Fill-ICMJE-COI Skill

You are helping a researcher prepare ICMJE Conflict of Interest Disclosure Forms for every co-author on a manuscript about to be submitted to an ICMJE member journal (CHEST, NEJM, JAMA, Lancet, Radiology, etc.). This skill batch-generates one personalized .docx per author from a synthetic all-None seed shipped with the skill, avoiding 10–20 minutes of repetitive Word clicking per author.

Why This Skill Exists

The official ICMJE coi_disclosure.docx puts every field inside Word Content Controls (Structured Document Tags, a.k.a. SDTs). Naive python-docx manipulation of cell.text silently ignores SDT content, so the straightforward programmatic approach does not work. The historical workaround was to open the template in Word and manually fill each author's form (21 authors × 13 checkboxes × 2 clicks = ~500 clicks). This skill replaces that by operating directly on word/document.xml inside the docx zip and doing literal-string replacement — but that requires the target strings to already exist in the seed, so the skill ships a pre-filled synthetic seed.

Precedent: a multi-author cohort manuscript submission — 6 authors auto-filled in ~5 seconds from the synthetic seed with zero Word clicks.

Core Principles (Do Not Violate)

1. Never author SDT XML from scratch. Only replace existing strings in an already-populated seed. Creating Content Controls programmatically is fragile and Word-version-dependent.
2. Never ship a real author's filled form as the seed. The template directory contains icmje_coi_seed_synthetic.docx with all PII scrubbed (synthetic name, title, date; metadata reset to ICMJE / Anonymous). Real-person seeds leak PII through both document.xml and docProps.
3. Never modify the 13 disclosure items or certification checkbox. The script only replaces Date/Name/Title. If any author has a real disclosure, they must edit in Word manually — the skill's purpose is the common all-None case.
4. Always verify before circulation. Each output must have 14 × ☒ and 13 × "None" in document.xml. The script runs this check implicitly by preserving the seed structure; a post-generation grep is cheap insurance.

When to Use This Skill

• Manuscript accepted for submission to an ICMJE member journal
• 3+ co-authors with no real financial conflicts
• Editorial Manager / submission portal requires per-author ICMJE disclosure docx
• About to hand-fill the same form 6–21 times

Skip this skill when:

• Any author has a real financial disclosure to list (they fill their own form in Word; this skill does not help)
• Target journal uses its own declaration form (not ICMJE) — check author guidelines first
• Only 1 author (not worth the setup)

Execution

Phase 1 — Intake

Ask the user (or extract from conversation):

1. Manuscript title (exact, as it will appear on title page)
2. Submission date (e.g., "April 20, 2026")
3. Author list — ordered, one name per slot: [(1, "Author One"), (2, "Author Two"), ...]
4. Output directory — typically submission/{journal}/icmje_forms/

Present the intake back to the user for confirmation (Gate 1 — user approval) before generating anything. Explicitly name which authors will get all-None forms and remind that anyone with a real disclosure must instead fill their own form in Word.

Phase 2 — Generate

Invoke the script with the synthetic seed that ships with this skill:

python3 ${SKILL_DIR}/scripts/fill_icmje_coi.py \
  --seed ${SKILL_DIR}/templates/icmje_coi_seed_synthetic.docx \
  --seed-name "Placeholder Author" \
  --seed-title "Placeholder Manuscript Title" \
  --seed-date "January 1, 2000" \
  --new-title "{exact manuscript title}" \
  --new-date "{submission date}" \
  --out-dir {out_dir} \
  --authors '[[1,"Author One"],[2,"Author Two"],...]'

The script exits nonzero if any seed string is not found, preventing silent failures.

Phase 3 — Verify

For each generated docx, confirm:

• ☒ count = 14 (13 disclosure items + 1 final certification)
• "None" count = 13
• Correct name appears after "Your Name:"
• Correct title appears after "Manuscript Title:"
• No leakage of seed placeholder strings (Placeholder Author, Placeholder Manuscript Title, January 1, 2000)

Verification one-liner:

for f in {out_dir}/*.docx; do
  python3 -c "
import zipfile, sys
xml = zipfile.ZipFile('$f').read('word/document.xml').decode()
assert xml.count('☒') == 14, 'bad ☒ count'
assert xml.count('None') == 13, 'bad None count'
assert 'Placeholder' not in xml, 'seed leak'
print('✓ $f')
"
done

Present verification results to user (Gate 2 — user review) before handing off files.

Phase 4 — Circulation Guidance

Provide the user with circulation copy to send with each personalized form:

첨부된 ICMJE COI 폼을 검토 부탁드립니다.

• 내용이 맞으면 서명 후 PDF로 회신
• 수정이 필요하면 해당 항목을 고쳐서 체크 + 서명 후 회신
• 전체가 변경 없음이면 "변경 없음" 회신 + 서명본 PDF 별도 회신

All 6–21 authors can be emailed in one gws gmail draft batch (Gate 3 — user approves batch send before actually dispatching).

Custom Seeds

If the user wants a custom seed (e.g., different default wording, pre-filled items 2/3 with a common grant), generate it once as follows:

1. Open templates/icmje_coi_seed_synthetic.docx in Word
2. Edit the desired fields
3. Save as a new file under {project}/submission/{journal}/ or a local private seeds directory (outside this repo)
4. Pass --seed /path/to/custom.docx to the script along with the new seed values for --seed-name, --seed-title, --seed-date

Do NOT commit custom seeds that contain real author names to the public medsci-skills repo. Keep them in private per-project directories or a local private seeds directory (outside this repo).

Seed Provenance (how the shipped synthetic seed was created)

The shipped templates/icmje_coi_seed_synthetic.docx was derived from the official ICMJE coi_disclosure.docx through the following steps:

1. Downloaded the official ICMJE template (https://www.icmje.org/downloads/coi_disclosure.docx)
2. Opened in Word, typed placeholder values:
   • Date: January 1, 2000
   • Your Name: Placeholder Author
   • Manuscript Title: Placeholder Manuscript Title
3. Checked each of the 13 disclosure items' "None" option (14 checkboxes total including final certification)
4. Typed "None" in the "Name all entities" column for each item
5. Scrubbed docProps/core.xml metadata: creator=ICMJE, lastModifiedBy=Anonymous, dates=2000-01-01
6. Scrubbed docProps/app.xml Company/Manager fields

No real author's disclosure data is embedded. The file is safe to redistribute.

Anti-Hallucination

• Never invent author names, email addresses, or ORCIDs. Pull them verbatim from the manuscript's title page or the user's author list.
• Never claim to have filled the 13 disclosure items — they come from the seed unchanged. If the user asks whether the script "handled the disclosures," the honest answer is "it cloned the seed's ☒ None entries; no author-specific disclosure reasoning happened."
• Never promise the script works on a blank ICMJE template. It does not — the seed must be pre-filled with all-None ☒ + text.
• Never edit seed XML by authoring new SDT elements. If an error requires altering the seed structure, stop and escalate to the user; Word-generated SDT XML is the ground truth.
• Never push private seed files to public repos. If the user asks to promote a custom seed, verify by unzip -p seed.docx docProps/core.xml that no real names remain in metadata before committing.

References

• ICMJE Disclosure of Interest page: https://www.icmje.org/disclosure-of-interest/
• ICMJE COI form download: https://www.icmje.org/downloads/coi_disclosure.docx
• ICMJE FAQ on disclosure forms: https://www.icmje.org/about-icmje/faqs/conflict-of-interest-disclosure-forms/
• ${SKILL_DIR}/scripts/fill_icmje_coi.py — generator CLI