skills

Discover model capabilities and generate inference code. Use when the user wants to list available models, inspect model methods and signatures, generate working inference code, or use Clarifai's OpenAI-compatible API. Covers model discovery, code generation, featured model URLs, method selection, and media input handling.

Orchestrate multi-step Clarifai workflows. Use when the user needs to combine multiple Clarifai operations (e.g., "upload AND train", "train THEN deploy", "classify THEN describe"), build end-to-end ML pipelines, create orchestrator models for single-endpoint access, or design composite workflows. Triggers on sequential language ("then", "after", "followed by"), conjunctions combining operations ("and", "plus"), workflow/pipeline planning, multi-model chaining, or any request involving 2+ distinct Clarifai capabilities. Route here FIRST when detecting composite intent, then delegate to building-block skills.

Execute Clarifai operations using the `clarifai` CLI. Use when the user needs to authenticate, check identity, manage apps, compute clusters, nodepools, deployments, pipelines, pipeline runs, artifacts, or initialize models. Covers model init (toolkits, MCP, GitHub templates), pipeline templates, compute orchestration CLI, and all model/pipeline operations.

Upload and manage datasets on Clarifai platform. Use when the user wants to upload image datasets with annotations (classification, detection, segmentation), convert data formats (COCO, VOC, YOLO), create custom DataLoader classes, or manage dataset inputs. Covers load_module_dataloader pattern, Feature classes, and dataset operations.

Manage the end-to-end lifecycle of Clarifai deployments. Use when deploying, updating, monitoring, or removing model deployments. Covers the CLI-first deploy/status/logs/undeploy workflow, version tracking with metadata, the Delete-then-Post cycle, and state transition monitoring.

Create MCP (Model Context Protocol) servers on Clarifai. Use when the user wants to create tool-calling servers, implement custom tools for agentic AI workflows, bridge external stdio MCP servers (GitHub, filesystem), or deploy MCP servers for agentic models. Covers MCPModelClass, StdioMCPModelClass, FastMCP, tool definitions, and integration with agentic models.

Upload and deploy custom models to Clarifai platform. Use when the user wants to (1) deploy LLMs using toolkits (vLLM, SGLang, Ollama, HuggingFace), (2) convert existing Python models to Clarifai format, (3) create custom model implementations with ModelClass, or (4) configure compute resources. Covers toolkit templates, ModelClass implementation, config.yaml, checkpoints, GPU configuration, and local testing.

Debug and monitor Clarifai resources (Runners and Pipelines). Use when a resource is stuck, crashing, or failing to build. Covers the CLI logs command, interpreting builder logs, runtime stdout/stderr, and infrastructure events to resolve scheduling and execution issues.

Create and run Clarifai pipelines for batch processing and multi-step workflows. Use when the user wants to create custom pipelines with containerized steps, transfer data between steps using Artifacts API, or orchestrate long-running workflows. NOT for training - use clarifai-training-pipelines for model training.

Train models using Clarifai pipeline templates. Use when the user wants to train image classifiers (ResNet-50), object detectors (YOLOF), or create custom training pipelines. Covers built-in templates (classifier-pipeline-resnet, detector-pipeline-yolof), custom training implementation, hyperparameter configuration, and training monitoring.