安全

Dawn Proactive Agent v1.5

试用

Dawn Agent v1.5 self-evolution proactive architecture. P0-P4 framework for autonomous ETF trading agent with self-reflection, state machine, audit trail, mul...

它能做什么

Dawn Agent v1.5 self-evolution proactive architecture. P0-P4 framework for autonomous ETF trading agent with self-reflection, state machine, audit trail, multi-dimension scoring, and safety guardrails.

技能文档

Dawn Proactive Agent v1.5

An autonomous, self-evolving ETF trading agent architecture designed for the A-share market.

Architecture Overview

┌─────────────────────────────────────────────────────┐
│              Dawn Proactive Agent v1.5              │
├─────────────────────────────────────────────────────┤
│  P0: dawn_reflect.py    Hermes-inspired reflection  │
│  P1: dawn_state.py      LangGraph state machine     │
│  P2: dawn_audit.py      RagaAI audit trail          │
│  P3: dawn_analysis.py   ai-berkshire 4D scoring     │
│  P4: dawn_guardrails.py OpenAI Agents safety        │
├─────────────────────────────────────────────────────┤
│       Active in: 华泰柏瑞杯ETF AI交易巅峰赛          │
│       Period: 2026/6/11 - 2026/7/20                 │
│       Portfolio: 8 ETFs, ~¥1M AUM                    │
└─────────────────────────────────────────────────────┘

Components

P0 - Self-Reflection Engine (dawn_reflect.py)

Hermes-inspired post-trade reflection. After every portfolio adjustment, automatically extracts lessons learned and loads them into the next decision cycle.

  • Trigger: post-trade callback
  • Output: structured learnings with recall scoring
  • Integrates with .learnings/ structured log system

P1 - State Machine (dawn_state.py)

LangGraph-inspired workflow orchestration with checkpoint/resume.

  • States: IDLE → ANALYZE → DECIDE → EXECUTE → REFLECT → IDLE
  • Checkpoint recovery on timeout/crash
  • Timeout downgrade: auto-fallback to safest state

P2 - Audit Trail (dawn_audit.py)

RagaAI-inspired immutable decision logging.

  • Every decision recorded: timestamp, reasoning, data sources, signals, outcome
  • Queryable by date/strategy/symbol
  • Exports as structured JSON for backtesting

P3 - Multi-Dimension Scoring (dawn_analysis.py)

ai-berkshire-inspired composite scoring for ETF selection.

  • Technical (30%): momentum, trend, volume
  • News Sentiment (25%): real-time financial news analysis
  • Capital Flow (25%): sector money flow tracking
  • Volume-Price (20%): volume-price divergence detection

P4 - Safety Guardrails (dawn_guardrails.py)

OpenAI Agents SDK-inspired transaction safety checks.

  • Blacklist: 科创板(688) and ST stocks blocked
  • Position limits: max 40% single ETF
  • Daily loss limit: max -5% stop-loss
  • Available cash check before execution

Usage

Daily Strategy Run (09:28)

python scripts/dawn_proactive.py --action morning

Post-Market Review (15:05)

python scripts/dawn_proactive.py --action afternoon

Midday Check (11:30)

python scripts/dawn_proactive.py --action midday

Manual Trade

python scripts/dawn_etf_rotator.py --execute

Files

FilePurpose
scripts/dawn_proactive.pyMain orchestrator
scripts/dawn_etf_rotator.pyETF rotation strategy
scripts/dawn_reflect.pyP0: Self-reflection
scripts/dawn_state.pyP1: State machine
scripts/dawn_audit.pyP2: Audit trail
scripts/dawn_analysis.pyP3: 4D scoring
scripts/dawn_guardrails.pyP4: Safety guardrails
scripts/dawn_memory_sync.pyL1↔L3 memory sync
scripts/dawn_collector.pyMarket data collection
scripts/dawn_selector.pyETF selection logic
scripts/dawn_monitor.pyReal-time monitoring

Results (2026-07-06)

  • Cron jobs: 10 automated tasks running daily
  • Portfolio: 8 ETFs, 38.2%仓位, ¥1,003,751
  • Key wins: 科创板V反守住(+0.69% on 07-03), 全天候自动运行
  • Self-healing: 收盘复盘cron timeout自动修复 (agentTurn→command模式)
  • Delivery: 飞书推送自动修复 (delivery.to user:前缀)

See samples/proactive_demo.md for a full session trace.

Changelog

v1.5 (2026-07-06)

  • P0-P4 framework complete
  • Cron self-healing: timeout 120s→300s, agentTurn→command mode
  • Delivery fix: feishu user: prefix
  • Git cleanup: 380K lines of old archive deleted

v1.0 (2026-06-24)

  • Initial dawn agent framework
  • LM Studio + DeepSeek-R1 local inference
  • Memory sync and knowledge injection

相关技能

agentic-trader

Consent-gated autonomous trading agent for Robinhood agentic accounts. Pluggable signal sources (own commands, any JSON feed, or AgentHC's sats-priced paper-...

1 星标

agentrade-skill

Use this runbook to operate Agentrade through the authenticated `agentrade` CLI/API. Agentrade is an agent-native, human-out-of-loop collaboration platform w...

27 次安装

drt-self-improving-agent

Self-improving DRT/ICT trading agent — journals every trade (setup type, bias, R:R, outcome), analyzes its own win/loss patterns, and builds a personal trading memory that makes the agent smarter over time. 100% lokal — ingen netværkskald.

作者 Northcap Group2 次安装

dawn-memory-arch

Three-layer storage + four mechanisms + P0-P3 truth hierarchy for long-term agent memory. MEMORY.md central index, memory/core/ structured facts, local vector DB semantic search. Engramory curation, WAL protocol, context management. Zero external API, pure local filesystem + SQLite + LanceDB.

self evolving agent

Build a goal-driven self-learning loop for OpenClaw and coding agents. Use when the agent should not only log mistakes, but diagnose capability gaps, maintai...

28 次安装

crypto-sentiment-pulse

Crypto Fear & Greed index + market sentiment analysis. Know the market mood before your agent takes a position. 100% lokal analyse — ingen netværkskald, ingen API-nøgle.

作者 Northcap Group1 次安装