pypcfx-glitch/risk-rule-design

## Core Features
- The plugin runs a risk-rule workflow on a specified X/y dataset through data quality checking, single-rule mining, parallel rule-set combination, and HTML report generation.
- It registers three tools: rrd_profiling for confirming features and removing ID, serial number, constant, near-unique, or high-missing columns, rrd_mining for single-rule evaluation and optimal rule-set combination, and rrd_report for regenerating or adjusting a report from a mining snapshot.
- It provides one expert skill named risk-rule-design that contains role definitions, workflow steps, evaluation criteria, and report structure for model or user invocation.
- The mining step can output an optimal combination, an HTML report, and a snapshot, while the report step can reuse the snapshot to regenerate the report.
- The generated HTML report includes dataset quality checks, univariate rule statistics, combination-rule metrics, all-combination ranking, scatter and precision-recall views, candidate rule pools, and appendix configuration details.

## Use Cases
- Risk-control strategy personnel can replace manual strategy adjustment with a rule-mining flow on prepared sample data.
- Data analysis personnel can inspect feature quality and candidate rules before selecting rule combinations for a target variable.
- Report consumers can open the self-contained HTML file directly in a browser without external resources.

## Technical Details
- The MIT-licensed implementation is pure JavaScript ESM with zero build steps and zero external dependencies.
- The package includes an entry file for registering tools and skills, an engine file for CSV parsing, quality checks, single-rule evaluation, and combination search, and a report file for self-contained HTML generation with inline CSS and SVG.
- Installation can use the npm package dsh-plugin-risk-rule-design through dsh plugin --profile web install, which automatically registers the bundle into the profile.
- GitHub installation adds the dependency but requires manually appending the package name to the profile bundle list and restarting dsh web.

## Notes
- The README states that the plugin is still being improved and may have issues during use.
- Users must provide a dataset path, specify the target field, and identify fields to exclude when invoking the skill.
- The GitHub installation path does not automatically register the bundle, so manual profile configuration is required.