Skip to content

AGENTS.md

Latest commit

 

History

History
26 lines (22 loc) · 2.12 KB

AGENTS.md

File metadata and controls

26 lines (22 loc) · 2.12 KB
Raw

Myna — Agent Guide

Commands

PYTHONPATH=. pytest tests/ -v               # all 21 tests (8 domain + 11 API + 2 serialization)
python src/main.py                          # start local server on :8000

# Vercel dev:
vercel dev                                  # serverless mode (no session persistence)

Critical Quirks

  • Hexagonal Architecture — strict layer isolation:
    • src/core/ = pure domain — models.py, ports.py (ABCs), domain_services.py, agents/base.py.
    • src/adapters/ = infrastructure — api/, fs/, repositories/, visualization/.
    • Core must NEVER import from adapters. Adapters depend on core ports (ABCs), not the other way around.
    • New features: domain logic goes in core/domain_services.py, orchestration in core/agents/, API endpoints in adapters/api/router.py.
  • Two execution modes, selected by VERCEL env var:
    • Local dev (VERCEL unset): server-side state via cookies + pickle persistence in storage/.
    • Vercel stateless (VERCEL=1): no server-side session. Frontend sends df_json (DataFrame as JSON split format) with every request. Backend deserializes, processes, and returns results — no persistence.
  • Zero scikit-learn/scipy dependency: K-Means, Z-Score, and Kurtosis are implemented in pure NumPy. This is intentional — scikit-learn exceeds Vercel's 250MB serverless function limit.
  • No Docker, no CI/CD, no Makefile — this is a Vercel-deployed app. The vercel.json and .vercelignore control deployment.
  • Agent/Skill system: DataPrepAgent (in core/agents/base.py) orchestrates via AgentManager.execute_skill(). Skills are registered with @register_skill decorator. New skills go in core/agents/skills/ and must be registered via register_skill.
  • Super-Skills are parametric — compute_stats(stat_type=...), plot(type=...) instead of separate files per skill. Adding a new stat_type or plot type extends the existing super-skill file.
  • Upload flow: POST file → stored in memory as DataFrame → session holds reference. To see data in the frontend, upload first, then use other endpoints.