Skills/01

Autonomous Agents

Agents that run for weeks, not minutes.

The work

Most "AI agents" are a chat window with a for-loop. Mine are long-running processes with memory, schedules, tools, a sandbox, and a way to reach a human when it matters.

I design the runtime first: what the agent is allowed to touch, how it recovers from a crash, how its state survives a restart, and how a person audits what it did. Then I give it a job.

What I deliver

  1. An AI employee for one role: inbox, scheduling, research, ops
  2. Fleets of agents with a broker that routes work and keeps a ledger
  3. Sandboxed agent runtimes (Docker) with memory and cron schedules
  4. Human-in-the-loop approval over SMS, email or a dashboard

Tools

Claude APIClaude Agent SDKOpenAI / CodexDockerTypeScriptPython

Proof

23 projects.

Cold Outreach Agent (Smartlead)

A daily Claude Code session that runs Black Lily's cold email program

internal2026

AutoNegotiate

An always-on supplier negotiator that stops at limits you set

product2026

Agent Update

Your agents text you; you answer from a black iPhone app

product2026

Busy Bee

An autonomous agent that runs a business like Claude Code runs a codebase

internal2026

Temperance

iOS executive assistant that acts in the real world and shows receipts

product2026

Recursive Momentum Trader

Claude agents autonomously research a regime-routed stock model

experiment2026

Claude Code Dash

Desktop launcher and secrets vault for running many Claude Code sessions

internal2026

Email Bee

A cold-outbound agent that has to prove its own numbers

internal2026

Temper

Boilerplate for sandboxed, long-running terminal agents

product2026

Quartermaster

Containerized broker agent that manages an agent fleet and your calendar

product2026

Librarian

Resident agent that files, vectorizes and answers questions over a vault

product2026

QM: Multiplayer Agent Harness

Slack and web agent harness with per-person sandboxes, memory and crons

product2026

Closer: iMessage Negotiation Coach

iMessage coach that calls a used-car seller's bluffs with a Bayesian model

experiment2026

agent-to-agent

Self-hosted, federated, signed email for AI agents

product2026

RIA Buddy

Chat-driven portfolio ops for advisors, with two human approval gates

product2026

Kalshi Trading Pod

Three Claude agents research, backtest and paper-trade Kalshi markets

experiment2026

Onyx Leopard

Open-source business simulation with AI CEO agents and real economics

product2026

Polymarket Autonomous Trading Agent

Rust agent that has Claude price prediction markets and sizes bets with Kelly

experiment2026

Marcus: Car-Seller Negotiation Agent

Texting agent that haggles as a private-party car seller, as a sparring partner

experiment2026

Temperance Beta Launch Research

Launch research for Temperance, plus a user-panel subagent skill

internal2026

Self-Improving Market Bot

Karpathy-style autoresearch loop for a quant strategy, gated by CPCV

experiment2026

Telegram to Claude Code

Drive a local Claude Code session from Telegram, with voice memos

product2026

Temperance Company Brain

Git repo where AI employees file dated research, leads and strategy pages

internal2026