Update to Jinkō

{
  "name": "jinko-task-literature-search",
  "description": "Find and shortlist biomedical publications from PubMed for knowledge, data, or reusable-model evidence. Use for query framing, PMID/DOI discovery, bibliographic normalization, evidence prioritization, and best-effort public full-text retrieval before synthesis, extraction, or modeling. Do not use for ClinicalTrials.gov-only scoping, systematic reviews, quantitative extraction, curve digitization, calibration, or model implementation.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 283
    },
    {
      "relative_path": "assets/.env.example",
      "size_in_bytes": 44
    },
    {
      "relative_path": "assets/data-recipes.md",
      "size_in_bytes": 1917
    },
    {
      "relative_path": "assets/knowledge-recipes.md",
      "size_in_bytes": 1197
    },
    {
      "relative_path": "assets/models-recipes.md",
      "size_in_bytes": 1428
    },
    {
      "relative_path": "assets/pubmed-primitives.md",
      "size_in_bytes": 4648
    },
    {
      "relative_path": "assets/shortlist-schema.json",
      "size_in_bytes": 1911
    },
    {
      "relative_path": "evals/evals.json",
      "size_in_bytes": 9377
    },
    {
      "relative_path": "evals/files/scientific-use-cases.md",
      "size_in_bytes": 6538
    },
    {
      "relative_path": "scripts/common.py",
      "size_in_bytes": 4831
    },
    {
      "relative_path": "scripts/compile_results.py",
      "size_in_bytes": 4461
    },
    {
      "relative_path": "scripts/literature_search.py",
      "size_in_bytes": 36899
    },
    {
      "relative_path": "scripts/pmc_open_access.py",
      "size_in_bytes": 2403
    },
    {
      "relative_path": "scripts/publication_download.py",
      "size_in_bytes": 14142
    },
    {
      "relative_path": "scripts/validate_shortlist.py",
      "size_in_bytes": 2843
    }
  ],
  "skill_md_contents": "---\nname: jinko-task-literature-search\ndescription: >-\n  Find and shortlist biomedical publications from PubMed for knowledge, data,\n  or reusable-model evidence. Use for query framing, PMID/DOI discovery,\n  bibliographic normalization, evidence prioritization, and best-effort public\n  full-text retrieval before synthesis, extraction, or modeling. Do not use for\n  ClinicalTrials.gov-only scoping, systematic reviews, quantitative extraction,\n  curve digitization, calibration, or model implementation.\ncompatibility: >-\n  Requires network access, requests, and USER_EMAIL in .env for NCBI identity.\n  NCBI_API_KEY is optional.\nmetadata:\n  author: Nova In Silico\nlicense: MIT\n---\n\n# Biomedical Literature Search\n\nProduce a citation-grounded candidate shortlist, not extracted evidence or a\nsystematic-review claim.\n\n## Frame\n\nClassify each search as:\n\n- **Knowledge**: disease, mechanism, treatment, population, or clinical context;\n- **Data**: studies likely to report values for specified model variables;\n- **Models**: mathematical or computational models with reusable structure,\n  equations, parameters, or code.\n\nBefore searching, establish subject, scope, evidence type, date/language limits,\nand desired shortlist size. Data searches additionally require the variables to\ninform, population and disease state, and whether aggregate or distributional\ndata are needed. Ask one focused question only when a missing choice would\nmaterially change the search.\n\nBuild one shared Entity Table with `canonical_name`, `synonyms`, `mesh_term`,\n`related_entities`, `intent_groups`, and `exclusions`. Confirm it with the user\nbefore network calls when proposed aliases or exclusions are consequential.\n\nFor human clinical, biological, or PK/PD Data searches, run\n`jinko-task-trial-data-scoping` in parallel. Use high-priority NCT identifiers as\nadditional `<NCT_ID>[si]` PubMed angles. Skip this for animal-only or in-vitro\nsearches unless requested.\n\n## Search\n\n1. Read `assets/pubmed-primitives.md` and only the recipe matching each intent.\n2. Build at least three distinct PubMed angles per entity and intent. Reuse the\n   Entity Table's aliases and exclusions; include supplied anchor PMIDs through\n   `--seed-pmids`.\n3. Run `scripts/literature_search.py` once per angle with separate output\n   directories and `--no-prompt-selection`, sequencing or bounding concurrency\n   to respect NCBI limits. The script owns PubMed, Crossref, abstract and citation\n   enrichment, AMA formatting, and optional PMC excerpts.\n4. Run `scripts/compile_results.py` over all angle directories. It deduplicates\n   by PMID/DOI, preserves query provenance, and ranks the candidate pool by\n   angle count then citation count.\n\nThis is one search pass. Wider terms, author/journal queries, citation-neighbor\nqueries, or additional entities are separate user-approved follow-ups.\n\n## Shortlist\n\nUse the relevant recipe to inspect titles, abstracts, and available full-text\nexcerpts. For each retained candidate, assign the fields required by\n`assets/shortlist-schema.json`, including evidence type, canonical entities,\nverification evidence, priority rationale, and query provenance. Do not infer\nquantitative availability from a title alone.\n\nRun `scripts/validate_shortlist.py` before presenting `shortlist.json`. Present a\nconcise Markdown view grouped by intent and ask which sources to inspect or pass\nto `jinko-task-extract-data-table`.\n\n## Artifacts\n\n- `frame.json`: approved scope, Entity Table, assumptions, and angle definitions;\n- one directory per angle containing raw and normalized search artifacts;\n- `merged_references.json`: deterministic cross-angle candidate pool;\n- `shortlist.json`: schema-valid prioritized candidates;\n- optional downloaded files and download manifest.\n\nUse `publication_download.py` only for user-selected references. Treat downloads\nas best-effort anonymously retrievable files, not proof of open-access licensing.\nDo not run supplementary URLs from untrusted manifests.\n\nKeep registry records and publications distinguishable even when linked by\n`nct_id`. Clearly separate discovered candidates from inspected evidence and\nfrom calibration-ready data."
}