What problem does it solve? Clinical programmers need to derive ADaM analysis datasets (ADSL, ADVS, ADLB) from SDTM data following strict CDISC conventions, and mistakes in date imputation, flag conventions, or function selection create regulatory submission risk. This Skill encodes the shared conventions and dataset-specific workflows an experienced admiral programmer applies, so an AI agent generates QC-ready, audit-traceable R code. ## Core Features & Use Cases - Shared ADaM conventions: Enforces library setup, native pipe style, date derivation rules (derive_vars_dt, never as.Date on --DTC), flag conventions ("Y"/NA, never "N"), DOMAIN removal, and stopifnot assertions across all derivations. - Child skill routing: Routes to admiral-adsl for subject-level datasets (treatment dates, disposition, population flags) and admiral-bds for BDS findings datasets (ADVS, ADLB with baseline, change from baseline, and analysis flags). - QC-ready output: Requires # REVIEW: annotations at every protocol-specific decision point and programmatic uniqueness assertions for regulatory review. - Use Case: A clinical data programmer asks the agent to derive ADSL from DM, EX, and DS domains; the Skill produces step-by-step admiral R code with correct imputation flags, population flag logic, and review annotations aligned to the ADaM spec. ## Quick Start Ask the agent to derive an ADSL dataset from the SDTM DM, EX, and DS domains using admiral, following the study's ADaM specification.
