Skip to content

AGENTS.md

Latest commit

 

History

History
71 lines (51 loc) · 2.32 KB

AGENTS.md

File metadata and controls

71 lines (51 loc) · 2.32 KB

YC Hackathon Codex Guide

This file makes Codex behave consistently with the existing Claude project rules in .claude/CLAUDE.md. If there is any conflict, use .claude/CLAUDE.md as the source of truth.

Project

JARVIS is a real-time person intelligence platform for the Browser Use + YC web agents hackathon:

  • Meta glasses or phone camera input
  • Face detection and identification
  • Browser-agent enrichment and report generation
  • Corkboard-style frontend for live demo output

Non-Negotiables

Before writing implementation code:

  1. Search for existing maintained repos and libraries first.
  2. Decide use, fork + extend, or build.
  3. Prefer existing high-quality tools because this repo is in hackathon mode.

Do not change these architecture choices without explicit approval:

  • Frontend: Next.js + Tailwind + Framer Motion on Vercel
  • Backend: FastAPI
  • Real-time: Convex
  • Persistent storage: MongoDB Atlas
  • Fast research: Exa API
  • Deep research: Browser Use agent swarm
  • Face pipeline: mediapipe or similar detect, ArcFace-style embeddings, reverse-image enrichment

Build Priorities

Order of importance:

  1. Demo-visible outcomes
  2. Working fallbacks
  3. Fast iteration
  4. Code quality

The live demo matters more than broad feature count.

Expected Structure

  • backend/capture: camera input and frame extraction
  • backend/identification: face detection, embeddings, lookups
  • backend/agents: Browser Use agents and orchestration
  • backend/synthesis: report generation
  • backend/enrichment: Exa and other fast lookups
  • backend/db: Convex and MongoDB integration
  • backend/observability: Laminar and tracing
  • frontend/app: Next.js routes
  • frontend/components: UI pieces such as the corkboard and person cards
  • frontend/convex: schema, queries, mutations
  • frontend/lib: app utilities and animation helpers

Before Implementing A Module

Read the matching docs first:

  • SYSTEM_DESIGN.md
  • DESIGN_HANDOFF.md
  • TECH_DOC.md
  • TASKS.md

Working Style

  • Favor libraries over custom builds when the library already solves most of the problem.
  • Keep implementations small and observable.
  • Add timeouts and error handling for every external dependency.
  • Preserve demo quality. If a live path is fragile, provide a recordable fallback.