What problem does it solve? Reproducing a machine learning paper's results is incomplete without the baseline rows in its tables and figures, but each baseline lives in a different repository with its own environment and entry points. This Skill automates that cross-repo work: it finds each baseline's official code, clones it beside the main repo, and drives the existing reproduction pipeline stages inside it, scoped to exactly the datapoints the paper's comparisons need. ## Core Features & Use Cases - Baseline manifest and scoping: Reads the main pipeline's experiment-scoping and run-experiment reports to decide which baselines and which exact table/figure datapoints must be reproduced, then writes a scope.md contract per baseline repo. - Sequential sub-pipeline orchestration: Spawns one subagent per baseline repo to run experiment-scoping, resource-download, environment-setup, run-validation, and run-experiment, reusing downloaded assets and the main conda environment where possible. - Auditable index output: Writes .paper-reproduction/baselines.md with per-baseline provenance, per-datapoint statuses, pending user actions, and comparability caveats for the result-analysis stage. - Use Case: After reproducing a paper's main method, run this Skill to also reproduce the L2X and INVASE baseline numbers for Table 1, each in its own cloned repo, without re-downloading shared datasets. ## Quick Start Reproduce the baseline methods for the experiments already completed in this paper's codebase.
