MCP prompts
The server lists 5 prompts.
Each prompt gives the agent the steps of one workflow of the xpd skill, with the tools to call and
the result that ends it. The server completes the package and rule arguments as it does in
resource templates.
lint_and_fix
Section titled “lint_and_fix”Lint a path with xpd, fix what it reports, and lint again until it is clean.
| Argument | Required | Description |
|---|---|---|
path |
no | A file or folder inside the project root. |
The agent lints, fixes, and lints again until only explained warnings are left. With
path set to samples/bad.py, the prompt starts like this.
Lint samples/bad.py with xpd and fix what it finds.
1. Call the `check` tool with paths ["samples/bad.py"].2. Call the `fix` tool with the same paths to write the safe fixes.review
Section titled “review”Review a file, a folder or a diff against xpd's rules and the Python review skill.
| Argument | Required | Description |
|---|---|---|
target |
yes | A path inside the project root, or a git range such as main..HEAD. |
package |
no | Look closest at this pack's rules, such as pandas. |
rule |
no | Look closest at this rule, such as chained-assignment. |
The agent reviews a path or a git range with the Python review skill and the check tool.
refactor_data_logic
Section titled “refactor_data_logic”Refactor data logic and prove with equiv that the result did not change.
| Argument | Required | Description |
|---|---|---|
path |
yes | The script to refactor, inside the project root. |
goal |
yes | What the refactor should achieve, such as "replace the row loop". |
The agent refactors a script and proves with the equiv tool that its result did not change.
add_to_dataflow
Section titled “add_to_dataflow”Add or change a step of step code or a node of Apache Hamilton modules, and check the dataflow with graph before and after.
| Argument | Required | Description |
|---|---|---|
path |
yes | The module with the steps or nodes, inside the project root. |
change |
yes | The step or node to add, move or rename, and what it computes. |
The agent adds or changes a step or a node, and calls the graph tool before and after.
build_pipeline
Section titled “build_pipeline”Build a scikit-learn model from a data file, save it with its contract, and show with gates that check reports no error, that a raw training refit agrees, and that a new Python process predicts from the saved file.
| Argument | Required | Description |
|---|---|---|
data |
yes | The data file, inside the project root. |
target |
yes | The column the model predicts. |
task |
no | The task in plain words, before choosing a library. |
The agent builds a scikit-learn model from a data file and saves it with its contract. It
chooses a library with find, api and discovery_gate before it writes code, and asks the
user before it adds a dependency. It runs the stages until the gates tool returns
passed: true. With code mode on, one execute call chains the stages.