Skip to content

AGENTS.md

Latest commit

 

History

History
61 lines (46 loc) · 2.37 KB

AGENTS.md

File metadata and controls

61 lines (46 loc) · 2.37 KB

AGENTS.md

Purpose

This repository is intended to build a specialized worker for vibe-trading using:

  • opencode
  • oh-my-opencode
  • pluggable MCP servers and SKILLS

The core goal is to support complex quantitative research and investment analysis workflows with reusable, automation-friendly components.

Working Mode For AI Coding Agents

  • Prioritize correctness over speed for quant or investment logic.
  • Keep every change auditable: clear assumptions, data source boundaries, and reproducible steps.
  • Prefer small, composable modules over large one-off scripts.
  • When requirements are unclear, ask for missing constraints before implementing financial logic.

Test Environment Dependencies

Use this default environment unless the user explicitly requests otherwise.

  1. OS
  • linux-based environment (WSL, Docker, or native)
  1. Python runtime
  • Conda environment: legonanobot
  • Run Python with: conda run -n legonanobot python <script.py>
  1. Tooling assumptions
  • zsh shell available
  • Project may rely on MCP servers and skill definitions; validate availability before wiring integrations

Basic Workflow

Follow this baseline loop for tasks in this repository.

  1. Clarify task scope
  • Confirm strategy goal, asset scope, time horizon, and risk constraints.
  1. Inspect current workspace state
  • Identify existing MCP/skills/instructions and any reusable components.
  1. Implement in small increments
  • Add or update one focused unit at a time.
  • Avoid broad refactors unless explicitly requested.
  1. Validate locally
  • Run targeted checks or scripts in the legonanobot environment.
  • Report what was validated and what was not validated.
  1. Summarize outcomes
  • Provide changed files, key behavior changes, and remaining risks/assumptions.

Guardrails For Quant/Investment Tasks

  • Never fabricate market data, backtest results, or performance metrics.
  • Explicitly label assumptions and data limitations.
  • Separate data collection, feature/signal logic, and execution/risk controls.
  • Prefer deterministic scripts and config-driven parameters for repeatability.

Expansion Guidance

As the repository grows, add focused customizations under .github/:

  • .github/instructions/*.instructions.md for language or folder specific rules
  • .github/skills/<skill-name>/SKILL.md for repeatable multi-step workflows
  • .github/agents/*.agent.md for specialized subagents with constrained tools