Skip to content

agentic-work/agenticwork-skills

agenticwork-skills

The source-of-truth repository for Agent Skills across the AgenticWork platform — portable, progressively-disclosed capability modules that any compatible agent runtime can load.

License Skills format


What this is

A skill is a self-contained Markdown module (SKILL.md) that teaches a model how to do something well — the conventions, constraints, and worked patterns it needs to produce expert-quality output faithfully and consistently, without the host application hard-coding a template for every case.

Skills use progressive disclosure: a skill's name + description stay resident in the model's context (cheap), and the full body is loaded on demand only when the task calls for it. The model decides, in-context, when to read a skill — there is no separate router.

This repository is the single source of truth for the skills AgenticWork ships, and it is designed to be consumed across runtimes:

Consumer How it loads skills
AgenticWork Synced into the platform skill registry; surfaced to chat / flows / codemode via the on-demand skill tools
OpenAgentic Imported as the canonical skill set
Claude (Anthropic Agent Skills) Format-compatible with anthropics/skills — installable as a skill set
OpenClaw & other open agent/persona systems Plain SKILL.md + resources; load directly from the repo

Skill format

Each skill lives in its own directory and is consumable by any Agent-Skills-compatible runtime:

skills/
  <skill-name>/
    SKILL.md          # YAML frontmatter (name, description, …) + the skill body
    resources/        # optional reference material loaded on demand
    scripts/          # optional helper scripts the skill may invoke

SKILL.md frontmatter (minimum):

---
name: inline-artifact-design
description: >-
  Use when a user asks for a rich inline/interactive visualization, dashboard, or
  scene rendered in the chat. Teaches how to author a single default-exported React
  component, using only the curated pinned libraries, that renders at production
  fidelity in the sandboxed artifact iframe.
---

The body is standard Markdown — instructions, constraints, and worked examples. Keep the name + description tight (they are always resident); push detail into the body and resources/ (loaded only when the skill is opened).

Flagship skill — inline/interleave artifact authoring

The headline skill in this set lets a capable model author unique, production-grade inline artifacts on demand — charts, dashboards, 3D scenes — as a single React component, rendered in a hardened sandboxed iframe, without the platform shipping a hard-coded template for every layout. The skill is what makes that output faithful and consistent rather than a lucky one-off: it pins the curated library set, the theme tokens, the default-export contract, and the design language the output must follow.

Governance

License

Apache License 2.0 — see NOTICE for attribution. Some skills may reference upstream skills (e.g. Anthropic's anthropics/skills) that carry their own licenses; those are noted per-skill where applicable.

About

No description, website, or topics provided.

Resources

License

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors