later, and reviewer references to line numbers only resolve against the exact render
+a human read. 🔴 The failure that made this rule exist: `agenticdev-2026` had the wrong submission
+PDF sitting there for a month (340,952 B vs 352,357 B), and when a reviewer cited `L. 166`, there
+was nothing to check it against.