ai-scaffolding

Invoke Anthropic's Claude Code CLI from Codex for external reviews, second opinions, and file-based analysis. Use when the user asks to call Claude or Claude Code, compare Codex with Claude, run a quick one-shot prompt through Claude, or launch a background Claude review over many files, diffs, or a prepared context file.

Triggers on `/council-of-bots` or when the user wants multiple AI agents to review code, diffs, documents, or technical questions. Gather outside perspectives from Codex, Gemini, and Claude, then synthesize them. Supports both Claude Code and Codex by using a shared review protocol with host-specific launch adapters.

Capture and persist lessons learned from a session to compound knowledge over time. Triggers on "/lessons-learned", "what did we learn", "save lessons", "update skills with what we learned", or at the end of a complex multi-session task. PROACTIVE USE: This skill should also be suggested or invoked (1) when resuming from context compaction (the previous context likely contained unrecorded lessons), (2) after resolving a non-trivial bug or debugging session, (3) after significant friction or failed approaches that yielded insight, (4) after a council-of-bots review that surfaced fixes. Identifies reusable patterns, bug fixes, workflow insights, and tool quirks, then persists them to the right places: auto-memory (project-specific), skill files (reusable across projects), or both.

Manage R-based HPC workflows on the LRZ Linux Cluster (CoolMUC-4) via SSH from the user's local machine. Use this skill whenever the user wants to run R scripts on the LRZ cluster, submit SLURM jobs, install R packages on LRZ, monitor running jobs, fetch results, or set up their LRZ compute environment. Also trigger when the user mentions CoolMUC, LRZ HPC, parallelized R, mclapply on a cluster, SLURM + R, or remote HPC job management. This skill assumes Claude Code is running locally and reaches LRZ through a pre-configured SSH multiplexed connection.

Finalize R packages for CRAN submission. Use when the user wants to: submit a package to CRAN, prepare for CRAN release, run pre-submission checks, do reverse dependency checks (revdep), run cross-platform checks via rhub, verify pkgdown builds, review CRAN policies compliance, or says "/cran-submission". Covers the full workflow from pre-flight checks through submission and post-acceptance steps. Complements r-package-coding (development) with release-specific procedures.

Edit prose to remove LLM/AI-writing tells (inflated vocabulary, formulaic phrases, vague authority, chatbot scaffolding, mechanical triads, and formatting tics) without changing content, numbers, citations, or technical meaning. Use when asked to de-slop, humanize, remove AI tells, or check a draft for chatbot phrasing, and proactively before finalizing a paper section, peer review, Gutachten, grant, teaching material, or email. For scientific manuscripts, also audit claim–evidence alignment, uncertainty, causal scope, reproducibility detail, and reporting completeness without inventing missing science. Preserve load-bearing domain terms such as robust, sandwich, leverage, penalized, coverage, efficient, significant, and power.

Create comprehensive quiz questions in R/exams format for FMM course slide sets. Use when the user asks to: create quiz questions, make Moodle quizzes, generate R/exams questions, write exam questions for a slide set, or says "/make-quiz". Produces schoice, mchoice, and num questions at mixed difficulty levels. Output renders to Moodle XML via exams2moodle(). Questions are in German.

R package development standards for AI coding agents. Use when writing R package code (not scripts/analysis), implementing S3 objects, writing roxygen2 documentation, or working with checkmate/cli/purrr/vctrs. Covers naming conventions, pure functions, validation patterns, and package workflow.

R coding standards for scripts and data analysis. Use for writing R scripts, data analysis, tidyverse workflows, or exploratory analysis. Includes guidance for creating teaching demos with knitr::spin (literate R scripts that render to PDF/HTML). Not for package development.

Review code and data supplements of scientific papers for computational reproducibility. Use when asked to: review a code supplement, check if a paper's code is reproducible, audit a simulation study, evaluate a scientific paper's data and code, or assess computational reproducibility. Actively executes code (fixing minor issues), runs reduced simulations, and compares outputs against reported results. Outputs a structured markdown review document.

Monte Carlo experiment design for comparing statistical methods. Use when the user asks to "set up a benchmark", "create a simulation study", "compare methods across DGPs", "design a Monte Carlo experiment", "run a coverage study", or needs guidance on: (1) benchmark architecture and data flow, (2) parallelization with furrr/future, (3) truth function design and constraint matching, (4) seed management and reproducibility, (5) performance measures and Monte Carlo SEs. R-specific with tidyverse/furrr patterns. Implements ADEMP framework (Morris et al. 2019).

Generate rigorous, evidence-based peer reviews for statistical and methodological papers. Use when asked to review a manuscript, evaluate a statistical paper, write referee comments, or assess methodological research. Produces structured reviews with Summary, Major Issues, and Minor Issues sections. Includes specialized checklists for evaluating simulation studies (Morris et al. 2019) and code/data supplements. Always outputs reviews using the provided Quarto template.