Voice
Clawock
By AI argues. Code settles. The losses stay on the page. A real HK + US brokerage account run by agents that must debate every call, settled by code the model never touches. Strategies are plugins whose parameter searches can only propose. Install the same workflow into OpenClaw, Claude Code, Codex, or DeepSeek Harness.
Install
pip install clawockConfiguration Example
{
"status": "rejected",
"validation_issues": [
{"code": "insufficient_opposing_evidence", "message": "requires at least 1 opposing evidence item(s)"},
{"code": "unsupported_bear_case", "message": "bear_case must cite opposing evidence"},
...
]
}README
<div align="center">
<h1><img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/logo-lockup.svg" alt="clawock" height="48"></h1>
### AI argues. Code settles. The losses stay on the page.
[](https://pypi.org/project/clawock/)
[](https://www.npmjs.com/package/clawock-dsh)
[](https://github.com/KCNyu/clawock/actions/workflows/ci.yml)
[](https://github.com/KCNyu/clawock/actions/workflows/dashboard-artifact-gate.yml)
[](https://github.com/KCNyu/clawock/actions/workflows/ci.yml)
[](https://github.com/KCNyu/clawock/blob/master/LICENSE)
[**Live dashboard**](https://kcnyu.github.io/clawock/) · [**Daily briefs**](https://kcnyu.github.io/clawock/briefs.html) · [**Evidence**](https://kcnyu.github.io/clawock/#reflect) · [**简体中文**](https://github.com/KCNyu/clawock/blob/master/README.zh.md)
<a href="https://kcnyu.github.io/clawock/">
<img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/social-card.png" alt="clawock — portable investment decision workflows for any external AI agent, proven on a live HK and US desk" width="820">
</a>
<sub><i>“The market doesn't care how confident the model was.”</i></sub>
<a href="https://kcnyu.github.io/clawock/"><img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/dashboard.gif" alt="clawock dashboard cycling through its tabs" width="820"></a>
<a href="https://kcnyu.github.io/clawock/#drill"><picture><source media="(max-width: 700px)" srcset="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/books-narrow.svg"><img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/books.svg" width="820" alt="US book and Hong Kong book, side by side: each book's return, the basis that return is divided by, and its daily P&L curve in its own currency; the combined return sits underneath on a mixed basis"></picture></a>
| **<!-- CW_M:days -->148<!-- /CW_M:days -->** | **<!-- CW_M:rows -->1013<!-- /CW_M:rows -->** | **<!-- CW_M:settled -->166<!-- /CW_M:settled -->** | **47** | **5** | **0** |
|:---:|:---:|:---:|:---:|:---:|:---:|
| days live on a real HK + US account | decisions on the public ledger | episodes settled by code | data modules across 8 layers | agent harnesses, one contract | scores the model wrote for itself |
<sub>Real positions, real P&L, losses included. Numbers and charts refresh weekly; the live dashboard updates through the trading day.</sub>
</div>
## What you get
clawock gives the AI agent you already use a continuing investment workflow for a HK + US stock book: **collect information → argue both sides → settle in code → feed the result into the next decision**. You still place the orders.
<p align="center"><img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/rsi-loop.svg" width="1016" alt="clawock on one page: eight source layers arrive through fallback routes; Python reconciles the book and certifies one context pack; Claude Code, Codex, OpenClaw, DeepSeek Harness or your own CLI argues a bull and a bear case over it; code reviews concentration and enforces the configured risk boundaries, records what happens when one breaks and enforces evidence requirements; the plan reaches you and you place the orders; code settles each call as a win, a loss or ungradeable; the record returns around two loops, one into the next judgment through history and earlier-date calibration, one into the evidence rules through a bounded proposal and a named review; on the author’s host patrol rounds, a filing gate, background agents and required CI maintain the desk’s own code, watched from DeepSeek Harness"></p>
**RSI means Recursive Self-Improvement**: each decision leaves evidence and an outcome for the next one to build on. **No promise of returns.** The active calls have yet to show an edge; what you get is an accumulating record you can check.
Strategies are [plugins mounted by configuration rows](#strategies-plug-in-a-gate-decides-what-they-may-do); a parameter search can only produce a proposal. Want to run it first? [Try it in five minutes](#try-it-in-five-minutes).
## Information in: sources before opinions
Before you open the morning brief, Python has collected quotes through fallback routes such as Tencent / Nasdaq, SEC / HKEX filings, Eastmoney capital flow, bilingual news, macro and sentiment, then reconciled your book, FX and risk. Old news is labelled separately from live information. A failed source says “not fetched,” never “no news.” Hong Kong has the base coverage, but **its research breadth is behind US**.
**You get an evidence pack.** On the live desk, `preflight` writes the core needed for this judgment and reference layers the agent can load on demand. The agent can trace the batch without reconstructing it from chat. In your own harness, `run prepare` produces `request.json`, pinning per-file and whole-context SHA-256 hashes, the workflow version and the whole-pack certificate.
**Certification pins what was used.** Citations carry the source and publication or observation time; hashes check content and generation, not whether a news story is true. Source access credentials and model API keys remain with your runtime, outside the public repository. The portable skill also lets your agent use its own research tools to add traceable evidence; scheduled desk jobs read the Python-assembled files.
**The live desk’s full information and delivery flow**
<p align="center"><img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/information-flow.svg" width="1016" alt="clawock data flow — eight information layers feed ordered fetch-fallback routes; Python reconciles the book and builds risk; the brief, session reports and intraday check-ins each run a preflight that assembles only the blocks that run can use; the agent reads those files and never fetches; a Python postflight validates, then publishes to master, to the data-plane branch the dashboard polls, and to WeChat and Telegram, with an LLM-free crontab watchdog as the delivery backstop"></p>
[Information layers and source catalog](https://github.com/KCNyu/clawock/blob/master/docs/how-the-desk-works.md#the-information-layer) · [Certification protocol](https://github.com/KCNyu/clawock/blob/master/docs/architecture/runtime-protocol.md).
## Decision out: both sides read the same evidence
Ask “Can I add to this position today?” The agent writes the supporting case, the opposing evidence that could overturn it, then an action, trigger, confidence and thesis invalidation conditions. On the live desk, analysts, bull/bear researchers, risk voices and a judge run that debate. The portable workflow takes the same evidence and opposing-case requirements into your current agent.
**You get a plan and a receipt.** The desk leaves a daily brief and `plan.json`; intraday cards compare fresh quotes with the morning's triggers. A portable run produces `decision.json`. Python checks citations, opposing evidence, invalidation and order / FX arithmetic before returning a generation-pinned publication receipt. Without an opposing case it refuses publication. Conversation verdicts on the live desk enter the ledger through `clawock record --source <harness>`. You decide which orders to place.
<p align="center"><img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/decision-card-example.png" width="640" alt="A clawock decision receipt: the subject, a bull case and a bear case each citing evidence, the thesis, the invalidation condition, confidence and action, and a published run id with a pinned certificate"></p>
<sub>Example output, not a settled call.</sub>
**The live desk’s full bull / bear debate**
<p align="center"><img src="https://raw.githubusercontent.com/KCNyu/clawock/refs/heads/master/site/assets/debate-flow.svg" width="1016" alt="clawock's multi-agent debate — one evidence pack feeds four analyst lenses; two researchers argue opposing bull and bear cases and record where they disagree; three risk voices and a judge name the strategy frame and resolve it into plan.json, which enters the next session's grading loop"></p>
The desk debate is adapted from [TradingAgents](https://github.com/TauricResearch/TradingAgents); the judge names the strategy frame behind each call.
## After the close: results feed the next decision
The close is not the end of the conversation. You use `mark-followed` to mark followed / not-followed execution evidence. Code checks triggers and outcomes on canonical bars using each market's calendar, groups repeated calls on the same thesis into one episode, and keeps the ungradeable cases visible.
**The next decision starts from this record.** The next live brief reads decision metrics, past reviews and confidence calibration. Calibration uses earlier dates only, shrinking or abstaining when evidence is thin. A failed call stays linked to its original evidence, so the agent can revisit yeste
... (truncated)
voice
Comments
Sign in to leave a comment