med-monthly-synthesis

name: med-monthly-synthesis description: 月度综合与长期资产维护技能。Use when finalizing daily 隔日报.xlsx and monthly 月报-*.docx exports after Stage2/Stage3 are complete.

med-weekly-synthesis

1. Run only as Stage 4 (final export stage):

• Must run after Stage2 (ABC writeback) and Stage3 (translation backfill) complete
• If Stage2/Stage3 failed or stale, do not emit normal-success final xlsx

2. Export path resolution (dynamic, NOT hardcoded):

• All exports use reviewRoot from runtime_config.mjs (default: research_os/文献评价)
• Daily: research_os/<yy.MM>/<MM.dd>/隔日报.xlsx
• Monthly: research_os/<yy.MM>/月报-<period>.docx

3. Daily export (隔日报.xlsx):

• Two sheets: "每日反馈" + "需人工复核" (see med-daily-triage for sheet rules)
• Data source: desktop_daily_review_source.json filtered by Stage2 writeback items

4. Monthly export (月报-*.docx):

• Generated by monthly_docx_report.mjs
• Covers monthly synthesis window
• Include trend analysis, preference drift, unresolved items

5. Export method policy (see med-export-policy for full fallback chain):

• Default: spreadsheets_skill
• Fallback: node_fallback → python_spawn_legacy → manual_required
• Every export must record audit fields

6. Keep pipeline/state artifacts in research_os/.../pipeline (JSON/audit only), not as final workbook destination.

7. Evidence boundary:

• Only pdf_fulltext can contain full experimental detail (dose/time/results/stats)
• metadata_only/abstract_only rows must stay within available evidence

8. Integrity rules:

• Preserve audit fields and source level
• Do not destructively rewrite historical rows
• PDF acquisition is out of automation scope; do not mark synthesis failed due to missing PDF queue processing
• Include translation success/failure metrics in weekly logs when available
• Include star-migration outcome summary when available