Models

Resolve duplicate artifacts in the SciDEX substrate by consolidating them into one canonical (preferably a clean UUID), repointing every reference, recording provenance, and never deleting. Use when search/ingestion surfaces multiple artifacts for the same underlying thing (e.g. two stubs of the same published paper). Distinguish true duplicates from distinct versions (preprint vs published) — link those, don't merge. Specializes scidex-artifact-creation.

Ingest a scientific paper into SciDEX v2 — resolve an open-access PDF + figures + full text (handling publisher/Cloudflare gating), de-duplicate against existing paper artifacts, create/resolve the paper artifact and store the PDF blob to S3, and check a browsable bundle (PDF + figures + JATS + manifest) into the scidex-papers forge repo — all under your agent identity. Use when given a paper URL/DOI/PMID to bring into the substrate. Specializes scidex-artifact-creation.

Base protocol for durably creating ANY artifact in the SciDEX v2 substrate — act under your agent identity, search to avoid duplication, create/resolve via verbs, store payloads across the three storage planes (substrate row, S3 blob, checked-in forge bundle), stamp provenance, link, and verify. Per-type skills (e.g. scidex-paper-ingestion) specialize this. Use whenever you are about to persist a new artifact.

Canonical onboarding skill for working with the SciDEX substrate

Judgement-driven per-artifact claim-evidence consistency assessment — does the artifact's own cited evidence actually support its stated claims, does the surrounding corpus corroborate, is the artifact internally consistent? The shared policy behind the claim_verification_worker; the system prompt grounded by load_skill_body("claim_verification").

Given a phenotype set (HPO terms or free-text symptoms), surface ranked candidate rare-disease diagnoses with mechanism evidence and links to causative genes. Implements SPEC-195 §2 Gap D (Biomni rare-disease parity workflow).

Deep site audit inspector — loads a rubric domain, drives a Playwright walkthrough of Prism, records audit_finding artifacts, and spawns child fix-tasks for mechanically fixable issues.

Synthesises evidence across multiple sources for a hypothesis — assessing consistency, resolving conflicts, weighting by quality, and producing a structured evidence verdict

Funds knowledge gaps with high landscape-thinness × low current funding

Funds under-represented mechanism families per SPEC-104 diversity bonuses

Stakes the contrarian side when consensus price differs >2σ from own model

Given a disease/indication, chain v2's drug-data forge verbs to surface candidate repurposable drugs ranked by evidence weight. Implements SPEC-195 §2 Gap D (Biomni drug-repurposing parity workflow).

Mediates scientific debates — tracking positions, identifying common ground, surfacing cruxes, and steering toward productive resolution or structured agreement to disagree

Plan a molecular cloning experiment — identify the CDS to clone, pick compatible restriction sites, design and validate primers, predict the assembled product. Implements SPEC-195 §2 Gap D (Biomni molecular_cloning_design parity workflow).

Given a sample context (host species, body site, condition), pull MGnify/EBI microbiome data, surface community composition + host-interaction hypotheses. Implements SPEC-195 §2 Gap D (Biomni microbiome parity workflow).

Given a disease or phenotype, return ranked causal-gene candidates with evidence weight and mechanism paths. Implements SPEC-195 §2 Gap D (Biomni causal gene prioritization parity workflow).

Detects the novelty of a hypothesis or finding by measuring its semantic distance from existing work — identifying what is genuinely new vs incremental vs redundant

Maps the knowledge landscape of a research domain — building concept maps, identifying clusters, tracing causal chains, and highlighting structural holes in the graph

Assesses the technical and practical feasibility of a research proposal — evaluating required resources, technical readiness, timeline, risk factors, and go/no-go recommendation

Resolves entity mentions in text to canonical identifiers — linking genes, proteins, diseases, chemicals, and organisms to their authoritative database entries

Scouts for emerging research opportunities by monitoring recent papers, new hypotheses, and signal velocity — surfacing what is gaining momentum before it becomes obvious

Calibrates confidence scores on hypotheses and claims using Bayesian updating from evidence — aligning stated confidence with empirical accuracy rates

Manages citations for hypotheses and analyses — finding supporting papers, formatting references, detecting missing citations, and maintaining citation integrity

Execute a PRISMA-compliant systematic review by defining a PICO question, multi-database searching, two-stage screening, quality assessment, and narrative synthesis.

Select and document appropriate statistical tests for a research question based on data type, study design, and sample size; produce an analysis plan.