{
  "version": 3,
  "created_at": "2026-07-10T23:02:47.637Z",
  "task_summary": "Make the drug-program pipeline archetype-adaptive for genetic/rare diseases",
  "agents": [],
  "phases": [
    {
      "name": "Plan",
      "delegations": [
        {
          "steps": [
            {
              "title": "Add archetype + mechanism model to kernel.py",
              "description": "Add helpers: classify_archetype(facts) returning genetic architecture (monogenic/oligogenic/polygenic/chromosomal/mitochondrial/somatic) + data_density (rich/moderate/sparse); recommend_modality_genetic(mechanism, constraint) implementing the mechanism→modality decision tree (LoF→replacement/upregulation, GoF/dominant-negative→knockdown/allele-specific editing, splice→splice-switching ASO, repeat-expansion→knockdown/editing, nonsense→read-through); and data_triage(n_cohorts, n_trials) returning which Stage-3/4 path to run. Extend init_dossier with disease_archetype, mechanism_of_pathogenesis, data_density fields and an ARCHETYPE LOCK gate. Keep all existing helpers and the common-disease path intact. Unit-test every new function in the python kernel before upload."
            },
            {
              "title": "Add the genetic mechanism→modality decision-tree asset",
              "description": "Author assets/genetic_modality_tree.csv (or .md): mechanism_of_pathogenesis × gene-dosage/constraint → recommended modality family → which design/validation skills apply → key delivery consideration (e.g. AAV tropism, LNP, intrathecal ASO). Cover gene replacement, ERT, ASO knockdown, siRNA, splice-switching ASO, base/prime editing, CRISPR nuclease, read-through, and small-molecule where applicable. Ground the rows in real approved-precedent examples (nusinersen, onasemnogene, patisiran, casimersen, etc.) with a caveat that mechanism must be human-confirmed."
            },
            {
              "title": "Add ARCHETYPE LOCK gate to checkpoint_map + protocol",
              "description": "Add the ARCHETYPE LOCK row (hard gate, fires after Stage 1) to assets/checkpoint_map.csv with its gate question, go-criteria, and no-go action; add a data-sparsity 'kill-or-pivot vs proceed-on-genetics' decision. Update checkpoint_protocol.md § to describe archetype classification as a hard gate and mechanism-of-pathogenesis as a claim requiring clinical validation. Update program_dossier.schema.json with the new fields and gate."
            },
            {
              "title": "Rewrite stage_spec.md stages 3–9 as archetype-adaptive",
              "description": "Add a new 'Stage 1b / ARCHETYPE LOCK' section and revise: Stage 3 (add natural-history + analogous-modality-precedent path for sparse indications), Stage 4 (branch: omics meta-analysis when data-rich vs genetics-led target/mechanism confirmation when sparse — gnomAD constraint, ClinVar, ClinGen), Stage 5 (route through recommend_modality_genetic for genetic archetypes), Stage 6 (modality routing for gene therapy/ASO/editing incl. delivery), Stage 8 (rare-disease regulatory toolkit: orphan, RPD PRV, accelerated approval on surrogate, natural-history external controls, platform IND, N-of-1 ASO path), Stage 9 (ultra-rare economics: tiny-N, one-time pricing, value-based annuity, registry launch). Keep the common-disease description as the parallel branch."
            },
            {
              "title": "Update SKILL.md: archetype branch + connector wiring",
              "description": "Update SKILL.md: add ARCHETYPE LOCK to the stage/gate table and the composition map; wire the four human-genetics connectors (mcp-variants, mcp-clinical-genomics, mcp-human-genetics, mcp-genes-ontologies) into Stages 1/3/4/5; document the two Stage-5 routers (common recommend_modality vs genetic recommend_modality_genetic) and when each fires; note the data-density triage. Keep description ≤1024 chars. Re-verify referenced skills/connectors all exist."
            },
            {
              "title": "Republish, re-save artifacts, and dry-run on a monogenic disease",
              "description": "Re-upload all changed files to the drug-program-orchestration draft, confirm the kernel.py sidecar gate passes, and publish (overwrite). Re-save changed artifacts as new versions. Then dry-run the archetype-adaptive planning logic on a real monogenic sparse-data disease (e.g. a specific ultra-rare LoF or repeat-expansion disorder) to confirm: archetype classifies correctly, data triage picks the genetics-led path, the mechanism→modality router returns a genetics-appropriate modality (not an EoE/antibody default), and rare-disease regulatory/commercial branches fire. Produce dryrun_<genetic_disease>.md."
            }
          ]
        }
      ],
      "id": "phase-0"
    }
  ],
  "feasibility": {
    "rationale": "All four human-genetics connectors (mcp-variants, mcp-clinical-genomics, mcp-human-genetics, mcp-genes-ontologies) already exist in the catalog, and the change is additive: a new early classification gate plus adaptive branches on existing stages, new kernel helpers, and dossier fields. No new compute or external dependencies. Main risk is keeping the common-disease path unchanged while layering the genetic branch — handled by making archetype a resolved parameter, not a fork of the whole skill.",
    "confidence": "high"
  }
}