From lllllllama-ai-paper-reproduction-skill
Resolves narrow reproduction gaps in AI repos by extracting specific paper details like dataset splits, preprocessing, evaluation protocols, checkpoint mapping, or runtime assumptions when READMEs lack them.
npx claudepluginhub lllllllama/ai-research-workflow-skillsThis skill uses the workspace's default tool permissions.
- README and repo files leave a reproduction-critical gap.
Orchestrates trustworthy, README-first reproduction of AI paper repositories: selects minimal inference/eval targets, coordinates intake/setup/execution/training/analysis, enforces conservative patches, records evidence/deviations, and generates standardized repro_outputs/ bundle.
Scans Python ML experiment repos to generate hierarchical paper outlines (H1/H2/H3) with user approval checkpoints at each level, then body text with evidence annotations, citations, and bilingual output.
Converts Arxiv ML papers to citation-anchored Python implementations with ambiguity audits, in PyTorch/JAX/TensorFlow, producing structured project directories.
Share bugs, ideas, or general feedback.
Use references/paper-assisted-reproduction.md.