Introduction

In the DSH (DeepSeek Harness) ecosystem, a plugin-based architecture is a common practice: tools, skills, and documentation are split into installable bundles. In data-centric tasks, the common pain point isn’t “whether the model can see the data,” but rather how to consistently organize CSVs, pasted tables, SQL query results, or DuckDB data into reusable reports.

duyanta123/dsh-data-insight provides a data insight skill plugin: it transforms raw data into a structured Markdown analysis report containing “business conclusions + metric data + charts”.

The DSH Community Directory is a standalone site with no official affiliation to DeepSeek or Flash. Here, we introduce the plugin’s capabilities and usage boundaries.

What is this

  • Plugin Name: duyanta123/dsh-data-insight
  • Maintainer: duyanta123
  • Type: DSH Data Insight Skill Plugin
  • License: MIT
  • Implementation: Pure prompt-based skill plugin, zero dependencies, zero build; computation is completed by the LLM driven by the host’s provided file / shell tools.

Core Features

Data Sources

Supports four data sources:

  • CSV files
  • Pasted table text
  • SQL query results
  • DuckDB direct database connection (optional)

Five-Stage Pipeline

The execution flow is:

Input Acceptance → Data Exploration → Metric Calculation → Chart Presentation → Report Generation

Metrics and Charts

Metric capabilities include:

  • Summary statistics
  • MoM/CoM (baseline period rules are hardcoded)
  • TopN
  • Outliers (Z-score / IQR)

Chart presentation has three channels:

  • Markdown tables + numbers
  • Mermaid
  • ASCII bar charts

Installation and Activation

Execute in the target profile:

dsh plugin --profile web add dsh-data-insight

After restarting the profile, if data-insight-runbook appears in the skills list, it is ready to use.

The plugin declares an optional peer dependency:

@deepseek-ai/dsh-skill-filesystem

You should check the source code and the MIT license before installing. The plugin runs with the current dsh process permissions; when involving local files, Shell tools, or DuckDB, you need to confirm that these tools and data sources are within the expected permission scope.

Typical Usage

Session Trigger

In a session, you can say:

  • “Analyze this CSV and generate a report”
  • “Look at this data”
  • “Help me calculate the metrics”

Attach the data source: file path, pasted table, or DuckDB connection. The model will load data-insight-runbook and execute the five stages.

Start with CSV Exploration

If you want to check sample data first, you can run:

node scripts/csv-profile.mjs examples/sample-sales.csv

This step will output an exploration report for examples/sample-sales.csv, covering schema, missing values, distribution, and anomalies.

DuckDB Optional Direct Connection

DuckDB direct connection is an optional capability. To use it, first install the DuckDB single-file CLI and add it to PATH.

Query without library files (CSV/Parquet) example:

duckdb -csv -c "SELECT * FROM read_csv_auto('data.csv') LIMIT 100"

When connecting to library files / remote libraries, you must observe the following strict rules:

  • Always use -readonly when connecting to a library file; write statements will be blocked.
  • Connection strings go through the environment variable DATA_INSIGHT_DB_URL.
  • Queries default to LIMIT 5000.
  • Queries without library files (CSV/Parquet) should not add -readonly; -readonly is only used for library files / remote libraries.

Suitable Scenarios and Notes

Suitable for: Scenarios where raw data needs to be organized into a Markdown report within a DSH session, especially CSVs, pasted tables, SQL query results, or DuckDB data.

Notes:

  • When CSV Chinese characters are garbled, prioritize UTF-8 (files with BOM can be handled correctly); for GBK encoded files, convert with iconv -f GBK -t UTF-8 before exploring.
  • When Mermaid charts do not render in local Markdown preview, you can open them with local rendering editors like Typora, or refer to docs/chart-spec.md to switch to the Markdown table / ASCII bar chart channel.
  • DuckDB is an optional dependency; you do not need to install DuckDB when not connecting to a database.

Conclusion

To summarize: duyanta123/dsh-data-insight converges the “raw data to Markdown report” steps into a DSH skill workflow; input, exploration, metrics, charts, and report are all completed within the same session.

Project URL: GitHub. If searching via the DSH Community Directory, you can search for the plugin name dsh-data-insight.