MLflow Skills plugin
Ten MLflow skills teach a coding agent to add tracing, diagnose failures and evaluate GenAI apps
- 91 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The skills run `mlflow` commands and scripts, call an MLflow server and read traces that hold user data
- The plugin hook fires inside the agent session and suggests invoking skills
Install
- Works in
Manual install
npx skills add mlflow/skillsThe skills installer, the first route in the README.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skills run `mlflow` commands and scripts, call an MLflow server and read traces that hold user data
- The plugin hook fires inside the agent session and suggests invoking skills
This is third-party code. Review the repository files before installing.
What it does
A skill set from the MLflow team for Claude Code, Cursor, Codex CLI, Gemini CLI and OpenCode. `mlflow-trace-instrumentation` adds tracing to Python and TypeScript apps, `mlflow-trace-retrieval` finds and filters traces, `mlflow-trace-analysis` breaks down spans, agent failures and multi-turn sessions, and `mlflow-trace-metrics` builds aggregate metrics such as token use. `mlflow-evaluation` handles scoring: it prepares datasets, designs scorers and runs evaluation. `mlflow-agent-improvement` fixes agent behavior from traces and locks the fix in with a regression test. The rest cover first-time setup (`mlflow-onboarding`), documentation search (`mlflow-docs`) and SageMaker connection. The `mlflow-agent` dispatcher routes a broad request to the right skill. The plugin also ships a `mlflow-suggest-hook` hook.
Who it is for. Developers of AI agents and LLM apps who already use MLflow or plan to add it for tracing and evaluation.
Good fit when
- A LangChain agent answers wrongly and you need to find the step where it went off through the trace
- You need MLflow tracing added to an app without hand-marking every call
- You want scorers and a dataset to measure agent accuracy before and after changes
- A business rule changed and the agent must be fixed with a regression test that fails before the fix
Not a fit when
- The project does not use MLflow and has no plan to adopt it
- You need general model metrics with no app tracing: the skills target GenAI agents
- The task has nothing to do with observability or evaluation, where the skills add nothing
Example request
Add MLflow tracing to my LangChain app and show why the last request failedLimitations
- The skills assume a running MLflow server or backend and the installed `mlflow` package: without them the agent has nothing to trace.
- Some scenarios are tied to Databricks and Amazon SageMaker and need access to those platforms.
- The `mlflow-agent` skill is marked as manually invoked only, and the agent does not run it on its own.
- Evaluation with an LLM judge spends paid model requests, with volume depending on dataset size.
How to disable. In Claude Code run `/plugin uninstall mlflow` or delete the skill folders from `~/.claude/skills/`. The `mlflow-suggest-hook` hook goes away with the plugin.
FAQ
How do I install the MLflow skills?
Run `npx skills add mlflow/skills`. Alternatives are copying the repository folders into the agent skills directory such as `~/.claude/skills/`, or adding the repository to a project as `.skills/mlflow`.
Which agents are supported?
Claude Code, Cursor, Codex CLI, Gemini CLI and OpenCode. Any agent that reads the skill format works.
How does `mlflow-trace-analysis` differ from `mlflow-agent-improvement`?
The first only diagnoses: it reads the trace and names the cause. The second makes the fix after diagnosis and proves it with a regression test.
Is Databricks required?
No, the skills work with plain MLflow. Databricks and SageMaker have their own reference files and a SageMaker connection skill.
What does `mlflow-trace-metrics` do?
It queries aggregate trace metrics such as token use and trends over time. For a single trace use `mlflow-trace-retrieval`.
Discussion
No questions or reviews yet.
Related
PluginsThe Roboflow plugin gives an agent computer vision skills: data, training, model evaluation, inference and platform API calls.
Flowtrace turns an agent's work into steps stored as files in a git repository, so you can follow, check, fix one by one and re-run them on new input
AI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
Skill Flow groups skills from many sources and deploys them to many agents, including Claude Code, Cursor and Codex.
A local MCP server that stores skills as versions and lets improved variants go live only after a human review
A tool that builds tested skills from a real workflow and governs their lifecycle through your own marketplace
README in short
The README explains why MLflow needs skills and lists ten of them in a table. Three install routes follow: the skills installer, copying from source and a project-level install. Separate folders hold reference material on evaluation, tracing and Databricks, and `.claude-plugin/plugin.json` describes the Claude Code plugin.