# Build trustworthy AI data analysis with Claude Code

**The oryxflow Claude Code plugin makes an AI coding agent's data analysis trustworthy and reproducible.** You get numbers you can believe and results you can regenerate months later — and because nothing recomputes twice, the trustworthy path is also the fast, cheap one. It ships as a skill plus slash commands — not an MCP server.

AI coding agents now write real data-science pipelines — feature engineering, model training, experiment sweeps. They write plausible code fast. The hard part isn't the code; it's knowing whether you can **believe the number** it just handed you, and whether you could **produce it again** next month.

The **oryxflow Claude Code plugin** is how you get both. It pairs the oryxflow library — which carries the state an agent can't hold in its head — with a skill that makes the agent verify its own work. Three things follow, in the order they matter:

1. **You can trust the result.** A change to the data, a parameter, or the code reruns exactly what that change affects, and the agent checks that the rerun actually happened — so a number can't quietly sit on stale inputs.
2. **You can reproduce it.** Every result is tied to the code and inputs that made it in a queryable lineage record, so "how did we get this?" is answerable next month, by someone else.
3. **It costs you less, not more.** Reproducibility is normally a tax you pay in rerun time. Here it isn't: completed work is reused rather than recomputed, so you're not waiting on the 10-minute data pull again or spending tokens watching an agent re-run it.

> [!TIP]
>
> **The short version**
>
> Install the plugin, then just describe what you want. It scaffolds a project, wires the DAG, and follows the [house conventions](<https://github.com/oryxintel/oryxflow-claude-plugin/blob/main/skills/oryxflow/conventions.md>) so the analysis is trustworthy and reproducible by default — no manual task wiring, no stale-data surprises.

## Install

```
/plugin marketplace add https://github.com/oryxintel/oryxflow-claude-plugin.git
/plugin install oryxflow@oryxintel
```

When `/plugin` asks where to install, pick **Install for you (user scope)** so the plugin works in every project, not just this one. From a shell, `claude plugin install oryxflow@oryxintel` installs at user scope by default.

Once installed, the `oryxflow` skill **auto-activates** whenever you work in an oryxflow project (editing `tasks.py` / `flow.py` / `run.py` / `cfg.py` / `flow_params.py`) — you don't invoke it, it's just on.

> [!NOTE]
>
> The skill activates inside an oryxflow **project**. If you install it into an empty directory and nothing happens, run [`/oryxflow:init-project`](<https://docs.oryxflow.dev/docs/claude-plugin/commands/index.md>) first (or ask the agent to set up an oryxflow project) so there's a pipeline for the skill to work on.

## Why can't I just trust the agent's numbers?

Because in data work a mistake usually doesn't announce itself. A coding agent's weakness here isn't syntax — it's **invisible state**. Across a long session it loses track of what's already computed and whether it's still valid, then trains on stale features or hands you a number built from last week's inputs. Nothing errors. The result just looks the same whether it's right or wrong.

So the useful question isn't *"is the agent trustworthy?"* — it's *"can I check its work without rebuilding it?"* oryxflow makes the answer yes by externalizing the state the agent can't hold, and the plugin makes the agent *use* that machinery correctly. It:

- **starts every session by reading what's actually current** — pending staleness warnings, last runs, recent failures — so it never mistakes a stale result for a fresh one;
- **verifies after each edit that the intended tasks actually reran**, so a change that should have invalidated downstream work can't pass silently;
- **answers every staleness warning with the right move** — recompute, accept an output-equivalent refactor, or pin — instead of guessing (see [Automatic code invalidation](<https://docs.oryxflow.dev/docs/managing-workflows/#automatic-code-invalidation>));
- **records decision-relevant results as lineage**, so any result stays reproducible and explainable across sessions.

These are the same disciplines documented in the [CLAUDE.md snippet for AI-agent projects](<https://docs.oryxflow.dev/docs/managing-workflows/#claudemd-snippet-for-ai-agent-projects>) — shipped as a skill so they load automatically and stay current with the library, instead of a copy you paste and forget. ([Claude Code skills for data science](<https://docs.oryxflow.dev/blog/posts/claude-code-skills-for-data-science/index.md>) explains why a skill, rather than a data connector or a notebook runner, is the piece that carries this discipline.)

## In this section

- **[Trustworthy AI data analysis](<https://docs.oryxflow.dev/docs/claude-plugin/trust/index.md>)**

  Why you shouldn't take an agent's numbers on faith — and how the plugin makes its work *cheap to verify* instead: session status, rerun verification, and durable lineage.
- **[Commands](<https://docs.oryxflow.dev/docs/claude-plugin/commands/index.md>)**

  The five slash commands — scaffold a project, migrate an existing analysis, check standards, and put data under Git LFS. The LLM-eval commands live in the separate `oryxeval` plugin.
- **[Scaffold an LLM eval](<https://docs.oryxflow.dev/docs/claude-plugin/evals/index.md>)**

  Four commands in the `oryxeval` plugin (`/plugin install oryxeval@oryxintel`) that turn "is this prompt better?" into a number you can defend — a plan, a scaffold wired to it, a real case set, and a run that prints its bill before it spends.
- **[Data-science project structure](<https://docs.oryxflow.dev/docs/claude-plugin/project-structure/index.md>)**

  The load-bearing scaffold that keeps AI-generated data-science code from rotting — separation of concerns that's the shape of the code, not just its filing — and grows with the project.
- **[Stop AI data analysis turning into a mess](<https://docs.oryxflow.dev/docs/claude-plugin/coding-standards/index.md>)**

  Canonical names, code grouped by subject, docstrings as documentation — loaded into the agent's context so they shape the analysis code as it's written, not audited after.
- **[Why library + plugin is a matched pair](<https://docs.oryxflow.dev/docs/claude-plugin/why/index.md>)**

  The division of labor: what the library carries, what the agent carries, and why the pairing gets *more* valuable as a project grows.

## Learn more

- Plugin repository and issues: [https://github.com/oryxintel/oryxflow-claude-plugin](<https://github.com/oryxintel/oryxflow-claude-plugin>)
- House conventions the plugin follows: [https://github.com/oryxintel/oryxflow-claude-plugin/blob/main/skills/oryxflow/conventions.md](<https://github.com/oryxintel/oryxflow-claude-plugin/blob/main/skills/oryxflow/conventions.md>)
- Plugin changelog: [https://github.com/oryxintel/oryxflow-claude-plugin/blob/main/docs/CHANGELOG.md](<https://github.com/oryxintel/oryxflow-claude-plugin/blob/main/docs/CHANGELOG.md>)
- New to oryxflow? Start with **[Why oryxflow](<https://docs.oryxflow.dev/docs/why-oryxflow/index.md>)** and the **[Quickstart](<https://docs.oryxflow.dev/docs/quickstart/index.md>)**.

This library is the engine the plugin drives; the full API is documented throughout the rest of these docs.