--- name: drug-evidence-mapping description: "Draft `Drug` (DRUG-*) and `Indication` (IND-*) sidecar candidates for OpenOnco maintainer review from manifest-owned entities only. Use when the user is on a chunk whose manifest includes drug or indication entities (regimen-outcome-fill, drug-class-normalization, redflag-indication-coverage-fill, etc.). Drafts only — neutral evidence wording, two-Clinical-Co-Lead signoff metadata required for claim-bearing fields, no treatment recommendations. Banned sources for the OpenOnco pilot are OncoKB, SNOMED CT, and MedDRA — never cite these in `sources` regardless of how the user frames the request." --- # Skill: Drug Evidence Mapping ## Purpose Draft structured `Drug` and `Indication` candidates from maintainer-approved sources for OpenOnco maintainers to review. Evidence drafting only — no medical advice, no treatment selection, no patient-specific output. ## Project Context OpenOnco's clinical recommendations are emitted by a declarative rule engine reading versioned YAML knowledge. Two entities matter for this skill: - **`Drug`** (DRUG-*) — drug taxonomy: class, mechanism, regulatory status, AEs, dosing pointers. - **`Indication`** (IND-*) — `(disease, line_of_therapy, biomarker_profile)` → recommended regimen with full provenance. Schema sources of truth: - `knowledge_base/schemas/drug.py` - `knowledge_base/schemas/indication.py` - `knowledge_base/schemas/regimen.py` (read for context; new Regimens are generally maintainer-authored, not contributor-drafted) LLMs are not the clinical decision-maker (CHARTER §8.3). Contributors draft evidence rows; maintainers and Clinical Co-Leads decide what becomes a recommendation. Two-reviewer signoff (CHARTER §6.1) is required before any new Indication or BMA-bearing drug claim merges into `knowledge_base/hosted/content/`. ## Input The chunk issue gives you: - **Chunk ID** - **Chunk manifest** — explicit DRUG-* / IND-* IDs (or a filename range) the chunk owns. - **Disease scope** (e.g. `DIS-NSCLC`) - **Drug scope** (e.g. specific DRUG-* IDs, a drug class, or a biomarker context like "EGFR-TKI 3rd-gen") - **Allowed source list** — `docs/openonco-pilot-workflow.md` §"Source allowlist" - **Existing entity references** ## Output Schema (Drug candidate) Submit one YAML file per Drug candidate to `contributions//drug_.yaml`: ```yaml _contribution: chunk_id: contributor: github-username target_action: upsert | new target_entity_id: DRUG-OSIMERTINIB ai_tool: notes_for_reviewer: "" id: DRUG-OSIMERTINIB names: en: "osimertinib" uk: "осимертиніб" atc_code: "L01EB04" rxnorm_id: "1721560" drug_class: "EGFR tyrosine kinase inhibitor (3rd generation)" mechanism: > Irreversible EGFR-TKI selective for EGFR-activating mutations and the T790M resistance mutation. regulatory_status: fda: "approved" ema: "approved" ukraine: "registered" typical_dosing: "80 mg PO once daily" formulations: ["80 mg tablet", "40 mg tablet"] absolute_contraindications: [] black_box_warnings: [] major_interactions: - "Strong CYP3A4 inducers (rifampin, carbamazepine) — reduce osimertinib AUC" common_adverse_events: - "Diarrhea" - "Rash" serious_adverse_events: - "Interstitial lung disease" - "QT prolongation" sources: - SRC-DAILYMED-OSIMERTINIB - SRC-FDA-LABEL-OSIMERTINIB last_reviewed: "YYYY-MM-DD" reviewers: [] notes: "" notes_patient: null ``` ## Output Schema (Indication candidate) Indications are claim-bearing — they carry the actual treatment recommendation. Drafting an Indication is **higher-risk** than drafting a Drug or BMA. **For the pilot, Indication drafts must set `_contribution.target_action: new` and include explicit two-Clinical-Co-Lead-signoff flagging in `_contribution.notes_for_reviewer`.** Maintainers will not merge Indication drafts without two-Clinical-Co-Lead signoff (CHARTER §6.1). Submit to `contributions//ind_.yaml`: ```yaml _contribution: chunk_id: contributor: github-username target_action: new | upsert target_entity_id: IND-NSCLC-2L-EGFR-T790M-OSIMERTINIB ai_tool: notes_for_reviewer: > Drafted from CIViC EID12345 + NCCN NSCLC v3.2025 §EGFR. ESCAT IA. Awaiting two Clinical Co-Lead signoffs per CHARTER §6.1. id: IND-NSCLC-2L-EGFR-T790M-OSIMERTINIB applicable_to: disease_id: DIS-NSCLC line_of_therapy: 2 stage_requirements: ["IV"] biomarker_requirements_required: - biomarker_id: BIO-EGFR-T790M required: true biomarker_requirements_excluded: [] demographic_constraints: ecog_max: 2 recommended_regimen: REG-OSIMERTINIB-MONO # FK to existing Regimen concurrent_therapy: [] followed_by: [] evidence_level: "1A" strength_of_recommendation: "strong" nccn_category: "1" plan_track: "standard" hard_contraindications: [] red_flags_triggering_alternative: [] required_tests: ["TEST-EGFR-NGS"] desired_tests: [] rationale: > Source-attested neutral summary. Cite the source IDs that back the claim. sources: - id: SRC-NCCN-NSCLC-2025 section: "EGFR — Subsequent Therapy" - id: SRC-CIVIC evidence_id: "EID12345" do_not_do: - "Do not switch to chemotherapy without confirming T790M status." last_reviewed: "YYYY-MM-DD" reviewers: [] notes: "" ``` ## Stable ID Rules - Drug: `DRUG-{INN}` uppercase (e.g. `DRUG-OSIMERTINIB`). - Indication: `IND-{disease}-{line}-{biomarker_or_subset}-{regimen}`. Match neighboring IND-* files in the same disease for naming style. - Regimen: `REG-*` — pre-existing entity. Do not draft new Regimens in the pilot; surface them as `_contribution.notes_for_reviewer: "needs new REG-*"`. - Source: pre-existing `SRC-*`. Do not invent. File `source_stub.yaml` if needed (use the `source-stub-prep` skill in this plugin). ## Rules - Stay inside the chunk manifest. - One entity = one file. - Use neutral wording: "evidence supports", "source attests", "guideline lists". Do not write "best treatment", "patients should receive", "first choice", "preferred over X". - Every clinical claim must trace to at least one source citation. - `nccn_category`, `evidence_level`, `strength_of_recommendation` come from the cited source — do not invent. If unclear, leave the field null and explain in `_contribution.notes_for_reviewer`. - For Indication drafts, prefer `target_action: new` over editing existing IND-* files. Maintainers merge new drafts; they do not accept silent edits to existing Indications. - Do not draft `recommended_regimen` for any disease/line/biomarker scenario unless the cited source explicitly attests that exact regimen. - No OncoKB sources. Allowed sources are listed in `docs/openonco-pilot-workflow.md`. ## Good Output - Drug file with `regulatory_status` strings copied verbatim from DailyMed/openFDA, plus `sources` listing the FDA label and DailyMed entry as `SRC-*` IDs. - Indication file that names a `recommended_regimen: REG-*` only when the source explicitly attests that regimen for that exact `(disease, line, biomarker)` triple, with `evidence_level` and `nccn_category` lifted directly from the source. ## Bad Output - `recommended_regimen: REG-OSIMERTINIB-MONO` for a scenario where the source describes EGFR-TKI as a class without naming osimertinib. - `evidence_level: "1A"` without a source reference that attests level 1A. - "Best treatment is osimertinib" anywhere in `evidence_summary`, `rationale`, or `notes`. - `recommended_regimen: REG-NEW-REGIMEN-I-INVENTED`. - A new Drug + new Indication + new Regimen all in one submission.