mirror of
https://github.com/velocitatem/cvfs.git
synced 2026-05-31 08:43:37 +00:00
2.0 KiB
2.0 KiB
UltiPlate - Agent Instructions
Scaffold for any project: webapp, API, ML pipeline, scraper, worker, CLI, or SDK. Deployable via Makefile and Docker Compose.
Project Layout
apps/webapp/ Next.js 15 + React 19 + Tailwind 4 (Bun, Turbopack, auth optional)
apps/webapp-minimal/ Streamlit prototype
apps/backend/fastapi/ FastAPI server
apps/backend/flask/ Flask server
apps/worker/ Celery worker (Redis broker)
ml/ PyTorch ML pipeline (arch, train, inference, etl)
alveslib/ Shared Python library: logger, scraper, agent
src/ Simple scripts / CLI
Rules for Agents
- Use
make initto bootstrap. Usemake devto run webapp. Usemake helpfor all targets. - Python deps: use root
pyproject.toml+uv.lock;make envlinkpropagates.envto sub-apps. - JS/TS: Bun is the package manager for
apps/webapp. Usebun add/bun install/bun dev. - Do not create rogue files or test scripts outside the established structure.
- All shared Python utilities go in
alveslib/. Import from there, never duplicate logic. - No emojis in code, comments, or logs.
AI / Agent SDK
ANTHROPIC_API_KEY is required for AI features. alveslib.agent provides:
from alveslib import ask, stream, Agent
ask("prompt") # blocking one-shot
stream("prompt") # iterator of text chunks
Agent(system="...").chat("prompt") # multi-turn
For full agentic loops with file/bash tools, use the Claude Agent SDK:
pip install claude-agent-sdk
from claude_agent_sdk import query, ClaudeAgentOptions
async for msg in query(prompt="...", options=ClaudeAgentOptions(allowed_tools=["Read","Bash"])):
print(msg)
Slash Commands (.claude/commands/)
Use in Claude Code sessions (type /):
/plan- plan an implementation within this boilerplate/build- implement a feature end-to-end/api- scaffold a backend endpoint/page- scaffold a Next.js page/review- review recent changes/ship- commit staged changes