dsh-data-insight
DSH (DeepSeek Harness) 数据洞察技能插件：将原始数据（CSV / 粘贴的表格 / SQL 结果 / DuckDB）转换为包含业务结论、指标和图表的结构化 Markdown 报告。
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:duyanta123/dsh-data-insight#4a9f3fef1bcaa2d4b3e3b0a4c1ea44d7bc65a573Data Insight 以 dsh-data-insight 发布，当前版本为 0.1.5。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
English | 简体中文
A DSH (DeepSeek Harness) data-insight skill plugin: turns raw data into a structured Markdown analysis report of "business conclusions + metrics + charts".
A pure instruction-type skill plugin with zero dependencies and zero build. Computation is done by the LLM driven by the host's file/shell tools; the package ships a zero-dependency CSV profiling script and full chart/report specifications.
dsh-data-insight covers the "data → report" step: given a data source, it produces a reviewable Markdown report. It does not write or clean data back to storage, and it never sends data to any external service.
It answers:
Rigor guarantees (hard gates baked into the report template): every conclusion backed by numbers, facts separated from inference, reproducible metrics definitions, no fabricated data.
As a DSH plugin (recommended):
dsh plugin --profile web add "github:duyanta123/dsh-data-insight#v0.1.5"
Or from npm:
npm install dsh-data-insight
Or manually in two steps (in the target profile directory): add "dsh-data-insight": "^0.1.5" to dependencies in package.json, and add "dsh-data-insight" to the dsh.profile.bundles array.
Compatibility tiers: the standalone CSV / DuckDB scripts run on Node.js >= 18; as a DSH 0.1.5-rc.2 plugin it is verified with Node.js >= 22.19. Run npm run test:compat to execute an isolated-profile add, dump-config, and startup smoke test.
After restarting the profile, the data-insight-runbook skill appears in the skill list and is ready to use.
Say "analyze this CSV and give me a report", "take a look at this data", or "compute these metrics for me" with a data source attached (file path / pasted table / DuckDB connection). The model loads data-insight-runbook and runs the five-stage pipeline: input intake → data profiling → metric computation → chart rendering → report output (hard gates at each stage). The deliverable is a Markdown report written to the workspace.
node skills/data-insight-runbook/scripts/csv-profile.mjs skills/data-insight-runbook/examples/sample-sales.csv
Prints a profiling report for the CSV (schema / missing / distribution / outliers); the corresponding full report sample is examples/sample-report.md.
Zero dependencies by default; to query databases directly, install the DuckDB single-file CLI (on PATH). Install scripts: Windows skills/data-insight-runbook/scripts/setup-duckdb.ps1, macOS/Linux skills/data-insight-runbook/scripts/setup-duckdb.sh.
# Query CSV/Parquet directly: no database file, no -readonly (v1.5.5: -readonly fails on in-memory databases)
duckdb -csv -c "SELECT * FROM read_csv_auto('data.csv') LIMIT 100"
# Database file / remote DB: connection string via env var, forced read-only (POSIX shell: "$DATA_INSIGHT_DB_URL")
duckdb -readonly -csv -c "SELECT ... LIMIT 5000" "$env:DATA_INSIGHT_DB_URL"
| Option | Default | Description |
|---|---|---|
<file> | - | Path to the CSV file to profile |
--sep <char> | auto-detect | Field separator (auto-detects ,, \t, ;) |
--encoding <enc> | utf8 | File encoding: utf8 / utf16le / latin1 (transcode GBK files first) |
--limit <N> | 0 (all) | Limit rows read, for sampling very large datasets |
--json | - | Output the profiling result as JSON |
The final artifact of the five-stage pipeline is a Markdown analysis report with a fixed skeleton defined in docs/report-template.md:
-readonly (write statements are blocked); connection strings go through the DATA_INSIGHT_DB_URL env var, never into commands, config, or reports; queries default to LIMIT 5000.DuckDB says Cannot launch in-memory database in read-only mode?
Don't pass -readonly for file-less queries (CSV/Parquet) — see the example above; -readonly is only for database files / remote databases.
duckdb: command not found?
The CLI isn't installed or isn't on PATH; run the platform install script (skills/data-insight-runbook/scripts/setup-duckdb.ps1 / setup-duckdb.sh) and reopen the terminal.
Garbled Chinese text in CSV?
Prefer UTF-8 (BOM is handled correctly); transcode GBK files first with iconv -f GBK -t UTF-8.
Metrics don't match expectations?
Run node skills/data-insight-runbook/scripts/csv-profile.mjs <file> first and check the profiling report's missing values / duplicate rows / outlier distribution — real-world sample data has surfaced three classes of issues (a single dirty row driving a spike, duplicates inflating counts, missing values dragging averages), all exposed at the profiling stage.
Mermaid charts don't render in local Markdown preview?
Over file://, CDN-loaded mermaid.js is blocked by same-origin policy; open with a local renderer such as Typora, or switch to the Markdown table / ASCII bar chart channels per docs/chart-spec.md.
Old sessions won't open after upgrading the DSH host to 0.1.5.x? The Session format V3 migration is irreversible and is host behavior; back up session logs before upgrading the host (see the 0.1.4 entry in CHANGELOG.md).