OpenAlphaStack is a Codex Desktop plugin for auditable A-share research and paper trading. Repository Skills define research workflows, bounded MCP tools expose data and domain actions, and Python owns deterministic execution.
- Do not launch Codex, Claude Code, another Agent, or an LLM API from the application.
- Put Agent workflow instructions in
skills/and live capabilities behind typed MCP tools. - Keep validation, T+1, fees, risk limits, state transitions, idempotency, backtests, and mechanical paper execution in Python.
- Public engine modes are paper and backtest only. MCP mutations are paper-only; historical live runs remain read-only.
- Missing or stale plans keep the engine in observation mode. Never invent a plan.
- The Dashboard is an observability surface, not an Agent terminal.
- Use one Codex Agent by default. Do not spawn or require subagents unless the user explicitly requests independent parallel work.
- Confidence and narrative reasoning are audit metadata, never execution gates.
- Bind local services to
127.0.0.1by default and never expose arbitrary shell execution. - Do not commit
.env, runtime data underdata/, secrets, or private Feishu content. - Treat per-run SQLite as canonical; JSON/JSONL files are human-readable projections.
- Preserve atomic state-plus-ledger commits and append-only ledger semantics.
- Preserve unrelated user changes. Commit or push only when the user explicitly requests it.
- Add tests for changed MCP validation, paper-only boundaries, idempotency, and state reloads.
- Run the smallest relevant checks first, then the available baseline:
npm run dashboard:test
npm run dashboard:build
python -m pytest -q
python -m compileall -q src\openalphastack
python -X utf8 C:\Users\Admin\.codex\skills\.system\skill-creator\scripts\quick_validate.py skills\market-analyzer