lllllllama

Deep learning debugging often leads to speculative patches that break research reproducibility. This skill provides conservative diagnosis with explicit approval gates before any code modification, keeping debug fixes separate from research contributions.

Deep learning training runs often lack reproducible evidence and clear status reporting. This skill executes a selected training command conservatively and writes standardized training evidence to train_outputs/.

Manually scanning AI repositories for reproduction commands is time-consuming and error-prone. This skill automates README-first analysis to extract documented commands and generate minimal trustworthy reproduction plans.

When reproducing AI research, repository READMEs often leave critical gaps about dataset splits, evaluation protocols, or preprocessing details. This skill resolves those narrow reproduction questions from primary paper sources while preserving README-first guidance.

Running AI paper reproduction experiments requires consistent command execution and standardized reporting. This skill executes smoke tests, inference runs, or evaluation commands while automatically generating structured output bundles for audit trails.

Deep learning researchers need to run quick exploratory trials without overclaiming results. This skill generates budget-aware variant matrices with fair-comparison caveats, keeping exploratory evidence clearly separated from trusted baselines.

Deep learning researchers need to explore code modifications without breaking trusted baselines. This skill creates conservative, auditable change plans with explicit rollback paths for isolated worktrees.

Setting up AI research environments for paper reproduction is complex and error-prone. This skill automates conservative conda-first environment creation and asset path planning to reduce setup friction.

Understanding a new deep learning repository is time-consuming and error-prone. This skill provides read-only static analysis to map model structure, training entrypoints, and suspicious patterns without modifying code or running expensive training jobs.

Reproducing deep learning papers is slow and error-prone because commands, datasets, and assumptions are scattered across READMEs. This skill reads the repository first, selects the smallest documented target, and writes a standardized repro_outputs/ bundle with evidence, deviations, and human decision points.

Researchers struggle to systematically explore and rank novel deep learning ideas with scientific rigor. This skill provides auditable candidate exploration with idea gating, fair comparison, and governed experiment workflows on top of current_research.