Skip to content

Skills Marketplace

MemoRizz agents can discover and use external agent skills at runtime through pluggable marketplace providers. When a marketplace is enabled, the agent receives search and fetch tools that let it find relevant skills, read their instructions, and apply them to complete tasks.

Available Providers

Provider Identifier API Key Required Tools Registered
Vercel Agent Skills vercel No (search uses skills.sh; optional GITHUB_TOKEN raises GitHub rate limits) vercel_skills_search, vercel_skill_fetch
SkillsMP skillsmp Yes (SKILLSMP_API_KEY) skills_marketplace_search

Vercel Agent Skills

Vercel Agent Skills is an open ecosystem of reusable capabilities packaged as SKILL.md files in GitHub repositories. Each skill contains structured instructions (with YAML frontmatter for metadata and Markdown content for steps) that an agent can follow to complete a specific type of task.

The ecosystem covers skills for React, Next.js, AI SDK, deployment, design, browser automation, commerce, and more, with 269+ published skills from Vercel and the community.

How It Works

  1. The agent receives vercel_skills_search and vercel_skill_fetch tools
  2. When given a task, the agent searches for relevant skills by keyword
  3. The agent fetches the SKILL.md from the matching repository
  4. The agent reads and follows the instructions to complete the task

Enabling via Code

from memorizz.memagent import MemAgent

agent = MemAgent(
    llm_config=llm_config,
    memory_provider=provider,
    skills_marketplace_provider="vercel",
)

print(agent.run("Build a performant React component following best practices"))

Or with the builder:

from memorizz.memagent.builders import MemAgentBuilder

agent = (
    MemAgentBuilder()
    .with_instruction("You are a frontend developer.")
    .with_memory_provider(provider)
    .with_llm_config(llm_config)
    .build()
)

# Enable marketplace after build
agent.with_skills_marketplace_provider("vercel")

Enabling via Web UI

  1. Navigate to the Vercel Skills page in the sidebar to browse and search skills
  2. When creating or editing an agent, select Vercel Agent Skills (skills.sh) from the Skills Marketplace dropdown
  3. The agent's playground will show the Vercel provider status in the configuration panel

Fetching from a Specific Repository

Users can point the agent at any GitHub repository containing a SKILL.md:

# The agent fetches the SKILL.md and follows its instructions
agent.run("Use the skill from vercel/ai-chatbot to scaffold a chatbot")

The vercel_skill_fetch tool accepts:

  • repo – GitHub repository in owner/repo format or a full URL
  • skill_name – optional name for multi-skill repositories (e.g., "nextjs-app-router")
  • branch – Git branch to fetch from (default: main)

Multi-Skill Repositories

Some repositories contain multiple skills under a skills/ directory. When the agent fetches from such a repo without specifying a skill_name, it receives a list of available skills and can select the appropriate one.

Environment Variables

Variable Purpose
GITHUB_TOKEN Optional GitHub personal access token for higher API rate limits
MEMORIZZ_DEFAULT_SKILLS_MARKETPLACE_PROVIDER Set to vercel to auto-enable on all agents

How Skills Are Discovered in Repositories

The provider searches these standard paths in order:

  • SKILL.md (root)
  • skills/SKILL.md
  • skills/.curated/*/SKILL.md
  • skills/.experimental/*/SKILL.md
  • .claude/skills/*/SKILL.md
  • .agents/skills/*/SKILL.md
  • .cursor/skills/*/SKILL.md

This aligns with the Vercel skills CLI specification.

Tool Reference

Search the Vercel skills ecosystem for relevant skills.

Parameter Type Default Description
q str required Search query (e.g., "react", "deployment", "AI SDK")
limit int 20 Maximum results (1–100)

Returns a dict with ok, skills (list of repo/name/description/stars), count, and total_count.

vercel_skill_fetch

Fetch a skill's instructions from a GitHub repository.

Parameter Type Default Description
repo str required Repository in owner/repo format or GitHub URL
skill_name str "" Target a specific skill in a multi-skill repo
branch str "main" Branch to fetch from

Returns a dict with ok, name, description, instructions (full SKILL.md content), repo, path, and metadata.

SkillsMP

The SkillsMP provider connects to a hosted skills marketplace with keyword and AI-powered search.

Configuration

  1. Obtain an API key from skillsmp.com
  2. Set it via the Settings page in the web UI or export SKILLSMP_API_KEY
agent = MemAgent(
    llm_config=llm_config,
    memory_provider=provider,
    skills_marketplace_provider="skillsmp",
    skills_marketplace_config={"api_key": "your-key"},
)

Tool Reference

Parameter Type Default Description
q str required Search query
mode str "keyword" "keyword" or "ai" for semantic search
page int 1 Page number
limit int 20 Results per page (max 100)
sort_by str "" "stars" or "recent"

Web UI: Skills page

The local web UI has a Skills page in the sidebar. It searches the skills.sh directory, ranked by installs, with no token needed. Paste an owner/repo or GitHub URL to fetch a skill directly, and read the full SKILL.md before using it.

Attach a skill to an agent

Select a result and choose Attach skill. MemoRizz saves the skill's SKILL.md under ~/.memorizz/skills/<owner>/<repo>/<skill>/, records where it came from in .memorizz-source.json (repository, path and SHA-256), and adds the file to the agent's skill_paths. The agent then lists it with list_skills and reads it with read_skill. Attaching runs nothing; skill code only runs in a sandbox through the skill tools.

The same search and Attach button are in the playground under Settings → Skills. To detach a skill, remove its row there and save.

Skills from collections and plugins are found wherever they live in the repository (for example skills/notion/knowledge-capture/SKILL.md); when a name matches nothing, the error lists the skills the repository does contain.

The HTTP API behind these buttons:

GET    /api/agents/{agent_id}/skills
POST   /api/agents/{agent_id}/skills        {"repo": "owner/repo", "skill_name": "name"}
DELETE /api/agents/{agent_id}/skills?path=/absolute/path/SKILL.md

Attaching pins a reviewed skill to one agent. Enabling the marketplace instead lets the agent search and fetch skills on its own at run time.