Execution goal: produce an MDRGNB in-depth briefing every day, and a Top3 weekly report every week.
Fixed directories
- Project root:
/home/chenyechao/.openclaw/workspace/mdrgnb_proj - Deduplication file:
/home/chenyechao/.openclaw/workspace/mdrgnb_proj/evaluated_papers.json - Single-article output:
/home/chenyechao/.openclaw/workspace/mdrgnb_proj/papers/<pmid>/summary.mdscores.md
If the directory/file does not exist, create it first.
Daily workflow (strict)
- Search only same-day newly published PubMed literature (time zone:
Asia/Singapore)。 - Keywords center on MDRGNB (multidrug-resistant Gram-negative bacteria), and obviously irrelevant topics are filtered out.
- Deduplicate against
evaluated_papers.json. - Read through each candidate article one by one:
- Prefer the full text;
- When the full text cannot be obtained, read the abstract in full and note the limitation "full text unavailable" in the summary.
- Select the Top 3 by relevance + value.
- Generate for each article
summary.md: must answer the 10 core questions (seereferences/ten_questions.md)。 - Generate for each article
scores.md: four-dimensional scoring (1-10)- Engineering application value
- Architectural innovation
- Theoretical contribution
- Result reliability and give an overall assessment (brief rationale).
- Update
evaluated_papers.json(at least record pmid, title, date, source). - Output the "MDRGNB Daily In-Depth Briefing" for push notification:
- Number of newly retrieved items today
- Top3 (title + PMID + link)
- Each article must include: journal name + ranking information (at least the CAS ranking and the JCR/SCI ranking; if only one is available, note that)
- Each article's 10-question summary key points + four-dimensional score
- Conclusion of the day and suggestions for tomorrow
- If there are no high-quality new additions: state this clearly and provide 1 high-value review knowledge point
Weekly workflow
- Every Sunday at 10:00 (Asia/Singapore), based on this week's evaluated literature, generate the "MDRGNB Weekly Report Top3":
- Reasons for Top3 selection
- Each article's core contribution + journal name + ranking information (CAS/JCR) + 10-question highlights + four-dimensional score
- This week's trend assessment + next week's search suggestions (3 items)
Push requirements
- By default, push to Feishu + QQBot.
- Output the final content directly, and avoid requiring "call the message tool" again in isolated tasks.
- Any send failure must report the failure reason and the suggested repair path (target ID, channel, permissions, length limit).