Nitro
Community@nitrogen216 · Edmonton
Nitro maintains a 97-skill autonomous ML research registry covering literature discovery, GPU experiment orchestration, LaTeX paper production, patent drafting, and integrity auditing.
Agent Skills by Nitro
Showing 97 vetted skills indexed across 2 GitHub repositories.
Nitrogen216paper-write
Drafts modular LaTeX research papers from verified evidence plans and citations.
Nitrogen216askgpt-governor
Gate and submit Oracle consultations for unresolved major scientific decisions.
Nitrogen216paper-polish
Revise compiled research manuscripts through two bounded issue-driven passes and one final factual check.
Nitrogen216oracle
Executes one runtime-governed browser ChatGPT Pro consultation for a bounded major decision.
Nitrogen216evidence-audit
Audit research evidence for integrity, reproducibility, and claim support before promotion.
Nitrogen216experiment-plan
Plan screening and confirmation training experiments with precommitted metrics, budgets, and decision rules.
Nitrogen216experiment-monitor
Inspect background experiment job status and collect completion signals without altering research state.
Nitrogen216discover-research-idea
Converts a broad research direction into one literature-grounded, experimentally testable idea.
Nitrogen216literature-research
Searches and deep-reads academic papers to extract transferable mechanisms for research pipelines.
Nitrogen216improve-existing-project
Enters an autonomous research pipeline from an existing runnable project to improve it through verified experiments.
Nitrogen216experiment-run
Collects experiment results and routes screening or confirmation outcomes through the research pipeline.
Nitrogen216paper-compile
Compile LaTeX manuscripts, repair build failures, and verify the rendered PDF.
Nitrogen216contribution-hypothesis-map
Maps literature contributions to baseline failures and ranks testable adaptation candidates.
Nitrogen216paper-figure
Generate reproducible manuscript plots, tables, and editable diagrams from raw result artifacts.
Nitrogen216paper-plan
Builds a source-grounded paper plan with claims-evidence matrix and related-work positioning.
Nitrogen216experiment-bridge
Execute research experiment plans through DSH background jobs with run tracking.
Nitrogen216promotion-review
Audits confirmed research evidence and decides whether to freeze a candidate method.
Nitrogen216baseline-assessment
Establish the runnable baseline, evaluation protocol, and target metrics before research changes.
Nitrogen216minimal-adaptation
Implements the smallest literature-grounded code change for one selected research candidate.
Nitrogen216paper-writing
Orchestrates evidence-bounded research paper writing from promoted results into a compiled LaTeX manuscript.
Nitrogen216depth-research-loop
Orchestrates an autonomous performance-research pipeline from baseline through promotion and paper handoff.
Nitrogen216paper-poster-html
Generates print-ready academic conference posters as HTML/CSS with measurement-driven quality gates.
Nitrogen216meta-apply
Applies staged self-modification patches to a skill corpus after cross-model jury review and human approval.
Nitrogen216proof-checker
Verifies and fixes LaTeX mathematical proofs via cross-model adversarial review with audit reports.
Frequently Asked Questions About Nitro
FAQPage SchemaWhat tasks can I accomplish with Nitro's skill registry?▼
You can run the full ML research lifecycle: discover and novelty-check ideas, search arXiv/Semantic Scholar/OpenAlex literature, plan and launch GPU experiments, audit evidence and citations, write and compile LaTeX papers, build posters and slides, draft patents and grant proposals, and verify mathematical proofs.
Who is the target user for these skills?▼
ML researchers, PhD students, and research engineers preparing NeurIPS/ICML/ICLR/OSDI submissions, plus inventors drafting CN/US/EP patent filings. Chinese-language triggers are supported throughout, and robotics/embodied-AI and communications-domain variants exist for specialized literature and idea discovery.
How do the experiment and compute skills run in practice?▼
Skills like run-experiment, experiment-queue, and experiment-bridge launch training on local machines, SSH servers, Vast.ai rentals, Modal serverless GPUs, or the Qizhi platform via qzcli. Monitoring skills poll WandB metrics and job status, then route completed results into claim evaluation and ablation planning.
What external services and dependencies do the skills require?▼
Many skills call a Codex MCP backend for cross-model review, with optional MiniMax, Gemini, or OpenAI-compatible endpoints. Literature skills use arXiv, Semantic Scholar, OpenAlex, DeepXiv, and Exa. Paper skills need a LaTeX toolchain; posters use headless Chromium; notifications use Feishu/Lark webhooks.
How do the skills ensure research integrity before submission?▼
Dedicated audit skills verify paper-to-evidence fidelity: paper-claim-audit checks every number against raw results, citation-audit validates bibliography entries, experiment-audit detects fabricated results, integrity-forensics produces a policy gate, and kill-argument runs adversarial reviewer simulation before submission.