Skip to content

CMF Agent Skills¤

CMF ships a set of agent skills that let AI coding assistants (Claude Code, GitHub Copilot in agent mode, Cursor, Gemini CLI, and others) guide users through every CMF workflow — from first-time setup to querying artifact lineage.

Skills are prompt files that follow the agentskills.io standard. Once installed, they appear as slash commands in your coding assistant's chat interface.

Installing the skills¤

From your ML project directory, run:

npx skills add HewlettPackard/cmf --full-depth -y

This detects your coding agent and copies skills to the right directory automatically. Works with Claude Code, GitHub Copilot, Cursor, Codex, OpenCode, and any agentskills.io-compatible tool.

To install a single skill:

npx skills add HewlettPackard/cmf/tree/master/skills/cmf-instrument -y

To install globally across all projects:

npx skills add HewlettPackard/cmf --full-depth -y --global

Manual installation¤

Copy each skill directory into your agent's skills folder:

Tool Project-level directory Global directory
Claude Code .claude/skills/<skill-name>/ ~/.claude/skills/<skill-name>/
GitHub Copilot .agents/skills/<skill-name>/ ~/.copilot/skills/<skill-name>/
Cursor .agents/skills/<skill-name>/ ~/.cursor/skills/<skill-name>/
Codex .agents/skills/<skill-name>/ ~/.codex/skills/<skill-name>/
OpenCode .agents/skills/<skill-name>/ ~/.config/opencode/skills/<skill-name>/

Available skills¤

Skill Slash command Purpose
cmf /cmf <task> Router — describe your task in plain language and get routed to the right skill
cmf-init /cmf-init Install cmflib and run cmf init to configure a storage backend (local, S3, MinIO, SSH, OSDF)
cmf-instrument /cmf-instrument Add CMF tracking to existing Python ML pipeline code
cmf-sync /cmf-sync Push/pull metadata and artifacts to/from a CMF Server
cmf-query /cmf-query Query pipeline history and artifact lineage with CmfQuery
cmf-server /cmf-server Deploy CMF Server locally with Docker Compose, or connect to an existing server
cmf-mcp /cmf-mcp Deploy the CMF MCP server for natural language metadata queries

Skill details¤

/cmf — Router skill¤

The entry point for all CMF tasks. Describe what you want to do in plain language and the skill will route you to the correct specialized skill.

Usage examples:

/cmf I want to start tracking metadata in my ML pipeline
/cmf how do I push my artifacts to the CMF Server?
/cmf show me the lineage for my trained model

/cmf-init — Initialize CMF¤

Walks through installing cmflib and running cmf init with any supported storage backend:

  • Local — single-machine development, artifacts in a local directory
  • MinIO / Amazon S3 — shared object storage for teams
  • SSH remote — artifacts stored on a remote server
  • OSDF — distributed data federation across institutions (Pelican Platform)

After this skill completes you will have a working configuration (cmf init show returns valid output) and be ready to instrument your pipeline.

/cmf-instrument — Instrument ML code¤

The core instrumentation skill. It adds CMF metadata tracking calls to your Python ML pipeline scripts:

  1. Imports Cmf and initializes the metadata writer
  2. Creates contexts (stage types) and executions (stage runs) with hyperparameters
  3. Logs datasets, models, and metrics as inputs and outputs
  4. Calls finalize() to record the Git commit SHA

Includes copy-paste templates for data preparation, training, and evaluation stages, plus a checklist for auditing existing scripts.

Example instrumented training stage:

from cmflib.cmf import Cmf

def train(train_path: str, model_path: str, params: dict) -> None:
    metawriter = Cmf(filepath="mlmd", pipeline_name="my_pipeline")
    metawriter.create_context(pipeline_stage="Train")
    metawriter.create_execution(execution_type="Train", custom_properties=params)

    metawriter.log_dataset(train_path, "input")

    model = fit_model(train_path, params)

    metawriter.log_model(path=model_path, event="output",
                         model_framework="scikit-learn",
                         model_type="RandomForestClassifier",
                         model_name="RandomForestClassifier:default")
    metawriter.finalize()

/cmf-sync — Sync metadata and artifacts¤

Guides through the push/pull workflow for sharing results with teammates and a central CMF Server:

# Push your results
cmf metadata push -p my_pipeline
cmf artifact push -p my_pipeline

# Pull a teammate's work
cmf metadata pull -p my_pipeline
cmf artifact pull -p my_pipeline

Covers solo and team workflows, error diagnosis, and the complete command reference.

/cmf-query — Query lineage and artifacts¤

Writes CmfQuery Python code for common metadata questions:

  • Listing pipelines and their stages
  • Comparing hyperparameters across training runs
  • Tracing a model's full data lineage
  • Finding all artifacts consumed or produced by an execution

Also covers the CLI equivalents (cmf pipeline list, cmf execution list, cmf artifact list).

/cmf-server — Deploy or connect to CMF Server¤

Handles two scenarios:

  1. Local deployment — walks through cloning the CMF repo, creating a .env file with the right variables (CMF_DATA_DIR, REACT_APP_CMF_API_URL, POSTGRES_*, MCP_EXTERNAL_PORT), starting the Docker Compose stack, and verifying the web UI.

  2. Existing server — if a shared CMF Server is already running, guides the user through pointing their local cmf init at it with --cmf-server-url, verifying connectivity, and pushing the first pipeline.

/cmf-mcp — CMF MCP server¤

Covers two scenarios:

  1. Setting up — deploys the CMF MCP Server with Docker and connects it to your AI assistant (Claude Code, GitHub Copilot, Cursor, Codex).
  2. Already running — if the CMF MCP server is already configured, shows how to use it from within your agent using natural language prompts like "What pipelines are available?" or "Show the artifact lineage for models/rf.pkl".

After setup, the assistant calls CMF tools directly — no manual CmfQuery scripting required.

Skill files¤

The skill source files live in the skills/ directory of the CMF repository:

skills/
├── README.md               # Installation and usage instructions
├── cmf/
│   └── SKILL.md            # Router skill
├── cmf-init/
│   └── SKILL.md
├── cmf-instrument/
│   └── SKILL.md
├── cmf-sync/
│   └── SKILL.md
├── cmf-query/
│   └── SKILL.md
├── cmf-server/
│   └── SKILL.md
└── cmf-mcp/
    └── SKILL.md

See the skills/README.md for full details on installation, usage, and contributing new skills.