gaasherCommunity·25 Agent Skills Included

Agent-Loop-Skills

Self-verifying loops for research, coding, and analysis

Runs iterative, verification-gated loops that improve ML models, research proposals, scientific papers, datasets, SQL queries, prompts, and code until a real measured signal confirms each change. Eliminates blind trial-and-error and confidently wrong outputs by scoring every step against tests, metrics, judges, or ground-truth data. Covers autoresearch, literature surveys, hypothesis generation, red/blue/purple teaming, data cleanup, and end-to-end software builds with pull requests.

150|19

↗

npx skills add gaasher/Agent-Loop-Skills --all -g -y

All Skills in This Repository (25)

Pure Emerald Level Indicators

📦 In Repo

gaasher

research-question

Refine broad topics into specific research questions using rubric scoring and novelty checks.

claim-verify

Verify data-backed claims from markdown documents against datasets.

purple-team

Orchestrate red-team attacks and blue-team patches for security guardrails and APIs.

research-proposal

Evaluate and rewrite research proposals with iterative, literature-grounded feedback requiring Python 3.9+ and literature-search skill.

red-team

Generate adversarial inputs and log discrepancies between target systems and oracles.

hypothesis-gen

Generate literature-grounded research hypotheses using a multi-agent loop.

anomaly-investigation

Identify root causes of data anomalies by testing candidate causes.

scientific-writer

Coordinate expert judges, an independent grader, and a revising author to refine scientific writing.

tournament-autoresearch

Automate machine learning research with a tournament loop for architecture changes.

plan-loop

Decompose coding prompts into structured execution plans with schema validation.

blue-team

Apply patches to fix code failures from CI/CD test reports and create pull requests.

karpathy

Iterate training script changes to minimize a scalar metric.

Community

Advanced

Frequently Asked Questions

FAQPage Schema

How to install Agent-Loop-Skills?▼

Run `npx skills add gaasher/agent-loop-skills --all -g -y` in your terminal to install every loop globally for your agent host.

What do these agent loops actually do?▼

Each loop proposes one change, runs it in your environment, scores it against a real signal like tests or a metric, keeps it only if it improves, and repeats until the work is verifiably better.

Can it run autonomous ML experiments?▼

Yes. Loops like karpathy, ml-autoresearch, and tournament-autoresearch iterate on your training code against a metric, keeping only changes that measurably improve results.

Does it work with Claude Code and Cursor?▼

Yes. Every loop follows the open SKILL.md standard and runs on Claude Code, Codex, Cursor, and other compatible hosts, using real subagents where supported.

Do I need coding experience to use these loops?▼

No. You describe your task in plain English, confirm a few setup values, and the loop handles execution, verification, and logging on its own.