{
  "version": 3,
  "created_at": "2026-07-11T22:18:43.877Z",
  "task_summary": "Author 5 skills + 1 specialist from the EoE project and publish to public GitHub",
  "agents": [],
  "phases": [
    {
      "name": "Plan",
      "delegations": [
        {
          "steps": [
            {
              "title": "Reconstruct workflows from project history",
              "description": "Read the key frames/artifacts to extract the REAL methodology for each skill so they encode actual practice, not generic templates: (a) binder campaign — CCL26/POSTN + celiac frames, EoE_binder_design_report.md, S1_binder_designs.csv, S5_binder_liabilities.csv, binder_design_targets.csv (backbone gen tool, MPNN settings, fold-back filters like pLDDT/ipTM/PAE thresholds, output CSV schema); (b) developability — celiac_binder_developability.csv, celiac_binder_protease_map.csv, celiac_binder_hardened.fasta (liability rules, gut-protease mapping, hardening logic); (c) multivalent display — eoe_pmhc_nanoparticle_fusion_architecture.csv, assembly_scenarios.csv, table3_nanoparticle_specs.csv; (d) pitch/brand — business_plan.md, company_brand_brief.md, Tolera pitch decks; (e) GPU audit — gpu_stalled_tasks_costs.csv, gpu_stalled_tasks_report.md. Produce a short internal notes file capturing thresholds, tool chains, and output schemas to feed skill drafting."
            },
            {
              "title": "Author three protein-engineering skills",
              "description": "Draft via host.skills.edit: (1) binder-design-campaign — orchestrates target-epitope selection → backbone generation → ProteinMPNN/SolubleMPNN sequence design → AlphaFold2/Boltz/Chai fold-back validation → ranking/filtering → standardized design + liability CSVs; composes the existing model skills rather than reimplementing them; kernel.py with ranking/filter helpers and the output schema. (2) binder-developability-hardening — sequence-liability scan (deamidation/isomerization/oxidation/N-glyc/aggregation/unpaired-Cys), gut-restricted protease-site mapping, and hardening mutation proposal; kernel.py with liability-motif scanners. (3) multivalent-display-design — payload→scaffold/fusion architecture and valency/assembly scenario design with an assembly model. Each SKILL.md has a crisp description, numbered workflow, and worked EoE example; validate kernel.py via the edit sidecar gate."
            },
            {
              "title": "Author two program-operations skills",
              "description": "Draft via host.skills.edit: (4) biotech-pitch-and-brand — investor-audience deliverable: company naming, brand brief, and a VC pitch-deck narrative (.pptx) that pulls numbers from an existing market report + dossier; complements scientific-poster (conference audience). (5) compute-cost-audit — enumerate stalled/planned GPU tasks, estimate per-model runtime and cost, and prioritize against a stated budget, emitting a costs CSV + summary report + a cost bar chart (using figure-style). Each SKILL.md with description, workflow, and a worked example; kernel.py helpers where they earn their place (e.g. cost-estimate table builder)."
            },
            {
              "title": "Create PROTEIN_ENGINEER specialist",
              "description": "Create the UPPER_SNAKE profile via host.agents.create as a structure-design bench counterpart to INFLAM_TARGET_ANALYST. system_prompt: identity as an in-silico protein engineer that takes a target+epitope to designed, validated, developable binders/scaffolds, with explicit design-validation discipline (fold-back thresholds, developability gates, 'in-silico is not experimentally validated' honesty). Ask the user whether it should have full catalog access or a curated subset bundling the three protein skills + the model skills (proteinmpnn/solublempnn/ligandmpnn/alphafold2/boltz/chai1) + remote-compute-modal before finalizing loadout."
            },
            {
              "title": "Publish skills to the live catalog",
              "description": "Run host.skills.publish on all five new skills so they are immediately usable in this project and visible to the specialist and main agent. Read back host.skills.list() to confirm each shows as published, and confirm the specialist resolves its bundled skills."
            },
            {
              "title": "Publish to public GitHub and verify",
              "description": "Using the configured GitHub token (never written to any file), create a public repository organized as a single well-structured monorepo (proposed: one repo with one directory per skill, an AGENTS/ or specialists/ dir holding the PROTEIN_ENGINEER profile JSON, and a top-level README indexing all six items with install/usage notes and provenance to the EoE hackathon project). Commit the SKILL.md + kernel.py for each skill and the specialist definition, push to the default branch, and verify the repo is public and the files are present via the GitHub API. Report the repo URL. If repo-creation scope or account/org choice is ambiguous at push time, confirm with the user before creating."
            }
          ]
        }
      ],
      "id": "phase-0"
    }
  ],
  "feasibility": {
    "rationale": "All six items are reconstructable from existing project frames and artifacts (CCL26/POSTN + celiac binder campaigns, nanoparticle-pMHC design, pitch/brand deliverables, GPU cost audit). Authoring via host.skills SDK and publishing via the configured GitHub token are both available. Main uncertainty is GitHub org/account naming and repo-creation permissions, which I'll confirm at push time.",
    "confidence": "high"
  }
}