This file maps EvidenceForge skills to the methods sources they should draw from.
| Skill | Main job | Primary sources | Key guardrail |
|---|---|---|---|
evidence-synthesis-forge |
Orchestrate review design from question to protocol/report | PRISMA 2020, Cochrane Handbook, JBI Manual | Do not jump to meta-analysis before defining question, eligibility, search, screening, extraction, appraisal, and traceable PRISMA counts |
meta-analysis-forge |
First-order statistical synthesis, IPD meta-analysis planning, and mega-analysis audit | Cochrane Handbook, Borenstein et al., Hedges & Olkin, Harrer et al.; Bayesian/IPD synthesis traditions; ecology/evolution stock-versus-flux examples such as ant-mediated soil-carbon meta-analysis | Use validators and machine-readable coding sheets before running scripts; do not pool incompatible outcomes/effect metrics, ignore dependence, collapse stock and flux into one endpoint, or call a project mega-analysis without harmonized data reprocessing/remodeling |
umbrella-review-skeptic |
Review of reviews, umbrella review, second-order meta-analysis, temporal review-level meta-regression | JBI umbrella review methods, Cochrane overviews, AMSTAR 2, ROBIS, agricultural diversification second-order meta examples | Do not double-count primary studies or treat overlapping reviews as independent; do not interpret review-level duration trends as primary-study causal effects without overlap and quality checks |
meta-ml-screener |
ML-assisted screening/extraction/classification | ASReview, Rayyan, AI-assisted review literature | Keep human-in-the-loop decisions and export machine-readable audit logs |
environment-life-review-forge |
Environmental/ecological/life-science adaptation, including PLS/VIP environmental indicator audits, ecosystem-service relationship threshold ML audits, air-quality food-security co-benefit audits, biodiversity-stability climate-stress audits, soil-fauna carbon meta-analysis, wetland methane scaling, cryosphere ground-ice map-product audits, food-system nexus reviews, food-waste geospatial ML forecasting, agroecosystem nutrient meta-analysis, crop-yield ML prediction, environmental causal ML, urban heat DML, agricultural irrigation optimization, scenario-model evidence synthesis, and land-use optimization trade-offs | CEE standards, JBI, ecology/evolution meta-analysis handbook, PLS/VIP environmental indicator modeling, ecosystem-service relationship and threshold-oriented ML examples, air-pollution crop-productivity examples, soil biodiversity and ecosystem-stability examples, soil-fauna SOC-plus-CO2 meta-analysis examples, wetland methane budgets, cryosphere and permafrost map products, food-system nexus, food-waste geospatial ML, environmental nutrient-meta, crop-yield ML, environmental causal ML/DML examples, irrigation optimization examples, scenario-model, and Pareto-frontier examples | Do not pool across incompatible species, exposures, endpoints, geography, measurement platforms, soil-depth conventions, map-depth conventions, spatial resolution, permafrost masks, durations, scale classes, causal estimands, or policy scenarios; separate stock, flux, stability, footprint, feedback, predictive-model, PLS/VIP screening, ESR threshold heuristic, air-quality counterfactual, biodiversity-stability moderation, map-product, causal-estimator, optimization-frontier, methane-budget, and policy-feasibility claims |
When adding a new skill, include:
- trigger conditions;
- sources it aligns with;
- workflow stages;
- output artifacts;
- guardrails;
- at least one template or structured output.
EmpiriForge uses the reproducibility paper to design skills for primary empirical research. EvidenceForge uses the same architecture for secondary research and evidence synthesis.
docs/knowledge-graph-navigation.md documents Graphify as an optional navigation layer for large repositories and literature folders. It can help locate relevant files and relationships before an agent reads raw documents, but it is not a source of methodological validity, evidence certainty, causal interpretation, or risk-of-bias judgment.