gioviat

gioviat/research-toolkit

Plugin Claude Code

Marketplace for the research-toolkit plugin: ML research skills and agents.

gioviat/research-toolkit

Plugin Claude Code

Skills and agents for the ML research pipeline: literature review, research ideation, math verification, experiment design, honest results reporting, and paper writing.

gioviat/research-toolkit

Agent

Surveys published work on a topic — searches arXiv, Semantic Scholar, and the web, verifies every citation exists, and returns a structured comparison with gaps and open problems. Use for SOTA overviews, related-work sections, and novelty checks on a research idea.

gioviat/research-toolkit

Agent

Edits drafts for AI-writing patterns and unclear exposition per the scientific-prose skill. Use as a final pass on any section before it's considered done. Returns a diff-style list of changes with reasons, not just a rewritten document.

gioviat/research-toolkit

Agent

Reviews research idea docs, experiment designs, and results drafts adversarially, looking for unchecked novelty claims, confounds, weak baselines, cherry-picking, and claims not supported by the reported evidence. Use before committing to a research idea, before running an expensive experiment, or before finalizing…

gioviat/research-toolkit

Agent

Checks that every numeric claim in a results draft traces back to a file in experiments/results/, that variation across seeds is reported, and that the stated statistical test matches what the data supports. Use before sharing any results draft.

gioviat/research-toolkit

Agent

Independently verifies mathematical derivations, proofs, and bounds. Use proactively whenever a derivation is added or changed, or when explicitly asked to check the math in a section. Returns a verdict per claim (correct / incorrect / unverified) with reasoning, never a general impression.

gioviat/research-toolkit

Hook

Runs after a tool call finishes for Write and Edit tool calls, executing check-claims.py via python3. From gioviat/research-toolkit.

gioviat/research-toolkit

Skill Claude CodeCodex

Use when designing a new experiment, planning an ablation, choosing baselines or metrics, or deciding how many seeds/runs are needed before any code is written. Trigger on "design an experiment", "how should I test", "what's the right baseline", "how many seeds".

gioviat/research-toolkit

Skill Claude CodeCodex

Use when surveying the state of the art on a topic, gathering related work, checking whether an idea has already been done, or comparing existing methods. Trigger on "what's the SOTA", "survey the literature", "related work on", "has anyone done", "find papers on", "is this novel".

gioviat/research-toolkit

Skill Claude CodeCodex

Use whenever asked to check, verify, derive, or re-derive a mathematical claim, proof, bound, or formula, including checking someone else's derivation, checking a step in a paper draft, or verifying that code implements a formula correctly. Trigger on words like "verify", "check the proof", "derive", "is this bound…

gioviat/research-toolkit

Skill Claude CodeCodex

Use when structuring or drafting a paper or assembling sections into a submission. Trigger on "write the paper", "draft the intro", "draft the abstract", "related work section", "structure the paper". The scientific-prose skill governs the sentence level and applies simultaneously.

gioviat/research-toolkit

Skill Claude CodeCodex

Use when developing research ideas, generating candidate research questions, or evaluating whether an idea is worth pursuing. Trigger on "brainstorm research directions", "what should we try next", "is this idea worth pursuing", "research question", "next project".

gioviat/research-toolkit

Skill Claude CodeCodex

Use when writing up experimental results, a results section, a table of numbers, or any summary of what an experiment found. Trigger on "write up the results", "summarize the experiment", "draft the results section".

gioviat/research-toolkit

Skill Claude CodeCodex

Use whenever drafting or editing prose for a paper, report, README, or any written explanation of methods or results. Trigger on "write up", "draft the section", "explain this in the report", "polish this paragraph".