skills.rest
kridaydave
kridaydaveCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

anchor

Pin environments, seeds, and data versions to build reproducible replication packages.

1|Updated Jun 1, 2026
Search Tags:#docker#reproducibility#lockfile#data versioning#artifact evaluation#replication package#random seed
One-click install
npx skills add https://github.com/kridaydave/My_Skills --skill anchor-kridaydave
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: anchor
Source: https://github.com/kridaydave/My_Skills/tree/main/anchor
Command: npx skills add https://github.com/kridaydave/My_Skills --skill anchor-kridaydave

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Results that run on one machine often fail elsewhere due to unpinned dependencies, unset random seeds, drifting data sources, and undocumented manual steps. Anchor audits these failure layers and produces a replication package that a reviewer or teammate can actually rerun. ## Core Features & Use Cases - Reproducibility Audit: Systematically checks environment pinning, randomness control, data versioning, and run steps, reporting risks and fixes per layer. - Replication Packaging: Produces lockfiles or Docker/conda environments, seeded run harnesses, data snapshots with hashes, a one-command run path, and a README with expected output and tolerance. - Use Case: Before submitting code with a paper for artifact evaluation, use Anchor to pin torch and CUDA dependencies, set and record seeds, snapshot the dataset with a content hash, and smoke-test a clean rerun from a fresh clone. ## Quick Start Ask Anchor to audit your project for reproducibility and build a replication package with pinned dependencies, a recorded seed, and versioned data.

Frequently Asked Questions about anchor

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make my machine learning experiment reproducible?▼

Pin every dependency to an exact version in a lockfile, set and record all RNG seeds, snapshot and hash your dataset, and document one command from clean clone to result. Then verify by rerunning from scratch in the pinned environment.

How do I pin Python dependencies for reproducibility?▼

Generate a lockfile with exact versions from the working environment rather than using version ranges like pandas>=1.0. Ranges resolve to different packages on each install, while a lockfile guarantees the same dependency set everywhere.

Why do I get different results with the same seed on GPU?▼

Some CUDA operations such as atomic reductions are nondeterministic by default, so a seed alone is insufficient. Enable deterministic algorithms and set CUBLAS_WORKSPACE_CONFIG for bit-identical output, or report mean and standard deviation over multiple runs.

What should a replication package for a paper include?▼

Include a pinned environment such as a Dockerfile or lockfile, the dataset with its version or content hash and license note, a single documented run command, the expected output with a tolerance for matching, and a README covering prerequisites and a smoke test.

When is a full Docker-based replication package overkill?▼

For internal one-off analyses, a requirements lockfile, a recorded seed, and a hash of the input data usually suffice. Reserve Docker images and data DOIs for artifacts submitted with papers or handed off externally.

Related Skills

Explore More Education & Research Skills→
danilomongedanilomonge
1

hlatyping

Run the nf-core/hlatyping workflow on samplesheet-based DNA or RNA sequencing data.

Community
Advanced
Elijah-JElijah-J

ar-closeout

Summarize perf autoresearch results, verify final state, and archive artifacts for handoff.

Community
Advanced
ttsmith21ttsmith21

qa

Run headless SolidWorks add-in test suites and report pass/fail results.

Community
Intermediate
wildwasserwildwasser
35

ml-experiment

Track machine learning experiments with structured logging and reproducibility.

Community
Advanced