dsh-quant-workspace

English | 中文

A self-contained quant research workspace for DeepSeek Harness. It ships a Python engine inside the package — fetch Yahoo Finance daily data, backtest rule-based strategies, and generate interactive visual reports, all from chat.

⚠️ Not investment advice. The workspace surfaces rule state and evidence; decisions are always yours. It never places orders and never changes positions.

Features

• Data — Yahoo Finance daily bars (2 years by default), with OHLCV + indicator export.
• Backtesting — per-trade table, total return, max drawdown, win rate, average hold, and a buy-and-hold baseline.
• Visual reports — self-contained interactive HTML charts: candlesticks with bands and entry/exit markers, volume, %B, and an equity curve. Zoom (anchored at the cursor), pan, crosshair, and a red-up/green-down toggle — no dependencies, open in any browser.
• Strategy registry — save strategies you have researched and reuse them by id.
• Read-only by design — no orders, no position changes, no market-data keys.

Requirements

• A DeepSeek Harness installation (web profile) with pnpm.
• uv (runs the bundled Python engine; first use syncs python/.venv).
• Internet access for Yahoo Finance data.

Install

Installation status: not yet published to npm. Until then, install from the git spec (dsh plugin --profile web add github:AllenCX/dsh-quant-workspace) or use the dev overlay below.

dsh plugin --profile web add dsh-quant-workspace

All configuration is optional. Example user patch ($DSH_HOME/profiles/web/cordis.patch.yml):

- id: quant-workspace
  config:
    ledgerPath: 'C:\path\to\trade_log.csv'   # optional: track your real positions
    reportsDir: 'C:\path\to\reports'          # optional: where visual reports go

Dev / local overlay

pnpm dsh web --patch ./dev.patch.yml

Quick start

In a Harness session:

• "Give me today's signal card for TSLA" — single_ticker, mode daily.
• "Backtest META and generate a chart" — single_ticker, mode backtest, chart: true.
• "Compare the review health check for TSLA" — single_ticker, mode review.

A call runs exactly one rule: an example rule family, a registered strategy id, or the configured defaultRule when neither is given. Without any of these, the workspace reports that no strategy is selected.

Tools

single_ticker

• ticker (required) — symbol, e.g. TSLA. Uppercased automatically; only letters, digits, dot and dash.
• mode (default daily) — daily signal card · backtest with per-trade table · review health check.
• rule — an example rule family (currently bollinger_mean_reversion).
• strategy — id of a strategy in the workspace registry.
• chart (default false) — also write an interactive HTML report and the state CSV under reportsDir; the artifact paths are included in the output.

Strategy registry

After research, save a strategy and reuse it by id:

dsh-quant strategy register --id tsla_dip --family bollinger_mean_reversion --bollinger-window 30 --note 'dip strategy after Aug-2026 research'
dsh-quant strategy list
dsh-quant strategy remove --id tsla_dip

Example rule

The bundled engine ships one example rule so the workspace works out of the box: Bollinger mean-reversion on daily bars — enter when %B <= 0, exit when %B >= 1 (Bollinger 20, 2σ, same-bar close fills, no transaction costs in v1). Rule parameters are CLI options, and more rule families (MA cross, Donchian, RSI, trend filters) are on the roadmap.

CLI reference

dsh-quant single-ticker --ticker <T> --mode <daily|backtest|review> (--rule <family> | --strategy <id>) [--ledger <path>] [--chart <dir>] [--export-state <dir>] [--registry <path>] [--data-file <csv>]
dsh-quant strategy register|list|remove [options]

• Exit 0 with plain-text report on success; exit 1 when data cannot be loaded; exit 2 for invalid invocation.
• --data-file reads a local OHLCV CSV instead of the network (used by the tests).

Development

pnpm install && pnpm run typecheck && pnpm run test && pnpm run build   # TS shell
cd python && uv run --project . pytest tests -q                          # bundled engine

License

MIT