A collection of Claude Code skills for building on the Alchemyst Context Layer — the API that turns context into a computable primitive.
These skills give Claude the domain knowledge and scaffolding it needs to plan and ship context-aware agents on Alchemyst, instead of falling back on generic RAG or memory-library recipes.
| Skill | Kind | What it does |
|---|---|---|
| skills/context-api | direct | Developer reference for the Context Layer SDK — search_context, groupName, document set arithmetic, staleness. Trigger when working at the primitive level. |
| skills/contextual-agent-generator | meta | Interviews the user about an agent they want to build (or retrofit) on Alchemyst, matches their goal to a documented archetype, and generates a project-scoped child skill that plans and scaffolds the implementation end-to-end. |
The two are complementary: context-api documents the primitive;
contextual-agent-generator plans projects built on top of it. See
how-to.md for the meta-vs-direct distinction and how to
install / author / extend skills in this repo.
Install it using the npx skills executable.
npx skills add alchemyst-ai/alchemyst-skillsClone the repo, then symlink the skills you want into Claude Code's skills directory:
git clone https://github.com/alchemyst-ai/alchemyst-skills.git
cd alchemyst-skills
# user-scoped: available to all your projects
mkdir -p ~/.claude/skills
ln -sfn "$PWD/skills/context-api" ~/.claude/skills/context-api
ln -sfn "$PWD/skills/contextual-agent-generator" ~/.claude/skills/contextual-agent-generatorOr scope to a single project:
mkdir -p .claude/skills
ln -sfn "<abs-path>/alchemyst-skills/skills/context-api" .claude/skills/context-api
ln -sfn "<abs-path>/alchemyst-skills/skills/contextual-agent-generator" .claude/skills/contextual-agent-generatorOnce installed, invoke from a Claude Code session by description (e.g.
"build a context-aware agent on Alchemyst for this repo") — Claude
matches the request against each skill's frontmatter description and
loads the relevant SKILL.md.
alchemyst-skills/
├── README.md this file
├── how-to.md install / author / extend skills
├── .github/workflows/validate.yml CI: SKILL.md frontmatter + kebab-case names
└── skills/
├── context-api/
│ └── SKILL.md
└── contextual-agent-generator/
├── SKILL.md
├── README.md
├── references/ authoritative reference material the skill loads on demand
├── scripts/ shell helpers (project detection, signal scan, slugify)
└── templates/child-skill/ templates rendered into the generated child skill
Every PR touching skills/** runs
.github/workflows/validate.yml, which
checks:
- each
skills/<name>/SKILL.mdexists and has YAML frontmatter - frontmatter contains
name:anddescription: - the directory name is kebab-case (
^[a-z0-9]+(-[a-z0-9]+)*$)
Run the same checks locally before pushing:
act -W .github/workflows/validate.yml # if you have nektos/act
# or just inline the script from validate.ymlSee how-to.md §Authoring a new skill.
The short version: create skills/<kebab-name>/SKILL.md with valid
frontmatter, keep references and scripts as siblings under that
directory, and link from this README's table.
This repository is MIT licensed — see LICENSE for details.