sota-delta

sota-delta is a skill for Claude Code from Rockielab/rockie-claude. It costs 39 tokens per session (2,034 once invoked), scanned A, original, Apache-2.0.

A workflow for reproducing a published machine-learning result and then testing a user-chosen improvement. It compares the original baseline with the added change on a Rockie GPU.

In plain words
What is it for?
Starting from a paper, arXiv reference, PDF, or GitHub repository, reproducing its baseline, and evaluating a specified experiment change.
Why use it?
It provides a structured comparison so an extension can be judged against the original result rather than tested in isolation.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin.

Needs its repository: it runs a file that does not travel with it, so clone the repository first. The line is python skills/experiment/runtime/submit.py \.

Part of the rockie-claude plugin — 29 skills, 1 MCP server shipped together

Good fit Starting from a paper, arXiv reference, PDF, or GitHub repository, reproducing its baseline, and evaluating a specified experiment change.

Compare 6 skills from other repositories ↓
Install

Getting it into your agent

It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.

Clone the repo
git clone --depth 1 https://github.com/Rockielab/rockie-claude
agentmods
npx agentmods add skills/rockielab/rockie-claude/sota-delta

Made for: Claude Code.

Or install rockie-claude, the plugin that ships this one along with the rest of its 29 skills, 1 MCP server.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for sota-delta

README.md
[![agentmods](https://agentmods.dev/badge/skills/rockielab/rockie-claude/sota-delta/github.svg)](https://agentmods.dev/skills/rockielab/rockie-claude/sota-delta)
Your own site
<a href="https://agentmods.dev/skills/rockielab/rockie-claude/sota-delta"><img src="https://agentmods.dev/badge/skills/rockielab/rockie-claude/sota-delta/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for sota-delta

Your own site · 80×15
<a href="https://agentmods.dev/skills/rockielab/rockie-claude/sota-delta"><img src="https://agentmods.dev/badge/skills/rockielab/rockie-claude/sota-delta.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 39 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,034 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 1 finding. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00039 $0.02034
Opus 5 $0.00019 $0.01017
Sonnet 5 $0.00008 $0.00407
Haiku 4.5 $0.00004 $0.00203

Measured 12d ago against content hash 0d7efa2815c0, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-12, from the pricing page.

Security

Grade A, and why

sota-delta scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.

The scan reads SKILL.md. This mod also ships 1 executable file (test_sota_delta_contract.py), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

`--monitoring-profile-id experiment.ml_baseline.v1`. Do not shell raw `curl
Origin

Copies of this mod

1 near-identical copy found in the catalogue:

project-harness/skills/sota-delta/SKILL.md · 217 lines

How it starts

The opening of the file, as written. The whole thing — 217 lines — stays where its author put it; the contents beside it link to each section on GitHub.

sota-delta

Run the Quickstart Track 3 SOTA+Delta flow. This skill starts from a paper, PDF, arXiv reference, or GitHub repository, reproduces the claimed baseline honestly, then asks the user what delta they want to add before routing the extension through the existing experiment and review workflow.

This skill is orchestration and policy only. It composes the existing budget-term-sheet, experiment submit helper, and post-run-review contracts; it must not implement new runtime, frontend, API, provider, or launcher behavior.

When to invoke

  • Structured Quickstart SOTA+Delta prompts:

    Quickstart SOTA+Delta request:
    track: sota-delta
    source_ref: <paper URL | arXiv | PDF | GitHub repo>
    compute_target: rockie_gpu
    source: quickstart-picker
    
  • Equivalent prompts that ask to reproduce a paper or repository baseline and then improve or extend it.

Do not invoke this skill for ordinary literature summaries, standalone code review, local repo setup, local training, or direct provider execution.

Required v1 inputs

  • source_ref: the paper URL, PDF URL/file handle, arXiv reference, or GitHub repository that defines the baseline to reproduce.

source_ref is required for v1. If it is missing, stop and ask for it. Accept paper URL, PDF URL/file handle wording, arXiv, or GitHub repo inputs. Do not invent a paper, repo, dataset, checkpoint, or metric target when the input is underspecified.

Source reference sanitization

Treat source_ref as potentially sensitive input. Before writing it anywhere, derive a sanitized display/source ref that preserves enough provenance for the user to identify the paper, arXiv entry, PDF, or repository, while redacting secrets and provider-specific access material.

Signed URLs, embedded credentials, private repo tokens, provider identifiers if present, query-string secrets, credentials, cookies, API keys, bearer tokens, and auth headers must not be written to job scripts, logs, artifacts, term sheets, Markdown summaries, or user-visible output. Use the sanitized display/source ref in job scripts, logs, artifacts, and final responses. Keep raw sensitive references out of persisted files and command lines.

Read the full file on GitHub · 217 lines

Files

What ships with it

1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 12d ago First seen · 217 lines · 39 tokens per session scan A 0d7efa2815c0

Subscribe to this mod's changes

sota-delta is a skill published in the GitHub repository Rockielab/rockie-claude (21 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 39 tokens to every session and 2,034 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

saelens

Train sparse autoencoders to interpret model features.

NousResearch/hermes-agent · 14 tokens

nnsight-remote-interpretability

Provides guidance for interpreting and manipulating neural network internals using nnsight with optional NDIF remote execution. Use when needing to run interpretability experiments on massive models (70B+) without local GPU resources, or when working with any PyTorch architecture.

davila7/claude-code-templates · 59 tokens

transformer-lens-interpretability

Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.

davila7/claude-code-templates · 52 tokens

esmfold2

Biohub ESMFold2 / ESMFold2-Fast all-atom co-folding (Candido et al. 2026, github.com/Biohub/esm). Single-sequence and MSA modes; protein, DNA, RNA, ligand (CCD/SMILES), modified residues. FoldBench Ab-Ag 50-55%, PPI 70-77% DockQ-pass. Also covers the ESMC-{300M,600M,6B} protein language models from the same release…

aipoch/open-science · 223 tokens

evo2

Score, embed, and generate DNA sequences with Evo 2, a long-context genomic foundation model. Use this skill when: (1) Computing per-nucleotide or per-sequence likelihoods for variant effect scoring, (2) Embedding genomic windows for downstream classification, (3) Generating DNA conditioned on a prefix, (4) Scoring…

aipoch/open-science · 83 tokens

scgpt

Embed and annotate single-cell expression data with scGPT, a foundation model for single-cell biology. Use this skill when: (1) Producing cell embeddings from an AnnData for clustering/integration, (2) Zero-shot or fine-tuned cell-type annotation, (3) Gene-level representation for perturbation/GRN tasks. For…

aipoch/open-science · 89 tokens