ai-paper-reproduction

Companion skill for AI paper reproduction that resolves paper context, references, and dependencies needed when reproducing AI research implementations.

Helper skill for README-first AI paper reproduction that scans a repository, extracts documented commands, and returns the smallest trustworthy inference, evaluation, and training plan.

Executes and audits the selected smoke test, documented inference, or evaluation command during README-first AI paper reproduction, writing standardized `repro_outputs/` evidence and patch notes.

Sub-skill for README-first AI paper reproduction that prepares a conservative conda-first environment plus checkpoint, dataset, and cache path assumptions before any run attempt.

A lane-aware skill repository for deep learning research workflows that separates trusted reproduction tasks from exploratory candidate work, shipping 11 skills with 42 test scripts across Windows and Linux for use with Claude Code, Codex, and Agent Skills.

Part of the AI research workflow skills collection, this skill analyzes research project structure and codebase to understand components, dependencies, and architecture before reproduction.

An explore-lane skill for making isolated, candidate-only code changes during deep learning research, requiring explicit researcher authorization and producing disposable outputs that don't affect trusted reproduction artifacts.

Part of the AI research workflow skills collection, this skill enables safe exploratory trial runs in a trusted lane model for reproducing and experimenting with AI research papers.

Part of the AI research workflow skills collection, this skill safely diagnoses tracebacks and failed training or inference runs with safety guardrails to prevent data loss.

Part of the AI research workflow skills collection, this skill manages training runs with safety guardrails in a trusted lane model for reproducing AI research experiments.