Chenbo Yu
Community@yyccbb · Singapore
Yu Chenbo \\ Computer Science and Statistics undergraduate student @ National University of Singapore
Agent Skills by Chenbo Yu
Showing 10 vetted skills indexed across 1 GitHub repositories.
yyccbbexperiment-scoping
Plans paper reproduction experiments by reconciling a paper PDF with its codebase's argparse entry points.
yyccbbenvironment-setup
Builds and repairs conda environments for research codebases with pinned lockfiles and verification reports.
yyccbbbaseline-reproduction
Orchestrates reproduction of baseline comparison methods in cloned sibling repositories.
yyccbbrun-experiment
Executes full-length paper reproduction runs sequentially and records structured evidence per run.
yyccbbrun-visible
Launch long ML training runs in a visible terminal and monitor progress without blocking the agent.
yyccbbresource-download
Verifies hardware feasibility and downloads datasets and checkpoints for planned ML experiments.
yyccbbresult-analysis
Extracts reproduction metrics with auditable scripts and compares them against paper tables and figures.
yyccbbreproduction-audit
Audits paper-reproduction pipeline reports and issues per-experiment reproduction verdicts.
yyccbbpaper-reproduction
Orchestrates a seven-stage pipeline that reproduces machine learning paper results from cloned codebases.
yyccbbrun-validation
Runs mock experiment commands and fixes code errors in a git-tracked loop until each passes.
Frequently Asked Questions About Chenbo Yu
FAQPage SchemaWhat tasks can I accomplish with yyccbb's paper-reproduction skills?▼
You can reproduce ML papers end to end: scope experiments from the PDF, download datasets and checkpoints, build conda environments, validate mock runs with git-tracked fixes, launch full-length training runs, reproduce baseline comparisons, regenerate paper tables from actual logs, and audit the evidence chain for a per-experiment verdict.
Who are these reproduction pipeline skills designed for?▼
ML researchers, graduate students, and reproducibility engineers who need to verify published results. The skills assume familiarity with research codebases, GPU training, and hyperparameter configs, and suit anyone validating a paper's claimed numbers against actual runs.
How do I run the paper-reproduction pipeline in practice?▼
Clone the paper's codebase, place the paper PDF inside it, then invoke the paper-reproduction driver. It runs experiment-scoping, resource-download, environment-setup, run-validation, run-experiment, optional baseline-reproduction, result-analysis, and reproduction-audit sequentially, gating each stage on the prior stage's report.
What are the prerequisites for using these skills?▼
You need a cloned paper repository containing the paper's PDF, plus hardware with sufficient VRAM, RAM, and disk for the planned experiments, which the resource-download stage checks. Individual stages require the earlier stages' reports written under the .paper-reproduction/ directory.
Can I run a single stage instead of the full pipeline?▼
Yes. Each stage skill, such as environment-setup or run-experiment, works standalone when its prerequisite reports exist under .paper-reproduction/. The paper-reproduction driver is only for full end-to-end runs and explicitly skips work already recorded by earlier stages.