AgentMark agent skills

Claude Code / agent skill that teaches AI agents how to use AgentMark — CLI commands, prompt authoring, datasets, experiments, evals, and git-based deploys.

Install

Pick whichever fits your tooling:

Universal (npx skills, cross-tool — Claude Code, Codex, Cursor, GitHub Copilot, Amp, Antigravity, +others):

npx skills add agentmark-ai/skills

Files land in ./.agents/skills/agentmark/ with symlinks into per-tool paths automatically.

Native Claude Code plugin marketplace:

claude
> /plugin marketplace add agentmark-ai/skills
> /plugin install agentmark

This uses Claude Code's built-in plugin system (.claude-plugin/marketplace.json in this repo declares the marketplace).

Mintlify well-known discovery (for any AgentMark docs URL):

npx skills add docs.agentmark.co

Once installed, Claude Code (or any agentskills.io-compatible tool) auto-discovers the skill at session start and invokes it when AgentMark context appears (.prompt.mdx files, agentmark.json, @agentmark-ai/* imports, or explicit mentions of AgentMark / CLI commands).

What's in this repo

skills/agentmark/
├── SKILL.md                          # entry point — read by Claude on auto-invoke
├── workflows/                        # one file per common task
│   ├── creating-prompts.md
│   ├── building-datasets.md
│   ├── running-experiments.md
│   ├── using-evals.md
│   └── deploying.md
├── reference/                        # auto-generated, do not hand-edit
│   ├── cli-commands.md               # ← from cli-src/index.ts
│   ├── frontmatter-schema.md         # ← from prompt-core schemas.ts
│   ├── api-commands.md               # ← from openapi-spec.json
│   └── models.md                     # ← from model-registry
└── evals/scenarios.json              # eval scenarios for skill validation

How it stays current

The skill is a thin pointer that delegates to live docs, not a static reference. It teaches agents to:

1. Run npx agentmark <cmd> --help for canonical CLI surface
2. Fetch https://docs.agentmark.co/llms.txt for doc navigation
3. WebFetch specific .md pages from docs.agentmark.co for content

The four reference/*.md files are auto-generated from the canonical sources in agentmark-ai/agentmark on every release, so encoded facts can't drift.

Do not edit directly

This repo is auto-generated from agentmark-ai/agentmark.

Changes belong in skills/agentmark/ of the source repo. On every CLI release, a GitHub Action mirrors the source skills/ directory here. Direct commits to this repo will be overwritten on the next sync.

For substantive proposals, open an issue or PR against agentmark-ai/agentmark.

License

Same as AgentMark — AGPL-3.0-or-later.