llmscan

Know what your machine can run.

A fast CLI that inspects your hardware and reports which local LLMs will actually run on it, before you download 40GB to find out. Python, installable via Homebrew.

python·24 stars·github (opens in new tab)
claude-on-a-leash

Deterministic guardrails for coding agents.

Six hooks for Claude Code that hard-block remote code execution, credential exfiltration, SSRF, and prompt injection before a command ever runs. Ships with an MCP server so any other agent gets the same protection.

shell·5 stars·github (opens in new tab)
gitatlas

A map of your whole codebase in one file.

Generates a self-contained, browsable atlas of a multi-repo codebase as a single HTML file. Includes an MCP server and token-budgeted briefs so coding agents get the right context without reading everything.

typescript·github (opens in new tab)
senior-agent-skills

Make your agent work like a senior engineer.

Eight portable SKILL.md playbooks that push coding agents toward careful, deliberate engineering. One-command install across Claude Code, Codex, Antigravity, OpenCode, Copilot, and Cursor.

agent skill·1 star·github (opens in new tab)
fuelgauge

A status line for Claude Code.

Cross-platform status line showing your folder, git branch, and color-coded progress bars for context window, 5-hour, and 7-day usage. Know how much runway you have without leaving the terminal.

powershell·8 stars·github (opens in new tab)
handcraft-html

HTML that reads like a person wrote it.

An agent skill for building HTML pages that don't look generated. Tuned for docs, references, runbooks, and portfolios, with light and dark themes and mobile handled by default.

agent skill·github (opens in new tab)

I work on backend, platform, and AI security, out of the San Francisco Bay Area. A large part of that has been anti-bot mitigation: telling automated traffic apart from real users at scale, and doing it without breaking the experience for the humans.

That work shaped how I think about AI agents. Both problems come down to the same question: how do you let automation do useful work while stopping it from doing damage? I believe safety should be deterministic, not probabilistic. A rule that blocks a dangerous action every time beats a model that usually declines. Most of my open-source work comes from that conviction.

More on LinkedIn and GitHub.

Currently using

  • Python: CLIs, security tooling, automation
  • JavaScript / Node: services and libraries
  • Go: platform and infrastructure work
  • Anti-bot mitigation: detection, rate limiting, fingerprinting
  • Cloud + CI/CD: pipelines, guardrails, deploys
  • Claude Code + MCP: agent hooks and servers

Let's build something
worth building.

Open to collaborations, consulting, and conversations about hard problems.