Clarifai Skills

Agent skills for the Clarifai AI platform. These skills give AI coding assistants (Claude Code, Cursor, Codex, and others) deep knowledge of Clarifai's model deployment, inference, MCP, pipeline, and data management capabilities.

Built on the open Agent Skills standard — works across 30+ agent platforms.

Quick Start

Option A: Clarifai CLI (recommended)

pip install clarifai
clarifai skills install

That's it. Auto-detects your agents (Claude, Cursor, Codex) and installs all 11 skills globally.

Option B: Claude Code Plugin

claude plugin marketplace add Clarifai/skills
claude plugin install clarifai-skills

Option C: Manual

git clone https://github.com/Clarifai/skills.git
clarifai skills install --source ./skills

What's Included

Skill	What it does
clarifai-cli	Full CLI reference: init, serve, deploy, status, logs, undeploy, predict, list-instances
clarifai-model-upload	Deploy models with vLLM, SGLang, HuggingFace, Ollama. ModelClass, config.yaml, GPU config
clarifai-inference	Model discovery, method signatures, code generation, OpenAI-compatible API, agentic models
clarifai-mcp	Build MCP servers with MCPModelClass, StdioMCPModelClass, FastMCP tools
clarifai-deployment-lifecycle	Deploy/status/logs/undeploy lifecycle, version patching, state monitoring
clarifai-observability	Debug stuck deployments: CLI logs, K8s events, common resolutions
clarifai-agentic-flows	Multi-step orchestration: "train THEN deploy", client scripts, server orchestrators
clarifai-datasets	Upload datasets with annotations (classification, detection, segmentation), format conversion
clarifai-pipelines	Batch processing pipelines with containerized steps and Artifacts API
clarifai-training-pipelines	Train classifiers (ResNet-50) and detectors (YOLOF) using built-in templates
clarifai-grpc	Low-level gRPC API with protobufs for advanced platform operations

CLI Commands

clarifai skills install                          # install all skills (auto-detect agents)
clarifai skills install --claude                 # install for Claude only
clarifai skills install clarifai-cli --claude    # install one skill
clarifai skills list --installed                 # see installed skills
clarifai skills update                           # update to latest
clarifai skills remove clarifai-grpc --claude    # remove one skill
clarifai skills remove --all                     # remove all

Flags:

• --claude, --codex, --cursor, --all-agents — target specific agents (default: auto-detect)
• --global / --local — user-wide or project-level (default: global)
• --source /path — install from local clone instead of GitHub

How It Works

Skills follow the Agent Skills standard:

clarifai-cli/
  SKILL.md              # Instructions (YAML frontmatter + markdown)
  references/           # Detailed reference docs
  examples/             # Working code examples

Each SKILL.md has a name and description in YAML frontmatter. Agents load these at startup for routing, then read the full content on-demand when a matching task is detected.

Install layout:

~/.agents/skills/           # Central (one copy)
  clarifai-cli/
  clarifai-model-upload/
  ...
~/.claude/skills/           # Symlinks per agent
  clarifai-cli -> ~/.agents/skills/clarifai-cli
  ...

Example: Deploy a Model in 3 Commands

With skills installed, just tell your agent:

"Deploy Qwen3-0.6B with vLLM"

The agent reads the clarifai-cli and clarifai-model-upload skills and runs:

clarifai model init ./qwen --toolkit vllm --model-name "Qwen/Qwen3-0.6B"
clarifai model serve ./qwen          # test locally
clarifai model deploy ./qwen         # deploy to cloud

Repository Structure

.
  AGENTS.md                              # Master index (loaded by agents)
  marketplace.json                       # Skills registry for CLI
  .claude-plugin/marketplace.json        # Claude Code plugin manifest
  .github/skills/                        # All 11 skills
    clarifai-cli/
      SKILL.md
      references/
      examples/
    clarifai-model-upload/
      SKILL.md
      references/
      examples/
    ...
  scripts/
    generate_marketplace.py              # Regenerate marketplace.json

Contributing

1. Edit or create a skill in .github/skills/your-skill/SKILL.md
2. Follow the SKILL.md format: YAML frontmatter (name, description) + markdown body
3. Keep SKILL.md under 500 lines; move detailed docs to references/
4. Run python scripts/generate_marketplace.py to update the registry
5. Test with clarifai skills install --source . --claude --force

Links

• Clarifai Platform
• Clarifai Python SDK
• Runner Examples
• Agent Skills Standard
• Clarifai Docs