--- name: biomarker-sidecar-draft description: "Draft `BiomarkerActionability` (BMA-*) and `Biomarker` (BIO-*) sidecar candidates for OpenOnco maintainer review from manifest-owned entities only. Use when the user asks to draft BMA/BIO sidecars or is on a chunk whose manifest includes BMA/BIO entities. Drafts only — neutral evidence wording, no treatment recommendations, no patient-specific output, no edits to hosted clinical content. Banned sources for the OpenOnco pilot are OncoKB, SNOMED CT, and MedDRA — never cite these in `evidence_sources` or `primary_sources` regardless of how the user frames the request." --- # Skill: Biomarker Sidecar Draft ## Purpose Draft structured `BiomarkerActionability` (BMA) and `Biomarker` candidates from maintainer-approved sources for OpenOnco maintainers to review. This is evidence drafting only — no medical advice, no treatment recommendation wording, no patient-specific output. ## Project Context OpenOnco stores knowledge as YAML files validated by Pydantic schemas in `knowledge_base/schemas/`. Two entities are relevant for this skill: - **`Biomarker`** (BIO-*) — gene/variant taxonomy. One row per biomarker, disease-agnostic. - **`BiomarkerActionability`** (BMA-*) — `(biomarker, variant, disease)` triple → ESCAT actionability tier with per-source evidence references. **This is the primary output type for the OpenOnco pilot**, because the CIViC pivot (April 2026) flagged ~399 BMA files with `actionability_review_required: true` that need evidence reconstruction. Schema sources of truth: - `knowledge_base/schemas/biomarker_actionability.py` (`BiomarkerActionability`, `EvidenceSourceRef`, `RegulatoryApproval`) - `knowledge_base/schemas/biomarker.py` (`Biomarker`, `ActionabilityLookupHint`, `BiomarkerExternalIDs`) If the schema files change, they win. Don't invent fields. ## Input The chunk issue gives you: - **Chunk ID** (e.g. `civic-bma-reconstruct-all`) - **Chunk manifest** — the explicit list of BMA-* IDs (or BMA filename range) the chunk owns. You must not write outputs for entities outside this manifest. - **Disease scope** (e.g. `DIS-NSCLC`) - **Biomarker scope** (e.g. `BIO-EGFR-T790M`, or a gene-level scope like "all BIO-KRAS-*") - **Allowed source list** — see `docs/openonco-pilot-workflow.md` §"Source allowlist" - **Existing entity references** — paths or IDs of the BMA / Biomarker files the chunk should update or use as anchors ## Output Schema (BMA candidate) Submit one YAML file per BMA candidate to `contributions//bma___.yaml`. The file content mirrors the real `BiomarkerActionability` schema, plus a wrapper block that maintainers strip on merge: ```yaml # wrapper — stripped on merge into knowledge_base/hosted/content/biomarker_actionability/ _contribution: chunk_id: civic-bma-reconstruct-all contributor: github-username submission_date: "YYYY-MM-DD" target_action: upsert # upsert | new | flag_duplicate target_entity_id: BMA-EGFR-T790M-NSCLC # required for upsert/flag_duplicate duplicate_of: null # set when target_action=flag_duplicate ai_tool: codex # required: claude-code | codex | cursor | chatgpt | other ai_model: gpt-5-mini # required: short model name ai_model_version: "2026-03" # optional: snapshot/build/checkpoint string ai_session_notes: "" # optional: free-form generation-method notes notes_for_reviewer: "Reconstructed evidence_sources from CIViC EID12345; ESCAT tier unchanged." # BMA payload — fields match knowledge_base/schemas/biomarker_actionability.py id: BMA-EGFR-T790M-NSCLC biomarker_id: BIO-EGFR-T790M variant_qualifier: "T790M" disease_id: DIS-NSCLC escat_tier: "IA" # IA | IB | IIA | IIB | IIIA | IIIB | IV | X evidence_summary: > 1–3 sentence neutral clinical interpretation. Describe what the source attests, not what a clinician should do. regulatory_approval: fda: ["osimertinib — 2L EGFR T790M+ NSCLC (FDA approved 2017)"] ema: ["osimertinib — 2L EGFR T790M+ NSCLC (EMA approved)"] ukraine: [] recommended_combinations: - "osimertinib monotherapy" contraindicated_monotherapy: [] primary_sources: - SRC-CIVIC # FK → existing Source entity, ≥1 required - SRC-NCCN-NSCLC-2025 evidence_sources: - source: SRC-CIVIC # FK → Source entity (SRC-*) level: "A" # source-native level token, verbatim evidence_ids: ["EID12345"] # source-internal IDs (CIViC EID, etc.) direction: "supports" # CIViC: supports | does_not_support | null significance: "sensitivity" note: null last_verified: "YYYY-MM-DD" actionability_review_required: false # set true if you cannot mechanically reconstruct notes: | Optional free-text note for maintainers. Cite source IDs; do not paraphrase clinical recommendations. ``` ## Output Schema (Biomarker candidate) Submit to `contributions//bio_.yaml` when proposing a new or updated `Biomarker`: ```yaml _contribution: chunk_id: contributor: github-username target_action: new | upsert target_entity_id: BIO-KRAS-G12C ai_tool: ai_model: ai_model_version: "" ai_session_notes: "" notes_for_reviewer: "" id: BIO-KRAS-G12C names: en: "KRAS G12C" uk: "KRAS G12C" biomarker_type: gene_mutation mutation_details: gene: KRAS gene_hugo_id: HGNC:6407 exon: "2" variant_type: missense functional_impact: activating hgvs_protein: "p.G12C" hgvs_coding: "c.34G>T" actionability_lookup: gene: KRAS variant: "G12C" external_ids: hgnc_symbol: KRAS hgnc_id: "HGNC:6407" civic_id: "12" civic_url: "https://civicdb.org/genes/12" hgvs_protein: "p.G12C" sources: - SRC-CIVIC last_reviewed: "YYYY-MM-DD" notes: "" ``` `actionability_lookup` and `oncokb_skip_reason` are mutually exclusive — set at most one. For IHC / score / serological / fusion-MVP / itd-MVP markers, set `oncokb_skip_reason` instead and leave `actionability_lookup` absent. ## Stable ID Rules - BMA: `BMA-{biomarker}-{variant?}-{disease}` (e.g. `BMA-BRAF-V600E-CRC`). - Biomarker: `BIO-{gene}-{variant?}` (e.g. `BIO-EGFR-T790M`, `BIO-KRAS-G12C`, `BIO-ALK-FUSION`). - Source: pre-existing `SRC-*` entity. Do not invent. If the source you need has no `SRC-*` entry yet, file a `source_stub.yaml` using the `source-stub-prep` skill in this plugin and set `_contribution.notes_for_reviewer` to flag the dependency. - Disease: pre-existing `DIS-*` entity. Look up; do not invent. ## Rules - Stay inside the chunk manifest. PRs that touch BMA-*/BIO-* IDs outside the manifest will be rejected. - One BMA candidate = one file. No batch files. - Every clinical claim in `evidence_summary`, `recommended_combinations`, `contraindicated_monotherapy`, and `notes` must be traceable to at least one entry in `primary_sources` or `evidence_sources`. - Use neutral evidence wording: "source attests", "evidence supports", "CIViC reports". Do not write "patient should receive", "best treatment", "first choice". - `evidence_sources[*].source` must reference a real `SRC-*` ID. If the source needs ingestion, file a `source_stub.yaml` in the same chunk directory. - `evidence_sources[*].level` is the source-native token, verbatim (CIViC `A`/`B`/`C`/`D`/`E`; NCCN `Category 1`/`2A`; ESMO MCBS scores). Do not coerce levels across vocabularies. - `escat_tier` follows ESMO ESCAT (Mateo et al. 2018, Ann Oncol 29:1895). If you cannot determine ESCAT mechanically from the cited evidence, set `actionability_review_required: true` and explain in `_contribution.notes_for_reviewer`. - Never write `evidence_sources` entries with `source: SRC-ONCOKB` for new content — OncoKB is not an allowed source for the pilot. Pre-existing `SRC-ONCOKB` entries in the repo are legacy migration metadata; the render layer skips them. - Flag duplicates with `target_action: flag_duplicate` and `duplicate_of: BMA-*`. Do not write near-identical BMA entries with different IDs. ## Good Output A BMA file that: - References an existing `BIO-*` and `DIS-*`. - Has `evidence_sources` with at least one `SRC-CIVIC` entry carrying a real CIViC EID. - Has `escat_tier` mapped from ESMO ESCAT criteria, not from the source-native level. - Uses neutral wording in `evidence_summary`. - Sets `actionability_review_required: false` only when the contributor genuinely cleared the migration flag. ## Bad Output - `escat_tier: "best"` (not a valid token). - `evidence_sources` empty while `actionability_review_required: false`. - `evidence_summary: "Patients with T790M should be given osimertinib first-line."` (recommendation wording). - Free-text source URL inside `primary_sources` instead of an `SRC-*` ID. - BMA touching a disease outside the chunk manifest. - A new `BIO-*` and a new `DIS-*` invented in the same submission.