minimal-run-and-audit
🎯SkillA skill that runs one short check command when you re-run a deep learning research project and saves the results as standard report files. It also records any edits made to the project's files.
Installation
npx skills add lllllllama/rigorpilot-skills --skill minimal-run-and-auditRun it in a terminal; it asks which AI coding tools to install it for (Claude Code, Codex, Cursor and more).
Uses skills.sh · Node.js 22.20+
After installing
Ask for related work and the agent uses it on its own, or call it yourself.
/minimal-run-and-auditInstall by hand
Copy the whole skill folder from GitHub to the place below. Copying only SKILL.md can leave out files the skill uses.
Open the skill folder →- This project
.claude/skills/minimal-run-and-audit/ - All projects
~/.claude/skills/minimal-run-and-audit/
These paths are for Claude Code. For other tools, the command above puts it in the right place.
What this skill does
Overview
minimal-run-and-audit is the Rigor Run skill for reproducing deep learning projects, which means re-running published research code to check its results. It runs a command you already picked, such as a smoke test (a quick run to see if the code works), an inference run (using a trained model to get results), or an evaluation run, and records what happened. It does not run model training.
Key Features
- Save a summary of the run and standard report files in the
repro_outputs/folder. - Get
SCIENTIFIC_CHANGELOG.md, which lists any change that could affect what the results mean. - Get
COMPARABILITY_REPORT.md, which shows how well your results can be compared with the README, the paper, and baseline results. - See clearly whether each result is verified, partial, or blocked.
- Get
PATCHES.md, which lists the edits whenever project files were changed.
Who is this for?
This skill is for researchers and engineers who re-run published deep learning code and need an honest record of each run that others can trace. Use it when you already know which command to try and only need it run and reported, not for training, a first look at the repository, or choosing what to reproduce.
Summarized by AI from this skill's SKILL.md. Suggest a better one
SKILL.md
View on GitHub →Rigor Run skill for README-first deep learning repo reproduction. Use when the task is specifically to capture or normalize evidence from the selected smoke test or documented inference or evaluation command and write standardized `repro_outputs/` files, including patch notes when repository files changed. Do not use for training execution, initial repo intake, generic environment setup, paper lookup, target selection, hidden scientific-meaning changes, or end-to-end orchestration by itself.
minimal-run-and-audit
Use this as the Rigor Run skill. The installed slug remains
minimal-run-and-audit for compatibility.
Use the shared operating principles in
../ai-research-reproduction/references/agent-operating-principles.md; this skill should make run
evidence auditable without turning every command into a rigid protocol.
When to apply
- After a reproduction target and setup plan exist.
- When the main skill needs execution evidence and normalized outputs.
- When a smoke test, documented inference run, documented evaluation run, or other short non-training verification is appropriate.
- When the user already knows what command should be attempted and wants execution plus reporting only.
When not to apply
- During initial repo scanning.
- When environment or assets are still undefined enough to make execution meaningless.
- When the task is a literature lookup rather than repository execution.
- When the user is still deciding which reproduction target should count as the main run.
Clear boundaries
- This skill owns normalized reporting for an attempted command.
- It may receive execution evidence from the main skill or a thin helper.
- It does not choose the overall target on its own.
- It does not perform broad paper analysis.
- It does not own training startup, resume, or long-running training state.
- It should not normalize risky code edits into acceptable practice.
- It must not hide changes that alter evaluation, preprocessing, checkpoints, metrics, or other scientific meaning.
Input expectations
- selected reproduction goal
- runnable commands or smoke commands
- environment and asset assumptions
- optional patch metadata
Output expectations
- execution result summary
- standardized
repro_outputs/files SCIENTIFIC_CHANGELOG.mdfor changed scientific meaning and evidence statusCOMPARABILITY_REPORT.mdfor README/paper/baseline comparability- clear distinction between verified, partial, and blocked states
PATCHES.mdwhen repo files changed
Notes
Use references/reporting-policy.md, ../ai-research-reproduction/references/research-rigor-principles.md, scripts/run_command.py, and scripts/write_outputs.py.
More from this repository10
A helper skill for rerunning a deep learning project from its README. When the README misses a key detail, it finds the answer in the original paper and notes where the two disagree.
A skill that reads a deep learning project's README and main files before you run anything and lists the commands they describe. It then suggests the smallest run you can trust to check the results.
A skill that plans the software setup and file locations you need before re-running a deep learning research project. It suggests install commands and where to put model and data files.
This skill helps deep learning researchers test new ideas on top of an existing project, with fair comparisons against a fixed benchmark. It marks each result as a candidate to check and saves the reports in `explore_outputs/`.
A skill that reads a deep learning research project and explains how it is built, without changing any code. It also points out code that looks suspicious so you can check it.
A skill that helps your agent rerun a deep learning research project and check whether it works as its README describes. It starts with the smallest test and records every step and change.
A skill that finds the cause of errors in deep learning research code, such as running out of GPU memory, before changing anything. It suggests the smallest fix and waits for your approval.
A skill that helps you try small, low-risk code changes in a deep learning research project on a separate branch. It records why each change matters and how to undo it.
A skill that carefully runs a training command you already chose for a deep learning project. It saves the command, settings, logs, and results in one standard folder.
A skill that plans quick, small trial runs for deep learning research, such as testing on part of the data or trying many settings. It reports the results without claiming more than they show.