agent-skills

Reusable agent skills for AI coding assistants. Compatible with Cursor, Claude Code, Codex, and 40+ agents via npx skills.

skills.sh

Skills

Download grabs a ready-to-upload .zip of just that skill — for Claude Desktop (Settings → Capabilities), which has no CLI and can't run npx skills. Rebuilt whenever a skill changes on main.

Skill Description Download
production-grade Principle-engineering posture for production-grade code. Plans before code, simplest-correct-solution-first, ACM-grade algorithms, EXPLAIN-first databases, idempotent-atomic writes, realtime-first, concurrent-by-default, TDD-steered E2E, runtime-coherent infrastructure, maintenance & remediation of inherited or generated code. Substrate-agnostic. .zip
tailwind-v3-to-v4-migration Migrate a project from Tailwind CSS v3 to v4 safely and completely. Codemod-first (@tailwindcss/upgrade), then the judgment it can't do: dependency/PostCSS/Vite/CLI plumbing, JS config → CSS @theme (or keep via @config), the v4 changed-defaults audit + compat shims, a residual sweep, and proving the migration is a visual no-op. Framework-agnostic; bundles the official upgrade guide. .zip
create-skill-autoresearch Factory for building production-grade agent skills: interviews you for purpose + gold standards, researches the domain, drafts the skill, autonomously improves it against an LLM-as-judge (or real-world) metric, and verifies it with an independent multi-agent panel. Extends the official single-pass skill creators. Orchestrates companion skills (autoresearch, premortem, handoff) — install those alongside it; the full batteries-included environment is the agent-skills-harness. .zip
tribunal Doer → verifier-panel → consensus delivery verification for any artifact (code slices, plans, documents, audits). An orchestrator freezes acceptance criteria before implementation, dispatches a doer, then convenes a context-walled panel of independent verifiers — including an adversary with a must-oppose mandate — for evidence-anchored review (citations grepped before consensus math) adjudicated to SHIP / SHIP_WITH_CAVEATS / ITERATE / BLOCK / ESCALATE. Principles-first: panel lenses, dimensions, and prompts derived per artifact from hard invariants. Platform-agnostic; degrades to sequential fresh-context sessions without subagents. Outcome-weighted A/B benchmarks (blind-judged, executed answer keys, honest negatives included) in its README. .zip
database-documentation Generate database documentation that reaches provable 100% parity with the real schema. Grounds every statement in the live database (system-catalog introspection) and cross-validates against ORM models, migrations, generated types, and seeds, then proves completeness with a mechanical count gate + identity-diff (re-introspect and diff until empty). Emits mermaid ERDs, per-table data dictionaries, and a machine-readable schema.json. Engine-agnostic (PostgreSQL, MySQL, SQL Server, SQLite) and ORM-agnostic (Prisma, TypeORM, Drizzle, Sequelize, Knex, Django, Rails). ≥3 context-walled adversaries (omission / hallucination / correctness) gate the result; platform-agnostic with graceful degradation. Reproducible dual-dialect benchmark with honest negatives in its README. .zip
app-ai-guardrails Scaffold a new production application with the full agentic-AI guardrail canon baked in from commit #1: a uniform 7-gate interface (lint, typecheck, test, coverage, build, e2e, audit) on each stack's native runner, strict types, maximal static analysis, coverage thresholds with teeth, pre-commit hooks, hardened CI, supply-chain pinning, and an agent-ready AGENTS.md — every gate verified green before the first commit. Native adapters for Next.js, NestJS, Django, Go, Rust, and Spring Boot, plus a discovery method for any other stack. Greenfield only (v1); retrofitting an existing repo is v0.2. Same-model uplift benchmark (bare vs with-skill, per-stack medians, honest negatives) in its README. .zip

Benchmarks

Skill benchmarks live in benchmarks/, one directory per skill — kept outside skills/ so they don't ship when a skill is installed. Each is self-contained (arms, scorer, configs) and measures the same model with the skill vs without. See

Install

npx skills add a-tokyo/agent-skills

Or install a specific skill:

npx skills add a-tokyo/agent-skills --skill production-grade
npx skills add a-tokyo/agent-skills --skill tailwind-v3-to-v4-migration
npx skills add a-tokyo/agent-skills --skill create-skill-autoresearch
npx skills add a-tokyo/agent-skills --skill tribunal
npx skills add a-tokyo/agent-skills --skill database-documentation
npx skills add a-tokyo/agent-skills --skill app-ai-guardrails

create-skill-autoresearch orchestrates companion skills at runtime — autoresearch (github/awesome-copilot), premortem (parcadei/continuous-claude-v3), and handoff (mattpocock/skills). Install those too, or just use the batteries-included agent-skills-harness where everything is wired up. Its craft layer (references/skill-craft-principles.md) is distilled from writing-great-skills (mattpocock/skills, MIT).

Claude Code plugin

Alternative to npx skills for Claude Code / IDE users — install via the native plugin system:

/plugin marketplace add a-tokyo/agent-skills
/plugin install production-grade@a-tokyo-skills

Or install every skill at once:

/plugin install all-skills@a-tokyo-skills

License

MIT Ahmed Tokyo