Preface¶

The plugin mechanism of DeepSeek Harness (DSH) allows registering specific capabilities to agents. In financial tasks, if number formatting, yields, valuation ratios, present/future values, and risk metrics all rely entirely on the model’s real-time calculations, discrepancies in standards and arithmetic errors can easily occur. dsh-plugin-finance-data consolidates these calculations into a single tool: the model handles interpretation and citation, while the plugin delivers deterministic calculation results.

What This Is¶

dsh-plugin-finance-data is a finance data toolkit designed for DSH agents. It registers a tool named finance_data covering number/currency/percentage formatting, return calculations, financial ratios, time value of money, risk metrics, data quality checks, DCA/compound interest planning, and portfolio rebalancing.

This plugin is maintained by GitHub user 863683348 and is licensed under MIT. It has been verified that it has no filesystem access and its behavior is deterministic and side-effect free.

Core Features¶

finance_data provides the following actions:

These capabilities are well-suited as a fixed interface for financial calculations within DSH conversations, rather than having the model construct formulas on the fly.

Installation and Enabling¶

First, add the plugin for the specified profile:

dsh plugin --profile <profile> add dsh-plugin-finance-data

Then restart DSH. After restarting, the finance_data tool will be registered at the host-wide scope.

All configuration options are optional and are placed in the composition row’s config:

personaSection: true
sectionOrder: 6

By default, personaSection is true and sectionOrder is 6.

Typical Usage¶

After installation and restart, you can select finance_data actions in DSH agents based on your task. Below are some examples you can directly relay to the agent:

1. When outputting amounts or percentages, ask it to use format and specify whether thousands separators or Chinese 万/亿 units are needed.
2. When calculating interval returns, ask it to use return and choose between simple, log, or annualized (CAGR-style) metrics.
3. When analyzing valuation and profitability, ask it to use ratio and specify the needed fields among PE, PB, ROE, ROA, gross/net margin, and debt-to-equity.
4. When making discounting or compounding judgments, ask it to use value and output present value, future value, or implied rate.
5. When checking risk, ask it to use risk and output mean, volatility, max drawdown, historical VaR(95), and Sharpe based on the return series.
6. When validating data, ask it to use checklist to check source, as-of date, units, and definitions.
7. When planning DCA or compounding paths, ask it to use plan to view the future value schedule and inflation-adjusted real value.
8. When maintaining a portfolio, ask it to use portfolio to view weights, drift vs targets, and generate rebalance buy/sell orders.

These examples only describe capability entry points and do not assume fixed parameter names; specific inputs should align with the data and metrics you provide.

Dependencies and Notes¶

The verified materials list the following peer dependencies:

@deepseek-ai/cordis ^4.0.1
@deepseek-ai/dsh-fs ^0.1.0-rc.6
@deepseek-ai/dsh-system-prompt ^0.1.0-rc.6
@deepseek-ai/dsh-tools ^0.1.0-rc.6
@deepseek-ai/schemastery ^3.18.1

The DSH ecosystem emphasizes “everything is a plugin,” but specific plugins should still be reviewed individually. This plugin comes from a community repository and does not represent DeepSeek or High-Flyer’s official application store. It is recommended to review the GitHub source code, MIT license, and dependency requirements before installation. Once enabled, it runs within the current dsh process and inherits the permissions and capabilities of the host process.

Related Links¶

The value of dsh-plugin-finance-data lies in delegating error-prone financial calculations to a deterministic tool: formatting, returns, ratios, time value of money, risk metrics, DCA/compound interest planning, and portfolio rebalancing all have corresponding actions. The model remains responsible for interpreting results and providing recommendations, while the plugin ensures the arithmetic steps are executed correctly.

• GitHub repository: https://github.com/863683348/dsh-plugin-finance-data
• Community directory: The directory page URL was not confirmed in the verified materials; you can search for the package name dsh-plugin-finance-data in the DSH plugin directory.