Stars
Continuously sync your AI setups with one command. Codebase tailor suited agent skills, MCPs and config files for Claude Code, Cursor, and Codex.
Spin up your whole herdr workspace layout — tabs, panes, commands, and all — from a single YAML file.
Char-level GPT from scratch on MLX — transformer internals, hand-written
Three Agent Skills: replayable audit logs, read-only instruments, shadow-mode validation — observation precedes intervention. The installable how-counterpart to AAP ADR-0006/0005.
NGC-Learn: Computational Neuroscience and NeuroAI in Python
Audit a project repo's non-code assets (tool configs, CI workflows, runbooks) for diminished value — Keep/Update/Retire/Merge. A semantic value layer above structural linters.
Cross-model code review — a read-only Claude Code skill wrapping the OpenAI Codex CLI for a decorrelated second opinion on a diff
Optional local MLX generation backend (Apple Silicon, mlx_lm.server) for contemplative-agent — on-device, interactive use. Mirrors contemplative-agent-cloud's LLMBackend Protocol injection (ADR-0070).
Structural skill-library debt scan (missing-artifact detection) — the deterministic code layer of the Agent Knowledge Cycle's Curate phase.
Agent Skill: five-question triage routing AI work to the right architecture (script / algorithmic search / LLM workflow / autonomous agentic loop) before an agent is built — the installable form of…
Audit always-loaded behavioral rules for quality — residency cost model, 7 verdicts incl. Demote-to-skill / Dissolve. AKC Curate-phase skill for the rules layer.
The Authorship Strategy framework as a single always-loaded behavioral rule — the deterministic counterpart to the authorship-strategy-skill. Part of the shimo4228 research ecosystem.
The Agent Knowledge Cycle as a single behavioral rules file — install all six phases without the six standalone skills.
Claude Code skill: operational form of the authorship-strategy four-layer judgment framework (three-axis inversion, four-layer stack, prohibited and encouraged actions, 11-item judgment checklist).…
Agent Skill: audit and sync the four citation layers of a research repo — in-text docs, .zenodo.json, graph.jsonld, Wikidata P2860 — bottom-up.
Claude Code Agent Skill bundle for writing & reviewing academic papers (SSRN / arXiv / Zenodo / journals). Bundles an orchestrator skill, a draft skill, and 5 reviewer subagents — first repo to pac…
Claude Code skill: orchestrator for human-facing writing & review — AI-slop banned list (JA + EN), Voice rules (だ/である × 発見調), title conventions, role boundaries for article-writing / editor / essay…
Agent Skill: four load-bearing patterns for layering deterministic code and LLM calls in one pipeline — guard, filter, judge+enforce, orchestrator+worker.
Agent Skill: structural security for autonomous LLM agents — Security by Absence, Untrusted Content Boundary, One External Adapter.
Claude Code skill: release workflow for DOI-registered research repositories following the authorship-strategy identifier-federation triplet (ADRs 0001-0003). Five-phase verify-and-deposit runbook.
Agent Skill: design a research intake filter that admits only information likely to change your next action. Search widely, intake narrowly.
Agent Skill: per-task decision framework — deterministic code or LLM? The structural-vs-semantic axis, false-positive test, and enumerate/decide split.
Abhidharma × Computational Phenomenology — a personal inquiry into attention and the non-self
Honest #1 on EinsteinArena's Prime Number Theorem — multiscale supports + a column-generation LP engine. Built with Claude Code / Opus.
New state-of-the-art bounds for open problems
Claude Code skill: writes & improves human-facing READMEs — the single canonical entrance for humans, search, and AI Overviews. Human-surface counterpart to llms-txt-writer. Deterministic structura…
A curated list of resources for Japanese natural language processing (NLP): Python libraries, LLMs, dictionaries, corpora, and datasets. Includes Claude Code skills to search resources.