Skip to content
  • Home
  • Plugin manual
  • clawock

KCNyu/clawock

81Last commit Aug 16, 2026

clawock DSH plugin

clawock is a reusable decision workflow extracted from a real multi-agent trading desk. It installs into any external harness (OpenClaw, Claude Code, Codex, DeepSeek Harness, or a custom one) and lets the model propose trades while Python owns pricing, risk limits, ledger, settlement, and grading. The system publishes all decisions and outcomes, including losses, in a public scorecard.

skills-workflows

How to install the clawock DSH plugin

dsh plugin --profile web add clawock-dsh

Copying does not run this command. Review the repository and version before installing the clawock DSH plugin.

clawock DSH plugin data source

clawock DSH plugin snapshot date: Aug 16, 2026

discovered

What the clawock DSH plugin can do

  • Certifies evidence and forces an opposing case before any trade proposal
  • Validates money and FX with a reconciliation ledger that must balance before any publish
  • Links outcomes back to the decision that caused them, keeping every improvement proposal reviewable and reversible
  • Provides 41 fetch and compute modules across 8 information layers (market, fundamentals, capital flow, news, macro, quant, book, backtest)
  • Supports bilingual Hong Kong and US market coverage with scheduled sessions and intraday check-ins
  • Runs boundary-gated: the model reads pre-assembled context files instead of fetching live data directly

Where the clawock DSH plugin fits

  • Running a live investment desk with an LLM-based agent that debates trades and records decisions
  • Backtesting an investment strategy with a transparent, graded decision pipeline
  • Integrating a portable decision contract into different AI agent harnesses without rewriting the workflow
  • Monitoring and auditing automated trading decisions with a public, verifiable scorecard
  • Building a research loop where the model proposes parameter changes that are reviewed and applied only after approval

Who the clawock DSH plugin is for

  • Quantitative traders and developers who want a auditable, harness-agnostic decision workflow for their AI agents
  • Users of AI agent frameworks (OpenClaw, Claude Code, Codex, DeepSeek Harness) looking to add investment decision capabilities
  • Researchers and enthusiasts interested in transparent, graded AI-driven trading systems

clawock DSH plugin limitations

  • Requires a real brokerage account with separate Hong Kong and US books (does not execute trades automatically)
  • Depends on external data sources (Tencent, Yahoo, Eastmoney, Polygon, etc.) that may require API keys or have rate limits
  • Currently only supports Hong Kong and US markets, with bilingual coverage
  • The live desk has not beaten buy-and-hold (losses are published)
  • Requires an external harness (OpenClaw, Claude Code, Codex, DeepSeek Harness, or custom) to provide the model call, conversation, memory, and tools

clawock DSH plugin: from the repository README

Quoted from the KCNyu/clawock README, the upstream source of the clawock DSH plugin. Copyright remains with the original authors.

### AI argues. Code settles. The losses stay on the page. A real Hong Kong + US brokerage account, run day after day by agents that have to argue their case — and graded by code the model never gets to touch. Every call is published, wins and losses alike, because a system that only shows you the wins is marketing, not evidence. Install the same decision workflow into your own agent, in any harness — OpenClaw, Claude Code, Codex, DeepSeek Harness, or one you write yourself. [![PyPI](https://img.shields.io/pypi/v/clawock?label=PYPI&style=flat-square&logo=pypi&logoColor=white&labelColor=252b35&color=4b91c8)](https://pypi.org/project/clawock/) [![npm](https://img.shields.io/npm/v/clawock-dsh?label=NPM&style=flat-square&logo=npm&logoColor=white&labelColor=252b35&color=4b91c8)](https://www.npmjs.com/package/clawock-dsh) [![Dashboard](https://img.shields.io/github/deployments/KCNyu/clawock/github-pages?label=DASHBOARD&style=flat-square&logo=githubpages&logoColor=white&labelColor=252b35&color=4b91c8)](https://kcnyu.github.io/clawock/) [![Tests](https://img.shields.io/github/actions/workflow/status/KCNyu/clawock/harness-regression.yml?label=TESTS&style=flat-square&logo=githubactions&logo

Read the full READMERepository license: MIT

clawock DSH plugin questions

How do I install clawock for DeepSeek Harness?

You can install clawock via npm: `npm install clawock-dsh`. Then configure it as a plugin in your DeepSeek Harness agent configuration. Alternatively, you can install the Python package via `pip install clawock` and use the CLI directly. Check the examples directory for a sample DeepSeek Harness agent setup.

Do I need a brokerage account to use clawock?

Yes, clawock is designed to work with a real brokerage account with separate Hong Kong and US books. The plugin does not execute trades automatically; it only proposes decisions and records outcomes. You need to provide your own account and credentials for the data sources (Tencent, Yahoo, Eastmoney, etc.) if you want live market data.

Can I use clawock with other AI agent frameworks?

Absolutely. clawock is harness-agnostic and works with OpenClaw, Claude Code, Codex, DeepSeek Harness, or any custom runtime. The core workflow is a CLI and file-based interface, so you can integrate it into any environment that can call Python scripts and read/write files.

What markets does clawock support?

Currently clawock supports Hong Kong (HKEX) and US (NYSE, NASDAQ) markets with bilingual coverage (Chinese and English). The data collection modules include Hong Kong and US specific sources like Eastmoney, SEC EDGAR, and HKEX. There is no support for other markets at this time.

How can I see the performance of the live desk?

The live desk's scorecard is public at https://kcnyu.github.io/clawock/. It shows real positions, P&L, decision history, and all eligible outcomes including losses. The dashboard updates through the trading day, and daily briefs are also published. The data is not a recommendation; it is an evidence record.