Customize Expertise

Tableau Agent expertise overview

Note coming at GA The following information describes a pattern similar to the way you can modify standard skills for the agent, but is not final.

In Tableau Agent, Expertise is conceptual Tableau knowledge that helps the agent understand and make decisions about the user’s request and the content in focus.

Tableau Agent comes loaded with a set of standard expertise for a broad range of Tableau practices and patterns. To suit your needs, you can customize the expertise your agent has by creating new expertise files or you can override the standard expertise.

Expertise files live in the ~/Documents/My Tableau Repository (Beta)/Tableau Agent/knowledge/ folder. As shown in the tree diagram below, custom expertise and expertise overrides can live at the root folder. The standard expertise files live in the .vendored/overridable/ and .vendored/protected folders.

Note Changing vendored expertise files themselves is not supported. Read on to understand how to customize overridable expertise, but changing protected expertise may disable Tableau Studio functionality, and changes to these files will heal themselves at the start of the next session.

~/Documents/My Tableau Repository (Beta)/
└── Tableau Agent/
    └── knowledge/
        ├── .markers/                                # one .sha256 per overridable expertise item
        ├── .vendored/                               # bundled standard expertise — rebuilt each session; do not edit
        │   ├── overridable/                         # standard expertise you may override
        │   │   ├── personalization/
        │   │   │   └── personalization.md
        │   │   └── strategy/
        │   │       └── strategy.md
        │   └── protected/                           # core expertise that cannot be overridden
        │       └── tactics/
        │           └── tactics.md
        ├── personalization/                         # your expertise override - path must match the overridden expertise file 
        │   └── personalization.md
        └── my-custom-expertise.md                   # custom expertise at the knowledge folder root

Add your own expertise

Following the format of the standard Tableau Agent expertise files makes the file easier for the agent to understand. You can manually create the file or chat with the agent in plan mode and ask it to interview you for the information. The agent can also assist in testing your expertise before installing it.

When you have completed your expertise file, paste it into ~/Documents/My Tableau Repository (Beta)/Tableau Agent/knowledge/ and it will be used on the next prompt that invokes it. You can also group expertise files in subfolders within knowledge/.

Override standard expertise

To override a standard expertise file, copy that file and its parent folder. The relative path of the standard file from the knowledge/.vendored/ folder and that of the override folder from the knowledge/ folder must match.

Edit the override version of the expertise file to suit your needs either manually or by chatting with the agent. The next section describes the format and agent usage of the expertise file.

Expertise file format

To match the standard expertise files, your file should have: one topic per file; a short H1 that names the decision; a one-line description, and the H2 headings as specified:

# Title That Names The Decision

One sentence: what this file is for.

## Scope Check
## When to Use
## Best Practices
## Common Mistakes
## Implementation

Add When to Say No under Best Practices when the point is to refuse an unsafe or off-standard ask. Add Related Expertise only if you must point at another module. Source and Confidence is optional for an internal standard; include a date if you want reviewers to know when it was last checked.

Scope Check

Use this exact bullet list:

Primary audience — who this is for
Authoring outcome improved — create, refine, govern, …
In-scope reason — why the agent should read this before authoring
Out-of-scope risk — what this file must not become
Tags — comma-separated keywords
Relevant user prompts/search terms — quoted phrases people actually type

Those last two lines are how search-knowledge finds the file. Put the words users will say, including the wrong ask you want to intercept (for example a request for a pie when your standard is a bar). Do not put internal taxonomy or a nickname only your team uses.

What each body section is for

Section	Write
When to Use	Situations that should load this file, not the full solution
Best Practices	Imperative rules, each with a one-line why
When to Say No	The condition, wording the agent can say, and what to offer instead
Common Mistakes	The wrong move and what breaks
Implementation	Numbered steps to run after reading

Dos and don’ts

Do put your default choices here: which chart type, how to lay out KPIs, naming, color, which starter workbook to use, and that the agent should inspect the data before building a viz.

Don’t document Tableau’s internal file format (XML), node names, or product internals unless you are recording a confirmed org-specific mechanic. Built-in tactics already cover how Tableau works.

Don’t use this file to change how tools behave (retries, session handling, API calls). Expertise is authoring guidance only.

Do write rules an agent can follow without guessing: a number, a chart type that is not allowed, a required order of checks.

Don’t hedge (“consider possibly limiting KPIs”). If it is a standard, state it as a standard.

Do put the words for common failures and non-standard requests in the search terms, so the agent finds this file when someone asks those things.

Don’t rely on the rest of the file to get the agent to load it. The agent finds expertise files from Tags, search terms, and When to Use. If those don’t describe the new behavior, the agent will not find the file.

Do when you forbid something, say what to do instead. A prohibition with no replacement is often ignored.

Don’t mix several unrelated standards into one file. Split topics.

Do strip customer names, account IDs, and proprietary identifiers from examples. Generic field names are enough.

Don’t upload or paste private data into an expertise file.

Do when the file is about inspecting data before building: limit how many clarifying questions the agent asks, and name what to look at first (the real fields in the data — do not invent columns).

Don’t paste an entire workbook as an example. One pattern per rule is enough.