aifinlab-FinClaw Reference Experience
状态:Docker OpenClaw full case rerun complete
本地路径:/Users/mlabs/Programs/aifinlab-FinClaw
GitHub:https://github.com/aifinlab/FinClaw.git
Context Snapshot
- 分支:
main - HEAD:
e7cee33c7406c845fbdb2282d4781c591c881237 - Ahead / Behind:
0 / 0 - 初始状态:clean
- 体验重点:Docker/OpenClaw Agent 入口、skillsChoice、真实 chat case、模型 telemetry、工具降级、action-state truthfulness。
- 安全边界:不把大规模 Skills 仓库当成
FinClawMVP 产品形态。
Resume Point
已完成 Docker/OpenClaw 口径重测:
- 容器:
aifinlab-finclaw-openclaw - 人工入口:
http://127.0.0.1:18790/ - Agent 入口:
docker exec aifinlab-finclaw-openclaw node /app/openclaw.mjs agent --agent main --message "..." - 模型:
moonshot/kimi-k2.6 - 完整 case run:
logs/aifinlab-20260511-docker-openclaw-fcmatrix-run/ - 正式报告:
projects/finclaw/reference-experience/aifinlab-finclaw-evaluation.md
下一步:如需继续 aifinlab,建议做 60 个 skillsChoice selected skills 的 script/tool-level smoke,而不是重跑 982-skill root。