JuliaLLMAgentSkills

Build, migrate, audit, review, update, and compact agent-facing repository documentation with nested AGENTS.md routers, selectively loaded Diátaxis-style .agents context, structured V-model specifications, and bidirectional verification traceability. Use when creating documentation for a new or existing repository, migrating a monolithic AGENTS.md, defining or auditing V-model traceability, interviewing developers to recover intended behavior, adding nested AGENTS.md files, organizing agent context, reviewing .agents consistency, or compacting stale and duplicated repository context.

Build and repair Julia user interfaces: simple scripts and installable command-line apps, Comonicon-based command trees in compatible existing projects, styled terminal output and Term.jl live widgets, and custom Makie 0.24 plot recipes. Use for CLI entry points, options and subcommands, terminal reports or TUIs, package plotting extensions, and focused interface tests; use the workload skill instead for subprocess, pipeline, task, or thread mechanics.

Develop and maintain Julia packages, including Project.toml and package layout changes, dependencies and compatibility, extensions, ParallelTestRunner-first tests, existing TestItemRunner suites, Documenter docs and doctests, JET/Aqua/coverage checks, Scratch runtime data, debugging, CI, and contribution workflows. Use for Julia package repositories and files such as Project.toml, Manifest.toml, src/*.jl, ext/*.jl, test/runtests.jl, docs/make.jl, and Julia GitHub Actions.

Build and repair focused Julia data workflows that consume Tables.jl sources, transform tabular data with DataFrames.jl, read or write delimited files with CSV.jl, handle TOML or YAML configuration, and render ordinary reports with PrettyTables.jl. Use for CSV/TSV ingestion, table transformations, configuration serialization, schema validation, and terminal/Markdown/HTML table output; do not use as a general statistics, database, plotting, or machine-learning guide.

Measure and improve Julia runtime performance with BenchmarkTools, profiling and allocation tools, inference diagnosis, data-layout changes, and guarded low-level optimizations. Use when Julia code is slow, allocation-heavy, type-unstable, or needs a reproducible performance baseline or regression check.

Coordinate Julia tasks, channels, threads, external numerical-library threads, and subprocesses with explicit lifetimes and failure propagation. Use when building producer-consumer workflows, parallelizing CPU work, fixing races or deadlocks, controlling oversubscription, or automating external commands from Julia.