dsh-quant-indicators: a small quantitative toolkit built on dsh's tool contract

[pengpengyi92]

Hi everyone! After studying the dsh architecture (the ctx.tools pipeline, capability seams, and the canonical-output contract), I built a small plugin that gives agents a complete quant workflow: fetch market data → compute technical indicators → backtest a strategy. Everything is structured so Code Mode can consume it programmatically.

Repo: https://github.com/pengpengyi92/dsh-quant-indicators · npm: dsh-quant-indicators (tagged dsh-plugin)

What's inside (9 tools)

Tool	What it does
quant_market_fetch	OHLCV candles from the Binance public API (no credentials), structured candles {openTime, open, high, low, close, volume}
quant_sma / quant_ema / quant_rsi / quant_macd / quant_bollinger / quant_atr	Pure-computation indicators with length-aligned outputs (leading nulls instead of padding) so the model can index-align without guesswork
quant_backtest	Dual-MA crossover backtest: trades, position, equity curve, total return, max drawdown, annualized Sharpe. Signals confirm on bar i and execute at bar i+1 close (no look-ahead)

Design decisions worth sharing (learned from the dsh contracts)

1. Canonical outputs, not prose. Every tool returns structured JSON (output.schema + output.render split). Code Mode gets typed values for free — await tools.quant_sma(...) resolves to the actual numbers.
2. Null-alignment instead of padding. All indicator arrays are input-length aligned with leading nulls, so the model aligns by index without re-requesting.
3. Pure functions decoupled from the harness. src/indicators.ts / src/backtest.ts have zero dsh imports — numerical correctness is tested standalone (classic Wilder RSI 70.46, hand-computed MACD/ATR/backtest trades), and the harness layer is just the defineTool registration.
4. Real Loader composition tests. Following the pattern from tool-todo's loader-composition.spec.ts, the plugin boots through a real cordis.yml + Loader with a module map — registration, pipeline execution, isError paths, and HMR-safety disposal are all covered.

Verified end-to-end (live)

quant_market_fetch(BTCUSDT, 1d, 120) → closes → quant_backtest(fast 5, slow 20)
→ 3 trades, total -0.6%, maxDD 6.1%  ✓  (flat BTC market — the strategy losing slightly is the honest result)

Tests: 30 unit cases + 4 loader-composition cases + live integration, all green.

Why this matters for the ecosystem

The official toolset is excellent for coding workflows, but quantitative analysis (fetch → indicator → backtest) was a gap. This plugin shows how quickly a domain-specific toolkit can be built purely on the documented extension points — no loop changes, no UI changes, just ctx.tools.register + defineTool + a README. It's a proper bundle (dsh.bundle + cordis.patch.yml), so dsh plugin add dsh-quant-indicators works and it can join the awesome-dsh-plugin list.

I'd love feedback on the contract choices (especially the null-alignment convention), and I'm happy to turn any part of it into a proper ecosystem contribution. Also very interested in contributing to dsh itself — I've written up study notes on the architecture (Cordis composition, seam roles, tool pipeline, goal domain, subagent activation) plus a Chinese deep-dive blog post, and would be glad to help with docs, issues, or plugins. Thanks for building such a clean harness!

[pengpengyi92]

Update: the package is now published on npm — install with dsh plugin add dsh-quant-indicators or npm i dsh-quant-indicators. Links: https://www.npmjs.com/package/dsh-quant-indicators · https://github.com/pengpengyi92/dsh-quant-indicators

[pengpengyi92]

Update: the UI is live! 🐋

dsh-quant-ui — a Jane Street-style workbench that renders dsh-quant chart data:

• Live demo: https://dsh-quant-ui.pages.dev (open in any browser, real BTC backtest preloaded)
• K-lines + SMA overlays + B/S markers (Lightweight Charts v5), equity curves, fund simulation cards (¥100M capital, NAV 1.00, mgmt + high-water-mark fees), metric selector
• npm: dsh-quant-ui · repo: github.com/pengpengyi92/dsh-quant-ui
• Easter egg: click the title 3 times 🐋

dsh-quant is now at 0.14.0 — 31 tools across data, indicators, factor lab (alphalens-style), backtests, metrics, risk (VaR/CVaR/beta/alpha), fund simulation, and chart data plane.

[pengpengyi92]

📢 本插件已大幅迭代：37 工具 · 6 域 · 0.17.1，最新展示帖见 → #2513